Skip to content

armgit / hypit

Public Imported from GitHub
0

Clone any viral video with AI agents. Not just a script, the whole workflow: swap the face, the words, the B-roll, ship 100 variants in one command, and get your 100M views.

refactor(image): align prompt kits and craft around visual intent

kashorinc54b43f94ae8
diff --git a/examples/interview/reference.svml b/examples/interview/reference.svml
index 2788bce5..cf0ac87d 100644
--- a/examples/interview/reference.svml
+++ b/examples/interview/reference.svml
@@ -52,66 +52,77 @@
</bitcoin-rule>
</script>
+ <text:Value id="shared-view-person">
+ The guest is a Japanese-Italian woman about twenty, with a glamorous mob-wife presence.
+ She is exceptionally beautiful, with very fair skin and the refined, striking beauty of a Hollywood leading actress.
+ She has broad shoulders and excellent head-to-shoulder proportions.
+ Her long black hair falls in loose waves. Sunglasses rest on her head. She wears an expensive-looking dark
+ fur coat over a leopard-print strapless short dress, a gold-set cross necklace, and a luxurious structured
+ Hermes handbag.
+ The interviewer is a cheerful young American college student, exceptionally handsome, with the looks of a young Hollywood actor.
+ He has fair skin, broad shoulders and excellent head-to-shoulder proportions.
+ He has tousled curls spilling from a backward American-flag baseball cap, a purple designer graphic T-shirt,
+ a studded belt and fashionable flared jeans.
+ </text:Value>
<text:Value id="shared-view-shot">
Generate a close street-interview photograph showing two people from roughly mid-thigh upward.
- The guest stands just left of center, partly facing the camera and looking toward the interviewer on the right.
+ The guest stands just left of center, partly facing the camera and looking toward the interviewer in profile on the right.
+ Her handbag hangs on her left arm. Her hands rest naturally near her bag, leaving room for an easy gesture.
+ He holds one microphone marked HYPIT toward her mouth.
Keep their faces, gestures and the handheld microphone readable, with comfortable room around them.
+ They stand near the front of a matte sky-blue Lamborghini parked at the curb with its nose toward the left.
+ Its badge is visible low in the frame between the two people.
</text:Value>
- <text:Value id="shared-view-direction">
- On an affluent New York street, a matte sky-blue Lamborghini is parked at the curb with its nose toward
- the left. Its badge is visible low in the frame between the two people standing near the front of the car.
- The guest has a glamorous mob-wife presence. She is about twenty, exceptionally beautiful, with
- Japanese-Italian features, very fair skin and the refined, striking beauty of a Hollywood leading actress.
- Her long black hair falls in loose waves. Sunglasses rest on her head. She wears an expensive-looking dark
- fur coat over a leopard-print strapless short dress, a gold-set cross necklace, and a luxurious structured
- Hermes handbag on her left arm. Her hands rest naturally near her bag, leaving room for an easy gesture.
- The interviewer is a handsome, cheerful young American college student in profile on the right.
- He has tousled curls spilling from a backward American-flag baseball cap, a purple designer graphic T-shirt,
- a studded belt and fashionable flared jeans. He holds one microphone marked HYPIT toward her mouth.
+ <text:Value id="shared-view-setting">
+ The setting is an affluent New York street.
The stone facades, railings and trees establish a real neighborhood. No other people are in the frame.
+ The street's palette is sandstone and green; the sky-blue car and purple shirt provide accents.
</text:Value>
<text:Render id="shared-view-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={shared-view-person}/>
<text:Set name="shot" text={shared-view-shot}/>
- <text:Set name="direction" text={shared-view-direction}/>
+ <text:Set name="setting" text={shared-view-setting}/>
</text:Render>
<gpt:Image id="shared-view" prompt={shared-view-prompt} aspect-ratio="9:16" resolution="2K"/>
- <text:Value id="guest-view-shot">
- Create a closer camera view centered on the woman from the reference. She looks toward the right side of the frame.
- </text:Value>
- <text:Value id="guest-view-direction">
+ <text:Value id="guest-view-person">
+ Use the first image as the complete identity reference for the woman.
Preserve her face, black hair, sunglasses, fur coat, leopard-print dress, jewelry and handbag.
- Keep the same street, car and natural light from this new camera position.
+ </text:Value>
+ <text:Value id="guest-view-shot">
+ Create a closer camera view centered on her. She looks toward the right side of the frame.
The interviewer's microphone enters from the right toward her mouth; her face and upper body remain clearly visible.
</text:Value>
- <text:Value id="guest-view-references">
- Use the first image as the complete identity and scene reference. This is another view of the same interview.
+ <text:Value id="guest-view-setting">
+ Use the first image as the complete scene reference. This is another view of the same interview.
+ Keep the same street, car and natural light from this new camera position.
</text:Value>
<text:Render id="guest-view-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={guest-view-person}/>
<text:Set name="shot" text={guest-view-shot}/>
- <text:Set name="direction" text={guest-view-direction}/>
- <text:Set name="references" text={guest-view-references}/>
+ <text:Set name="setting" text={guest-view-setting}/>
</text:Render>
<gpt:Image id="guest-view" prompt={guest-view-prompt} aspect-ratio="9:16" resolution="2K">
<gpt:Reference image={shared-view.image}/>
</gpt:Image>
+ <text:Value id="interviewer-view-person">
+ Use the first image as the identity reference for the man.
+ Preserve his handsome face, tousled curls, backward cap and purple shirt.
+ </text:Value>
<text:Value id="interviewer-view-shot">
- Create a closer camera view centered on the man from the reference, slightly larger in frame.
+ Create a closer camera view centered on him, slightly larger in frame.
He continues looking left and holding the microphone in the same interviewing pose.
- Keep part of the woman's body visible along the left edge.
- </text:Value>
- <text:Value id="interviewer-view-direction">
- Preserve his handsome face, tousled curls, backward cap, purple shirt and the same street and car.
- Keep the natural light and the left-right relationship between the two people coherent.
+ Keep part of the woman's body visible along the left edge so this view remains part of their shared space.
+ Preserve the left-right relationship between the two people.
</text:Value>
- <text:Value id="interviewer-view-references">
- Use the first image as the identity and scene reference. Preserve the woman at the edge so this view remains part of their shared space.
+ <text:Value id="interviewer-view-setting">
+ Use the first image as the scene reference. Keep the same street, car and natural light.
</text:Value>
<text:Render id="interviewer-view-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={interviewer-view-person}/>
<text:Set name="shot" text={interviewer-view-shot}/>
- <text:Set name="direction" text={interviewer-view-direction}/>
- <text:Set name="references" text={interviewer-view-references}/>
+ <text:Set name="setting" text={interviewer-view-setting}/>
</text:Render>
<gpt:Image id="interviewer-view" prompt={interviewer-view-prompt} aspect-ratio="9:16" resolution="2K">
<gpt:Reference image={shared-view.image}/>
diff --git a/examples/podcast/reference.svml b/examples/podcast/reference.svml
index b06b185d..5bf26ed7 100644
--- a/examples/podcast/reference.svml
+++ b/examples/podcast/reference.svml
@@ -41,50 +41,56 @@
</script>
<!-- The main host establishes the person, place and capture language. -->
+ <text:Value id="girl-person">
+ A Japanese-American woman in her early twenties, exceptionally, strikingly beautiful, with the looks of a top Korean girl-group idol.
+ She has a small refined face, very fair skin, large attractive eyes with blue contact lenses, and long straight
+ black hair with blunt bangs. Her strikingly muscular physique contrasts with her delicate face.
+ Her shoulders are very broad, her head-to-shoulder proportions excellent, and her arms have pronounced
+ muscle definition and visible veins. She wears a black strapless top with an angel-wing design and
+ relaxed tweed trousers with an asymmetrical pattern of pale stripes.
+ </text:Value>
<text:Value id="girl-shot">
- Generate a close half-body podcast view. A woman sits toward the right, her body angled slightly left,
+ Generate a close half-body podcast view. She sits toward the right, her body angled slightly left,
looking at her conversation partner outside the left edge. Her face remains clearly readable.
She sits back comfortably in a fabric armchair with her legs crossed, near enough to the camera for her
face and upper body to feel present, while her shoulders, arms and seated legs stay comfortably framed.
+ A half-eaten grape bingsu sits on a table in the lower left.
+ A podcast microphone extends from the left, marked HYPIT in attractive uppercase lettering.
</text:Value>
- <text:Value id="girl-direction">
- She is exceptionally beautiful, with Japanese-American features and Korean girl-group-idol presence:
- a small refined face, very fair skin, large attractive eyes with blue contact lenses, and long straight
- black hair with blunt bangs. Her strikingly muscular physique contrasts with her delicate face.
- Her shoulders are very broad, her head-to-shoulder proportions excellent, and her arms have pronounced
- muscle definition and visible veins. She wears a black strapless top with an angel-wing design and
- relaxed tweed trousers with an asymmetrical pattern of pale stripes.
+ <text:Value id="girl-setting">
The setting is outside a pleasant Korean street-side restaurant, with textured dark wooden walls,
- plants and other tables. A half-eaten grape bingsu sits on a table in the lower left.
- A podcast microphone extends from the left, marked HYPIT in attractive uppercase lettering.
+ plants and other tables.
+ Wood brown and coral red establish the background palette, with green plants as accents.
Natural light falls comfortably across her face; the wooden wall retains its own color and texture.
</text:Value>
<text:Render id="girl-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={girl-person}/>
<text:Set name="shot" text={girl-shot}/>
- <text:Set name="direction" text={girl-direction}/>
+ <text:Set name="setting" text={girl-setting}/>
</text:Render>
<gpt:Image id="girl" prompt={girl-prompt} aspect-ratio="9:16" resolution="2K"/>
- <text:Value id="guy-shot">
- Generate the complementary half-body camera view across this conversation.
- A slim young man sits toward the left, angled and looking right toward the woman outside the frame.
- Match the comfortable camera distance and scale of the first view. The microphone now enters from the right;
- a bowl of mango bingsu on the lower-right table balances him.
+ <text:Value id="guy-person">
+ A young American man, exceptionally handsome, with the looks of a young Hollywood actor and an endearing nerdy presence.
+ He has fair skin, a slim build, broad shoulders and excellent head-to-shoulder proportions, messy curly hair
+ and black-framed glasses. He wears a pale finely striped T-shirt with a small abstract
+ devil-tail graphic, black trousers and a cool necklace. Create this other host rather than mirroring the woman.
</text:Value>
- <text:Value id="guy-direction">
- He is a handsome, skinny American young man with an endearing nerdy presence, messy curly hair,
- black-framed glasses and an attractive face. He wears a pale finely striped T-shirt with a small abstract
- devil-tail graphic, black trousers and a cool necklace. He sits casually in the corresponding fabric chair.
- Keep the same restaurant, textured wooden surfaces and natural lighting, but show the other part of the
- space with different background objects appropriate to this camera's direction. The microphone reads HYPIT.
+ <text:Value id="guy-shot">
+ Use the first image for the conversational geometry and generate the complementary half-body camera view.
+ He sits casually in the corresponding fabric chair toward the left, angled and looking right toward the woman outside the frame.
+ Match the comfortable camera distance and scale of the first view. The microphone now enters from the right
+ and reads HYPIT; a bowl of mango bingsu on the lower-right table balances him.
</text:Value>
- <text:Value id="guy-references">
- Use the first image for the restaurant, capture language and conversational geometry. Create the other host and the opposite camera view; do not mirror the woman or duplicate her background.
+ <text:Value id="guy-setting">
+ Use the first image for the restaurant and capture language. Keep its textured wooden surfaces and natural
+ lighting, but show the other part of the space with different background objects appropriate to this
+ camera's direction rather than duplicating her background.
</text:Value>
<text:Render id="guy-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={guy-person}/>
<text:Set name="shot" text={guy-shot}/>
- <text:Set name="direction" text={guy-direction}/>
- <text:Set name="references" text={guy-references}/>
+ <text:Set name="setting" text={guy-setting}/>
</text:Render>
<gpt:Image id="guy" prompt={guy-prompt} aspect-ratio="9:16" resolution="2K">
<gpt:Reference image={girl.image}/>
@@ -108,134 +114,143 @@
</mimo:VoiceDesign>
<asset:Audio id="soundtrack" src="./assets/shared-soundtrack.m4a"/>
+ <text:Value id="split-person">
+ Preserve the man's identity and styling from the first image and the woman's identity and styling from the second.
+ </text:Value>
<text:Value id="split-shot">
Create a horizontal split-screen image: the man fills the upper panel and the woman fills the lower panel.
+ The first image supplies his camera view for the upper panel; the second supplies hers for the lower panel.
Both panels are close views of their faces and upper bodies, with their microphones visible.
- </text:Value>
- <text:Value id="split-direction">
- Keep each person's identity, styling, restaurant surroundings and natural lighting.
The man looks right and the woman looks left, as in their individual views.
Keep the horizontal dividing line clean and preserve the sense that both are present in the conversation.
</text:Value>
- <text:Value id="split-references">
- The first image supplies the man and his camera view for the upper panel. The second supplies the woman and her camera view for the lower panel.
+ <text:Value id="split-setting">
+ Keep each reference image's restaurant surroundings and natural lighting in its corresponding panel.
</text:Value>
<text:Render id="split-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={split-person}/>
<text:Set name="shot" text={split-shot}/>
- <text:Set name="direction" text={split-direction}/>
- <text:Set name="references" text={split-references}/>
+ <text:Set name="setting" text={split-setting}/>
</text:Render>
<gpt:Image id="split" prompt={split-prompt} aspect-ratio="9:16" resolution="2K">
<gpt:Reference image={guy.image}/>
<gpt:Reference image={girl.image}/>
</gpt:Image>
- <text:Value id="girl-hold-shot">
- Keep the first image's camera view, framing, natural lighting, seated pose and microphone position.
+ <text:Value id="girl-hold-person">
+ Image one supplies the woman. Preserve her beauty, identity and muscular physique.
</text:Value>
- <text:Value id="girl-hold-direction">
+ <text:Value id="girl-hold-shot">
+ Keep the first image's camera view, framing, seated pose and microphone position.
The woman holds the creatine tub from the second image with the hand on the right side of the frame,
while the hand on the left points toward it. She still looks left as if talking to her partner.
- Preserve her beauty, identity and muscular physique. The tub belongs naturally in her hand and the scene.
+ Image two supplies the exact product, including its label and proportions. The tub belongs naturally in her hand and the scene.
</text:Value>
- <text:Value id="girl-hold-references">
- Image one supplies the host and the complete scene. Image two supplies the exact product, including its label and proportions.
+ <text:Value id="girl-hold-setting">
+ Image one supplies the complete scene. Keep its surroundings and natural lighting.
</text:Value>
<text:Render id="girl-hold-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={girl-hold-person}/>
<text:Set name="shot" text={girl-hold-shot}/>
- <text:Set name="direction" text={girl-hold-direction}/>
- <text:Set name="references" text={girl-hold-references}/>
+ <text:Set name="setting" text={girl-hold-setting}/>
</text:Render>
<gpt:Image id="girl-hold" prompt={girl-hold-prompt} aspect-ratio="9:16" resolution="2K">
<gpt:Reference image={girl.image}/>
<gpt:Reference image={creatine.image}/>
</gpt:Image>
- <text:Value id="guy-hold-shot">
- Keep the first image's camera view, framing, natural lighting, seated pose and microphone position.
+ <text:Value id="guy-hold-person">
+ Image one supplies the man. Preserve his identity and clothing.
</text:Value>
- <text:Value id="guy-hold-direction">
+ <text:Value id="guy-hold-shot">
+ Keep the first image's camera view, framing, seated pose and microphone position.
The man holds the creatine tub from the second image with the hand on the left side of the frame.
He still looks right as if speaking to his partner. Keep his other hand naturally at rest.
- Preserve his identity and clothing; the tub belongs naturally in his hand and the scene.
+ Image two supplies the same product held by the woman. The tub belongs naturally in his hand and the scene.
</text:Value>
- <text:Value id="guy-hold-references">
- Image one supplies the host and the complete scene. Image two supplies the same product held by the woman.
+ <text:Value id="guy-hold-setting">
+ Image one supplies the complete scene. Keep its surroundings and natural lighting.
</text:Value>
<text:Render id="guy-hold-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={guy-hold-person}/>
<text:Set name="shot" text={guy-hold-shot}/>
- <text:Set name="direction" text={guy-hold-direction}/>
- <text:Set name="references" text={guy-hold-references}/>
+ <text:Set name="setting" text={guy-hold-setting}/>
</text:Render>
<gpt:Image id="guy-hold" prompt={guy-hold-prompt} aspect-ratio="9:16" resolution="2K">
<gpt:Reference image={guy.image}/>
<gpt:Reference image={creatine.image}/>
</gpt:Image>
+ <text:Value id="coffee-person">
+ The first image supplies her identity and physique. Preserve her exceptionally attractive face,
+ muscular shoulders and strong arms.
+ She has changed into a lightweight pale-gray sleeveless zip-front sports layer over a beige bandeau,
+ with the hood and an additional pink baseball cap.
+ </text:Value>
<text:Value id="coffee-shot">
- Generate a lively raised-arm selfie of the woman turning toward her phone inside a cafe.
- The view feels comfortably close and casually composed.
+ Generate a lively raised-arm selfie of the woman turning toward her phone.
+ The view feels comfortably close and casually composed. Her expression is playful and sociable,
+ with one casual hand gesture.
</text:Value>
- <text:Value id="coffee-direction">
- She has changed into a lightweight pale-gray sleeveless zip-front sports layer over a beige bandeau,
- with the hood and an additional pink baseball cap. Her expression is playful and sociable, with one casual
- hand gesture. Preserve her exceptionally attractive face, muscular shoulders and strong arms.
+ <text:Value id="coffee-setting">
+ Change her location to a cafe.
Other customers are faintly present in the cafe. Afternoon light through the windows gives her face
natural illumination; cafe furniture and surroundings retain their color.
</text:Value>
- <text:Value id="coffee-references">
- The first image supplies her identity and physique. Change her clothing, pose and location to the cafe selfie described here.
- </text:Value>
<text:Render id="coffee-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={coffee-person}/>
<text:Set name="shot" text={coffee-shot}/>
- <text:Set name="direction" text={coffee-direction}/>
- <text:Set name="references" text={coffee-references}/>
+ <text:Set name="setting" text={coffee-setting}/>
</text:Render>
<gpt:Image id="coffee" prompt={coffee-prompt} aspect-ratio="9:16" resolution="2K">
<gpt:Reference image={girl.image}/>
</gpt:Image>
+ <text:Value id="smoothie-person">
+ Preserve the woman from the first image, with her very beautiful face and clearly muscular arms and shoulders.
+ Change her outfit to a navy Adidas sports tank and charcoal over-ear headphones.
+ </text:Value>
<text:Value id="smoothie-shot">
- Generate a lively mirror selfie of the woman standing in a dessert shop.
+ Generate a lively mirror selfie of the woman standing.
She holds her phone in one hand, watches its screen, and holds a bowl of strawberry bingsu in the other.
+ Her phone has a playful CASETiFY case with a cute rabbit design.
+ She parts her lips with an amused expression as if about to taste the bingsu.
+ The composition feels spontaneous rather than rigid.
</text:Value>
- <text:Value id="smoothie-direction">
- She wears a navy Adidas sports tank and charcoal over-ear headphones. Her phone has a playful CASETiFY
- case with a cute rabbit design. She parts her lips with an amused expression as if about to taste the bingsu.
- Keep her very beautiful face and clearly muscular arms and shoulders. The shop is lit naturally through
- its windows, with a readable background. The composition feels spontaneous rather than rigid.
- </text:Value>
- <text:Value id="smoothie-references">
- Preserve the woman from the first image while creating this different outfit, activity and location. This dessert scene broadly supports her food routine; it is not a literal smoothie demonstration.
+ <text:Value id="smoothie-setting">
+ Change the location to a dessert shop, lit naturally through its windows, with a readable background.
+ This dessert scene broadly supports her food routine; it is not a literal smoothie demonstration.
</text:Value>
<text:Render id="smoothie-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={smoothie-person}/>
<text:Set name="shot" text={smoothie-shot}/>
- <text:Set name="direction" text={smoothie-direction}/>
- <text:Set name="references" text={smoothie-references}/>
+ <text:Set name="setting" text={smoothie-setting}/>
</text:Render>
<gpt:Image id="smoothie" prompt={smoothie-prompt} aspect-ratio="9:16" resolution="2K">
<gpt:Reference image={girl.image}/>
</gpt:Image>
+ <text:Value id="pasta-person">
+ Use the first image for her identity and physique. Keep her exceptionally attractive face,
+ strong arms and broad muscular shoulders.
+ She wears a pink spaghetti-strap apron over a textured pale T-shirt, with a pale hair clip holding back
+ part of her hair.
+ </text:Value>
<text:Value id="pasta-shot">
- Generate a low-angle photograph as if a phone has been set on a table in the woman's home courtyard.
+ Generate a low-angle photograph as if a phone has been set on a table.
She is close enough for her attentive expression and practical hand movements to be clearly readable.
+ She is preparing macaroni at a stovetop, with pasta ready to go into a pot.
</text:Value>
- <text:Value id="pasta-direction">
- She wears a pink spaghetti-strap apron over a textured pale T-shirt, with a pale hair clip holding back
- part of her hair. She is preparing macaroni at a stovetop, with pasta ready to go into a pot.
- Keep her exceptionally attractive face, strong arms and broad muscular shoulders.
+ <text:Value id="pasta-setting">
+ Create this home-cooking scene independently of the other lifestyle images, in her home courtyard.
The covered courtyard has a warm modern Korean domestic feel, with a slatted wooden roof and hanging plants.
Afternoon sunlight enters the courtyard and gives her face natural illumination, while the covered area
and surrounding wood keep the scene tonally grounded.
</text:Value>
- <text:Value id="pasta-references">
- Use the first image for her identity and physique; create the home-cooking scene independently of the other lifestyle images.
- </text:Value>
<text:Render id="pasta-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={pasta-person}/>
<text:Set name="shot" text={pasta-shot}/>
- <text:Set name="direction" text={pasta-direction}/>
- <text:Set name="references" text={pasta-references}/>
+ <text:Set name="setting" text={pasta-setting}/>
</text:Render>
<gpt:Image id="pasta" prompt={pasta-prompt} aspect-ratio="9:16" resolution="2K">
<gpt:Reference image={girl.image}/>
diff --git a/examples/ranking-football/reference.svml b/examples/ranking-football/reference.svml
index 8f6cb129..7beb87d0 100644
--- a/examples/ranking-football/reference.svml
+++ b/examples/ranking-football/reference.svml
@@ -36,19 +36,22 @@
</messi>
</script>
- <text:Value id="presenter-shot">
- Generate a close half-body view of a woman seated to the right, speaking to the camera with a handheld microphone.
- Keep her head, broad shoulders and arms comfortably inside the frame, with her seated legs visible below her torso.
- She is comfortably near the camera, with natural head-to-body proportions. Her face sits toward the upper right.
- A small desk at her left balances the composition and leaves useful space for the player's icon on the left.
- </text:Value>
- <text:Value id="presenter-direction">
- She is exceptionally beautiful, with Japanese-American features and the captivating presence of a Korean girl-group idol.
+ <text:Value id="presenter-person">
+ A Japanese-American woman in her early twenties, exceptionally, strikingly beautiful, with the looks of a top Korean girl-group idol.
She has a small, refined face, very fair skin, broad shoulders and excellent head-to-shoulder proportions.
Her light goth makeup has a sweet, cool attitude and a feline quality around the eyes.
Her long black hair is straight with blunt bangs. She wears attractive earrings, a choker and layered jewelry,
a spiked baseball cap marked HYPIT in uppercase Gothic lettering, and a black cotton goth-style short-sleeved top.
- She sits on a chair in a university classroom. The desk beside her has a cluttered open storage compartment.
+ </text:Value>
+ <text:Value id="presenter-shot">
+ Generate a close half-body view of her seated on a chair to the right, speaking to the camera with a handheld microphone, her free hand gesturing naturally.
+ Keep her head, shoulders and arms comfortably inside the frame, with her seated legs visible below her torso.
+ She is comfortably near the camera. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper right.
+ A small desk at her left balances the composition and leaves useful space for the player's icon on the left.
+ The desk beside her has a cluttered open storage compartment.
+ </text:Value>
+ <text:Value id="presenter-setting">
+ The setting is a university classroom.
Behind her are a classroom door and casually arranged football posters in varied sizes:
a sky-blue-and-white striped jersey, a football, a Ballon d'Or and an abstract football design.
There are no players, school names or school logos on the posters.
@@ -56,8 +59,9 @@
her black outfit and hair give her a clear presence against it.
</text:Value>
<text:Render id="presenter-prompt" template={image-kit.phone-ugc-v1}>
+ <text:Set name="person" text={presenter-person}/>
<text:Set name="shot" text={presenter-shot}/>
- <text:Set name="direction" text={presenter-direction}/>
+ <text:Set name="setting" text={presenter-setting}/>
</text:Render>
<gpt:Image id="presenter" prompt={presenter-prompt} aspect-ratio="9:16" resolution="2K"/>
diff --git a/packages/gpt-image-kits/README.md b/packages/gpt-image-kits/README.md
index 518e9875..ff90251b 100644
--- a/packages/gpt-image-kits/README.md
+++ b/packages/gpt-image-kits/README.md
@@ -2,7 +2,8 @@
`phone-ugc-v1` is a data-only Text Template for assembling image prompts. Its
[source](kits/phone-ugc-v1.svs) emits a fixed iPhone-video capture paragraph, followed by the supplied
-Text slots, with paragraph separators. Import it as `@hypit/gpt-image-kits/phone-ugc-v1`.
+Text slots in `person` → `shot` → `setting` order, with paragraph separators.
+Import it as `@hypit/gpt-image-kits/phone-ugc-v1`.
For creative direction, see the Image direction page in the Hypit Skill.
@@ -13,17 +14,20 @@ selecting or authoring another Text Template. The caller supplies the variable d
| Text slot | Required | Content |
| --- | --- | --- |
-| `shot` | Yes | Camera view, framing, posture, gaze and interaction. |
-| `direction` | Yes | Subject, styling, setting and intended appearance. |
-| `references` | No | What each connected reference supplies or changes. |
+| `person` | No | Person or cast, appearance and styling; identity references and what they preserve. |
+| `shot` | Yes | Camera view, framing, posture, gaze, action and interaction with people or props. |
+| `setting` | No | Surrounding place, palette, structures and objects; scene references and what they preserve. |
-Each slot accepts an ordinary Text Output and may contain several paragraphs. The template produces
-Text; the image Surface consumes it. Actual reference Resources and model parameters are connected
-on that Surface.
+For a complete new character-and-scene image, supply all three as ordinary paragraphs. A derived
+view may inherit its person or setting from connected references and omit those slots. Omitted
+blocks add no empty paragraphs. State the necessary reference responsibilities in the slot they
+affect. Each slot accepts an ordinary Text Output and can describe more than one person or relationship.
+The template produces Text; the image Surface consumes it. Actual reference Resources and model
+parameters are connected on that Surface.
## Assemble the prompt
-Given authored Text Outputs `portrait-shot` and `portrait-direction`, this fragment assembles the
+Given authored Text Outputs `portrait-person`, `portrait-shot` and `portrait-setting`, this fragment assembles the
prompt and passes it to `gpt:Image`:
```svml
@@ -32,13 +36,14 @@ prompt and passes it to `gpt:Image`:
<import as="ugc" source="@hypit/gpt-image-kits/phone-ugc-v1"/>
<text:Render id="portrait-prompt" template={ugc.phone-ugc-v1}>
+ <text:Set name="person" text={portrait-person}/>
<text:Set name="shot" text={portrait-shot}/>
- <text:Set name="direction" text={portrait-direction}/>
+ <text:Set name="setting" text={portrait-setting}/>
</text:Render>
<gpt:Image id="portrait" prompt={portrait-prompt} aspect-ratio="9:16" resolution="2K"/>
```
-To explain reference inputs, add a `text:Set` named `references` to the Render. Connect those same
-images through `gpt:Reference` children on `gpt:Image`. Omit the optional Set when no reference
-direction is needed. The Text does not create Resource bindings.
+Connect reference images through `gpt:Reference` children on `gpt:Image` in the order described by
+the prompt. For example, a `person` paragraph can inherit identity from reference 1 while `shot`
+describes holding the product from reference 2. The Text does not create Resource bindings.
diff --git a/packages/gpt-image-kits/kits/phone-ugc-v1.svs b/packages/gpt-image-kits/kits/phone-ugc-v1.svs
index 0f306cab..ba316ace 100644
--- a/packages/gpt-image-kits/kits/phone-ugc-v1.svs
+++ b/packages/gpt-image-kits/kits/phone-ugc-v1.svs
@@ -8,27 +8,27 @@
text-template.phone-ugc-v1.block.capture {
kind: fixed;
order: 10;
- text: "A photograph with the texture of real iPhone footage, captured as a single frame from a video actually shot on an iPhone. The image looks real, without an oily, overprocessed finish, and has the texture of video footage. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, the lighting is natural, and the image is coherent and free of visual artifacts.";
+ text: "A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.";
}
- text-template.phone-ugc-v1.block.shot {
+ text-template.phone-ugc-v1.block.person {
kind: slot;
order: 20;
- slot: shot;
- optional: false;
+ slot: person;
+ optional: true;
}
- text-template.phone-ugc-v1.block.direction {
+ text-template.phone-ugc-v1.block.shot {
kind: slot;
order: 30;
- slot: direction;
+ slot: shot;
optional: false;
}
- text-template.phone-ugc-v1.block.references {
+ text-template.phone-ugc-v1.block.setting {
kind: slot;
order: 40;
- slot: references;
+ slot: setting;
optional: true;
}
</sheet>
diff --git a/skills/hypit/SKILL.md b/skills/hypit/SKILL.md
index 2e779db5..01ad9a51 100644
--- a/skills/hypit/SKILL.md
+++ b/skills/hypit/SKILL.md
@@ -50,8 +50,10 @@ unfold; semantic anchors locate events that belong to speech. A piece can use bo
Before writing or adapting an image prompt, read
[Image direction](references/playbooks/craft/image-direction.md) and apply it to the intended picture,
-including when starting from an existing project's prompt. For phone-footage images, use its
-recommended paragraph structure and assemble the fixed capture wording through its Kit.
+including when starting from an existing project's prompt. UGC, podcast and street-interview images
+usually share the Kit's fixed iPhone-video capture language. Follow it with Person, Shot and Setting:
+strong, specific casting; a deliberate camera encounter; and a place with a considered palette.
+The Craft owns the explicit appearance, proportion and reusable speaking-view guidance.
Before choosing, designing or changing a character's voice, read
[Voice direction](references/playbooks/craft/voice-direction.md). Before writing or adapting a video
prompt, performance Recipe or action, read [Video direction](references/playbooks/craft/video-direction.md)
diff --git a/skills/hypit/references/creation/reference-video.md b/skills/hypit/references/creation/reference-video.md
index d80dabc4..59d66c57 100644
--- a/skills/hypit/references/creation/reference-video.md
+++ b/skills/hypit/references/creation/reference-video.md
@@ -162,6 +162,10 @@ duration. Use [transformations](transformations.md) to think through that adapta
Selections for meaningful spans and Moments for events in the target Script, and let the accepted
performance establish their time. [Script and time](script-and-time.md) owns that authoring language.
+For generated camera imagery, [image direction](../playbooks/craft/image-direction.md#compress-the-idea-into-decisive-anchors)
+turns the observed appearance into a coherent styling and scene direction, with the decisive details
+the target needs. The reference notes retain the observations behind those choices.
+
The recorded account should explain the whole piece from opening to close and make its distinct
visual systems and their changes locatable. It should be concrete enough to direct the new work:
what to preserve or adapt, how it is expressed, and why it belongs. Continue investigating an
diff --git a/skills/hypit/references/playbooks/craft/examples/conversation-images.md b/skills/hypit/references/playbooks/craft/examples/conversation-images.md
index 58090da6..a2efe398 100644
--- a/skills/hypit/references/playbooks/craft/examples/conversation-images.md
+++ b/skills/hypit/references/playbooks/craft/examples/conversation-images.md
@@ -11,18 +11,15 @@ separately and imported as files.
Use `phone-ugc-v1` from `@hypit/gpt-image-kits` for this photographic language. Its fixed paragraph is:
```text
-A photograph with the texture of real iPhone footage, captured as a single frame from a video
-actually shot on an iPhone. The image looks real, without an oily, overprocessed finish, and has
-the texture of video footage. The background is clearly visible, with no depth-of-field blur.
-Skin texture is natural and fine, the lighting is natural, and the image is coherent and free
-of visual artifacts.
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
```
-Each direction below follows that paragraph. When using the Kit, put its framing sentences in `shot`,
-its scene and subject direction in `direction`, and any reference responsibilities in `references`.
-The Kit supplies the capture paragraph once. Set the output aspect ratio, such as `9:16`, on the
-model; do not spend prompt attention repeating it. Without the Kit, prepend the paragraph to the
-selected direction. Connect the stated reference images as actual model inputs.
+When adapting these production directions with the current Kit, arrange person and appearance in
+`person`, the camera encounter in `shot`, and the surrounding place in `setting`. Include reference
+responsibilities where they apply. A derived view can stay brief where its references already
+establish the facts; the `person` and `setting` blocks can be omitted. The Kit supplies Capture once.
+Set the output aspect ratio, such as `9:16`, on the model. Connect the stated reference images as
+actual model inputs. The historical directions below retain their production wording and order.
These renditions preserve complete production directions as evidence, including additional
time-of-day, brightness, exposure, or lighting language that belonged to those particular shots.
diff --git a/skills/hypit/references/playbooks/craft/examples/image-direction.md b/skills/hypit/references/playbooks/craft/examples/image-direction.md
index 6916147b..1ab8208c 100644
--- a/skills/hypit/references/playbooks/craft/examples/image-direction.md
+++ b/skills/hypit/references/playbooks/craft/examples/image-direction.md
@@ -1,188 +1,111 @@
# Image direction examples
-Read these with [Directing generated images](../image-direction.md) for complete phone-video image
-directions and visual examples.
+Read these with [Directing generated images](../image-direction.md). Each complete prompt uses
+Capture, Person, Shot and Setting. The Kit supplies Capture and assembles `person`, `shot` and
+`setting` in that order; the [Craft](../image-direction.md#use-a-concrete-capture-direction) explains
+assembly. For these full-screen portraits, set `aspect-ratio="9:16"` and usually `resolution="2K"`
+on `gpt:Image`.
-The first paragraph is the Kit's fixed capture direction. The rest describes the picture in ordinary
-language and can be assembled through its `shot` and `direction` slots. Use `aspect-ratio="9:16"`
-and usually `resolution="2K"` on `gpt:Image` for these full-screen portraits. The [Craft](../image-direction.md#use-a-concrete-capture-direction)
-shows the Kit assembly. Existing production wording, including particular lighting choices, remains
-with the result it describes.
+These production prompts belong to the accepted balcony portrait and five reviewed character
+portraits. They retain the same capture language while choosing the person, framing and setting
+for each picture: soft goth, clean girl, coquette, Harajuku decora and streetwear. The styling terms
+organize a particular appearance; the selected details make that direction specific.
-## Goth girl outside a Korean cafe
-
-![Goth presenter outside a Korean café, with a wooden shopfront, red architectural details, plants and grape bingsu](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/v1/goth-korean-cafe.webp)
-
-Production example: the prompt below was used directly to generate the shown image.
-
-The intended picture is a striking, sweet-but-rebellious goth girl casually talking to the viewer
-outside a Korean café. Idol beauty sets the strength of attraction; blunt bangs, feline makeup,
-jewelry and a spiked cap give it a particular character. Textured wood, coral-red accents and plants
-establish the place and palette. The half-eaten bingsu on the left both balances her rightward position
-and suggests a café visit already underway.
+## Japanese sailor on an apartment balcony
-In the result, wood belongs to a shopfront with openings and a recessed entrance. Sidewalk
-seating, plants and the foreground dessert establish different distances and a café in use. The red
-architectural detail and green foliage give color a physical role. Much of the background is still
-wall: its construction and relationship to the rest of the place make it convincing. These are useful
-qualities to seek in another setting, with the few spatial anchors that belong to that new view.
+The user selected the result generated from this exact prompt. The Person paragraph gives a
+specific casting direction, strong beauty, broad shoulders and head-to-shoulder proportions, with
+long straight hair and sailor styling. Shot establishes the frontal speaking state, free-hand
+gestures and face placement. The table gives the offset composition a physical relationship.
-The [four paragraph roles](../image-direction.md#compress-the-idea-into-decisive-anchors) are visible
-here. In the Kit, use the second paragraph for `shot` and the final two for `direction`.
+In the selected result, navy clothing, terracotta architecture and plants have distinct color roles.
+The glass door, railing and neighboring buildings establish a place with different distances.
+The model supplied individual tiles, leaves, reflections and clothing details within the authored
+direction; those incidental arrangements are not additional requirements for another portrait.
Production prompt:
```text
-A photograph with the texture of real iPhone footage, captured as a single frame from a video actually shot on an iPhone. The image looks real, without an oily, overprocessed finish, and has the texture of video footage. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, the lighting is natural, and the image is coherent and free of visual artifacts.
-
-Generate a vertical close half-body photo of an exceptionally beautiful young woman sitting slightly to the right, near the camera, with her body kept within the frame. She holds a handheld microphone and speaks to the camera, her free hand making a conversational gesture. She has a sweet, slightly rebellious charm, as though casually sharing her take with a friend. Her face is in the upper-middle part of the frame, slightly to the right.
-
-She looks Japanese-American and is extraordinarily beautiful, like a top Korean girl-group idol. She has a very small face, very broad shoulders, excellent head-to-shoulder proportions and very fair skin. Her hair is long, black and straight, with blunt bangs. She wears soft goth makeup with a captivating feline look around the eyes, beautiful silver earrings, a choker and layered necklaces. She wears a black goth-style T-shirt and a baseball cap decorated with silver spikes, with HYPIT in uppercase Gothic lettering.
-
-She is seated outside a relaxed street-side café in Korea. Behind her are beautifully textured wooden exterior walls with a pronounced grain, coral-red architectural accents, some green plants and a few other tables. A half-eaten grape bingsu sits on the table in the lower-left corner. Wood brown forms the main backdrop, with coral red and leafy green as accents.
-```
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
-For subsequent reusable frontal speaking images, the
-[idle guidance](../image-direction.md#choose-an-idle-state-for-the-encounter) adds an explicit pose:
-"Her face points straight toward the lens, with no head tilt or rotation."
+A Japanese woman in her early twenties. She is exceptionally, strikingly beautiful, with the looks of a top Korean girl-group idol and very fair skin. She has beautiful long, straight black hair with blunt bangs and pretty makeup, with bold eye makeup giving her an intensely alluring gaze. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a sweet-but-edgy navy-blue sailor-style outfit with white trim and a matching sailor cap.
-### A centered standing version
+Generate a vertical medium half-body shot. She sits on a chair, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper-right part of the frame. A small wooden table beside her enters the lower-left corner.
-This variation demonstrates the paragraph roles and has not been generated.
-
-Keep the capture and casting paragraphs above. The person can keep her strong appeal, proportions
-and speaking attitude while a new encounter paragraph places her standing in the center. Update the
-place paragraph to match that stance. A close half-body crop still suits this speaking image.
-
-Replacement encounter and place paragraphs:
-
-```text
-Generate a close half-body photo of an exceptionally beautiful young woman standing in the center of the frame, comfortably near the camera, with her shoulders and upper body visible. Her face points straight toward the lens, with no head tilt or rotation. She holds a handheld microphone and speaks to the camera, her free hand gesturing naturally. She has a sweet, slightly rebellious charm, as though casually sharing her take with a friend.
-
-She stands outside a relaxed street-side café in Korea. The wooden shopfront has a recessed entrance, coral-red architectural details and green plants beside the doorway. A café table beside her holds a half-eaten grape bingsu, with a few other seats farther along the pavement.
+She is on an apartment balcony. Behind her are a sliding glass door, textured exterior tiles, metal railings and green plants, with neighboring buildings visible beyond the balcony. The background color scheme is terracotta and green.
```
-The four paragraphs keep their roles; the scene determines their particular framing and objects.
-
-## Shanghai record-store owner
-
-![Shanghai record-store owner: green knit, bob, and headphones in a working record shop](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/v1/shanghai-record-store.webp)
-
-Here the striking beauty and cool music-lover presence have a different expression: a bob, jade-green
-knit, headphones, and a few record-shop details. Green clothing and walnut shelves distinguish the
-person from the setting; the lower-left turntable balances her rightward placement. Recognizing a
-record from its first two seconds gives her confidence a vivid social meaning.
+The frontal face and speaking gesture give a reusable direct-to-viewer state. The upper-right face
+and lower-left table are choices for this composition. A centered standing view keeps the casting
+and proportion direction while changing posture, placement and the scene relationship. A podcast
+view instead directs attention toward the partner. See the [idle guidance](../image-direction.md#choose-an-idle-state-for-the-encounter).
-The complete prompt below belongs to this particular portrait. Carry that decisiveness into another
-character while choosing the face, clothing, setting, and colors that suit the new image.
+## Goth girl outside a Korean cafe
-Complete prompt:
+Soft goth styling gives the person a clear visual character within an ordinary café setting. The grape bingsu and nearby table connect her rightward placement to a visit already underway.
```text
-A photograph with the texture of real iPhone footage, captured as a single frame from a video actually shot on an iPhone. The image looks real, without an oily, overprocessed finish, and has the texture of video footage. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, the lighting is natural, and the image is coherent and free of visual artifacts.
-
-Generate a half-body image of a Chinese woman in her late twenties inside the independent record shop she owns in Shanghai. She sits slightly to the right, comfortably near the camera, with her face level and directed toward it. A turntable on a low cabinet enters the lower-left part of the frame beside her chair. Her shoulders and upper body remain clearly visible.
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
-She holds a small handheld microphone while speaking to the camera, her free hand gesturing naturally. She has the effortless confidence of someone who recognizes a record from its first two seconds. She is exceptionally beautiful, with the striking appeal of a top girl-group idol and the cool presence of an independent film actress. She has a small face, broad shoulders, excellent head-to-shoulder proportions, and beautifully cared-for skin.
+A Japanese-American woman in her early twenties. She is exceptionally, strikingly beautiful, with the looks of a top Korean girl-group idol and very fair skin. She has beautiful long, straight black hair with blunt bangs and soft goth makeup, with bold eye makeup giving her an intensely alluring gaze. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a sweet-but-edgy black beret, a black goth-style outfit and a few pieces of silver jewelry.
-Her hair is cut into a neat chin-length bob. She wears a jade-green knitted polo, small silver hoop earrings, and a pair of over-ear headphones resting around her neck. Behind her are walnut record shelves, a few displayed album sleeves, and a handwritten recommendation card tucked into a record bin.
+Generate a vertical medium half-body shot. She sits on a chair, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper-right part of the frame. A small wooden table beside her enters the lower-left corner, with a half-eaten grape bingsu on it.
-The jade-green clothing gives her definition against the walnut surroundings. The shop has ordinary interior lighting: her face is clearly lit, while the shelves retain natural shadows and readable detail. It feels like a favorite local record-store owner taking a moment to tell you which album you should take home.
+She is seated outside a street-side café in Korea. Behind her are entrance steps, a wooden shopfront, metal railings, green plants and a few other tables. The background color scheme is navy blue and yellow.
```
-### The record-store shot with a supplied presenter
+## Clean Girl in a Korean cafe
-Imagine the user supplies a portrait of an adult woman with a bob, glasses, and broad shoulders,
-and asks to appear in the record-store setup. Her photo supplies identity; the styling and direction develop her into a striking,
-music-literate host. The capture language, strong appeal, comfortable framing, and coordinated palette
-carry forward. Her glasses become part of that composed presence.
-
-Connect the portrait as the first `gpt:Reference` image. The reference sentence below can live in the
-Kit's `references` input; the rest continues to use `shot` and `direction`.
+Clean-girl makeup, a sleek ponytail and navy knit give the person a clear style. The wooden counter and iced latte connect the seated composition to the café; pistachio green and walnut support the navy outfit.
```text
-A photograph with the texture of real iPhone footage, captured as a single frame from a video actually shot on an iPhone. The image looks real, without an oily, overprocessed finish, and has the texture of video footage. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, the lighting is natural, and the image is coherent and free of visual artifacts.
-
-Generate a half-body image of the presenter inside an independent record shop in Shanghai. She sits slightly to the right, comfortably near the camera, with her face level and directed toward it. A turntable on a low cabinet enters the lower-left part of the frame beside her chair. Her shoulders and upper body remain clearly visible.
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
-She holds a small handheld microphone while speaking to the camera, her free hand gesturing naturally. She has the effortless confidence of someone who recognizes a record from its first two seconds. She looks exceptionally beautiful and charismatic, with the cool, discerning presence of an independent film actress. Her broad shoulders and excellent head-to-shoulder proportions give her a strong presence in the frame.
+A Korean-Chinese woman in her early twenties. She is exceptionally, strikingly beautiful, with the looks of a top Korean girl-group idol and very fair skin. She has a sleek high ponytail and beautiful clean-girl makeup, with clear, expressive eyes and a relaxed, confident presence. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a navy-blue knitted top in a minimalist clean-girl style, with small gold earrings and a fine gold necklace.
-Her bob and glasses complement a jade-green knitted polo, small silver hoop earrings, and over-ear headphones resting around her neck. Behind her are walnut record shelves, a few displayed album sleeves, and a handwritten recommendation card tucked into a record bin.
+Generate a vertical medium half-body shot. She sits on a stool beside a café counter, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper-left part of the frame. The wooden counter beside her enters the lower-right corner, with a partly finished iced latte on it.
-The jade-green clothing gives her definition against the walnut surroundings. The shop has ordinary interior lighting: her face is clearly lit, while the shelves retain natural shadows and readable detail. It feels like a favorite local record-store owner taking a moment to tell you which album you should take home.
-
-Use the person in reference image 1 as the presenter, preserving her recognizable identity.
+She is in a neighborhood café in Seoul. Behind her are wooden shelves, a tiled counter, an open doorway and green plants beside a window. The background color scheme is pistachio green and walnut brown.
```
-Another supplied person may invite warmer energy, sharper humor, or different styling and colors.
-Look at that image and direct its appeal within the target work, using the same image craft.
-
-## Mob Wife: strong glamour in a captured world
-
-![Mob Wife: strong glamour in a captured world](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/v1/mob-wife.webp)
-
-The beauty comparison and "Stay out of my husband's business" give this person a vivid presence.
-Fur, layered gold, red lips, and an expensive restaurant reinforce it. Simple material names leave
-the model room to render their ordinary surfaces. Red and gold establish the room's atmosphere.
-The deep-red nails and dark-brown hair below preserve the production prompt that belongs to this
-example. Apply the current palette guidance when directing a new image.
+## Coquette girl in a bay-window bedroom
-Complete prompt:
+Coquette styling is carried by the hair bow, doe-eyed makeup and pink lace. A bay-window alcove supports the centered seat, while green and wood give the pink clothing a distinct place in the palette.
```text
-A photograph with the texture of real iPhone footage, captured as a single frame from a video actually shot on an iPhone. The image looks real, without an oily, overprocessed finish, and has the texture of video footage. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, the lighting is natural, and the image is coherent and free of visual artifacts.
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
-Generate a close half-body image of a Latina woman in her thirties, seated slightly to the right and near the camera. Keep her body within the frame. Her face is directed straight ahead without any tilt or rotation, and the lower part of her body is visible and unobstructed.
+An African-American woman in her early twenties. She is exceptionally, strikingly beautiful, with the looks of a young Hollywood film star and beautiful brown skin. She has long black waves with a pink ribbon bow and pretty coquette makeup, with captivating doe eyes and rose-colored lips. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a pink lace-trimmed top in a romantic coquette style, with a pearl necklace.
-She holds a handheld microphone in one hand and speaks to the camera, while her other hand gestures. Her commanding attitude says, “Stay out of my husband's business.” She is exceptionally beautiful, with the intense beauty of a top Italian actress in her youth. She has broad shoulders, excellent head-to-shoulder proportions, and a larger-than-life presence.
+Generate a vertical medium half-body shot. She sits in an upholstered chair in a bay-window alcove, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper center of the frame. The curved window alcove frames her chair.
-She wears a leopard-print fur coat over a black low-cut top, several chunky gold chains layered around her neck, and large gold hoop earrings. Her long nails are deep red. Her dark-brown hair falls in voluminous waves; she wears classic red lipstick and bold but clean eye makeup. She sits in a leather booth in an expensive-looking Italian restaurant. An oil painting hangs on the wall behind her, a bar is visible farther back, and candles stand on a nearby windowsill.
-
-The restaurant's background palette is mainly red and gold. It feels like an upscale private dining room where a mob wife meets her girlfriends for a meal.
+She is in an apartment bedroom. Behind her are a bay window with floral curtains, a wooden wardrobe and a small vanity with a vase of roses. The background color scheme is sage green and walnut brown, with pink accents.
```
-## Orange-cat CEO: comic authority through body and setting
-
-![Orange-cat CEO: comic authority through body and setting](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/v1/orange-cat-ceo.webp)
+## Harajuku Decora girl outside a fashion shop
-The boss's authority organizes the picture. A round face, large belly, crooked tie, straining suit
-buttons, and undersized cat tree make that authority funny. A leather chair, office window, laptop,
-and mug establish the workplace; gray and orange establish the palette.
-
-Complete prompt:
+Decora supplies the styling idea; pink double buns, a graphic hoodie and a few accessories make it specific. The shop entrance gives the centered standing view a natural frame.
```text
-A photograph with the texture of real iPhone footage, captured as a single frame from a video actually shot on an iPhone. The image looks real, without an oily, overprocessed finish, and has the texture of video footage. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, the lighting is natural, and the image is coherent and free of visual artifacts.
-
-Generate a close half-body image of a plump orange cat, seated slightly to the left in a large leather chair and near the camera. Keep its body entirely within the frame. Its face is directed straight ahead without any tilt or rotation, and the lower part of its body is visible and unobstructed.
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
-The cat wears a three-piece suit. One front paw rests on a small handheld microphone as it faces the camera, while the other gestures. Its expression carries the born-to-be-the-boss authority of an orange cat.
+A Japanese woman in her early twenties. She is exceptionally, strikingly beautiful and cute, with the looks of a top Japanese girl-group idol and very fair skin. She has pink hair in two high buns, a few star-shaped hair clips and pretty, playful makeup, with large, captivating eyes. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a lilac graphic hoodie in a playful Harajuku decora style, with colorful beaded jewelry.
-Its tie is crooked and its suit buttons strain around its large belly. It has classic orange-and-white tabby stripes, a round face, and long whiskers. Its whole body rests back in the leather chair. Behind it is a boss's office with a floor-to-ceiling window overlooking the city skyline. A desk to its side and rear holds a MacBook and a mug marked “BIG BOSS”; a cat tree in the corner is plainly too small for its owner.
+Generate a vertical medium half-body shot. She stands outside a fashion shop, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper center of the frame. The shop's recessed doorway frames where she stands.
-The office's background design combines gray and orange textures. It looks like a standard boss's office.
+She is on a shopping street in Harajuku, Tokyo. Behind her are a shop window displaying bags and plush toys, capsule-toy machines and neighboring storefronts farther along the pavement. The background color scheme is turquoise and yellow.
```
-## Mermaid: an impossible subject in a photographed world
+## Streetwear guy beside a basketball court
-![Mermaid: an impossible subject in a photographed world](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/v1/mermaid.webp)
-
-The mermaid identity, explicit beauty, and friendly curiosity direct the person. Wet hair, shells,
-pearls, coral, and fish establish the underwater world. Sea blue and coral orange give it a color
-relationship. The final sentence reaffirms that she appears to be physically sitting underwater.
-
-Complete prompt:
+A hip-hop-star beauty comparison, short twists and relaxed streetwear establish the person. The bench and open court explain the offset framing, with terracotta and teal giving the surroundings a deliberate palette.
```text
-A photograph with the texture of real iPhone footage, captured as a single frame from a video actually shot on an iPhone. The image looks real, without an oily, overprocessed finish, and has the texture of video footage. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, the lighting is natural, and the image is coherent and free of visual artifacts.
-
-Generate a close half-body image of a mermaid seated in the center and near the camera. Keep her body within the frame. Her face is directed straight ahead without any tilt or rotation, and the lower part of her body is visible and unobstructed.
-
-She holds a handheld microphone in one hand and speaks to the camera, while her other hand gestures. She has the curious, friendly presence of an underwater princess, as though she is teaching people on land about ocean life. She is exceptionally beautiful, with the delicate beauty of a mermaid princess, broad shoulders, excellent head-to-shoulder proportions, and graceful upper-body lines.
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
-Scales with a sea-blue gradient are visible on her neck and shoulders, with small fin-like decorations beside her ears. Her long, wet sea-blue-green hair is threaded with a few small shells and pearls. She wears a top made from shells and starfish and a necklace of real pearls. She sits on a coral rock on the seabed. Behind her is an underwater world with colorful coral colonies, a few small fish swimming past, and shafts of light filtering down from the water's surface farther away. A few shells lie scattered on the sandy seabed.
+An African-American man in his early twenties. He is exceptionally, strikingly handsome, with the looks of a young hip-hop star, brown skin and a magnetic, confident presence. He has short twists beneath a backward baseball cap and expressive eyes. He has an athletic build, very broad shoulders and excellent head-to-shoulder proportions. He wears relaxed New York streetwear: an oversized navy-blue graphic T-shirt and a silver chain.
-The background palette combines sea blue and coral orange, like a shallow-water coral reef reached by sunlight.
+Generate a vertical medium half-body shot. He sits on a courtside bench, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with his free hand gesturing. His face points straight toward the lens, with no head tilt or rotation, and sits toward the upper-left part of the frame. The bench extends beside him toward the open court on the right.
-The image looks real, as though this woman is actually sitting underwater.
+He is beside an outdoor neighborhood basketball court in New York. Behind him are the painted court, a chain-link fence, a mural and trees beyond the fence. The background color scheme is terracotta and teal.
```
diff --git a/skills/hypit/references/playbooks/craft/image-direction.md b/skills/hypit/references/playbooks/craft/image-direction.md
index 4dbbb9e4..e6029eb7 100644
--- a/skills/hypit/references/playbooks/craft/image-direction.md
+++ b/skills/hypit/references/playbooks/craft/image-direction.md
@@ -1,249 +1,229 @@
# Directing generated images
-Read this when creating a person, place, product, B-roll image, or camera view for a video.
-An image establishes the world that a reference-conditioned video model can bring to life.
+Read this before writing or adapting an image prompt, including one inherited from another project.
+An image establishes the person and world that a reference-conditioned video can bring to life.
-For the phone-shot social videos Hypit most often makes, the useful direction is compact and vivid:
-strong character appeal, a recognizable cultural or social presence, a clothing vibe, a few telling
-objects, and a clear color relationship. A short direction can be very specific; its force comes
-from what it chooses.
-Aim for the visible credibility of actual phone footage, including with invented people and
-fantastical subjects. Judge the generated picture against that photographic goal.
+UGC, podcast and street-interview work commonly share the same photographic foundation: the visible
+credibility of a frame from real iPhone video, with an appealing person and a readable environment.
+Use the full capture paragraph in `@hypit/gpt-image-kits/phone-ugc-v1`. Direct the particular person,
+camera relationship and setting around that foundation. A change of format usually changes those
+relationships while retaining the capture language. An invented or fantastical person can inhabit
+this photographed world too.
## Compress the idea into decisive anchors
-Think from the intended picture down to the choices that make it visible. What will this image do
-in the video? What makes the person compelling, and what encounter are we looking at? Let that
-whole-picture intention guide the casting, attitude, styling, setting, color and camera relationship.
-The prompt expresses those decisions in the language of a picture.
-
-Compression preserves the force of the idea. "As beautiful as a top Korean girl-group idol" carries
-a strong casting direction; "sweet, slightly rebellious, casually sharing her take with a friend"
-gives a speaking presence. Blunt bangs, a spiked cap and a goth T-shirt make that presence specific.
-A metaphor or cultural reference is useful when it evokes visible qualities the model can develop.
-Choose one that serves this person, then add the few details that distinguish this picture.
-
-Give each anchor the scope it needs: goth can describe the girl's styling while a Korean café
-supplies the setting. A few objects can establish the encounter and its colors together. Let the
-model complete incidental detail from those choices. Keep precise facts that the work depends on:
-recognizable identity, an exact product, HYPIT lettering, gaze toward a partner, or room for an icon.
-"Beautiful composition" would leave an intended rightward placement unstated.
-
-Judge the wording by how clearly it evokes the intended image. Preserve the full chosen capture
-paragraph and strong attraction language; omit detail that adds no useful direction. Group related
-choices so they are easy to revise. The [café example](examples/image-direction.md#goth-girl-outside-a-korean-cafe)
-shows one compact arrangement; a derived view can say even less when its reference already supplies
-the person and world.
-
-For these phone-video portraits, use this recommended four-paragraph structure. Keep the capture
-wording intact; write the other three with the few anchors this picture needs:
-
-| Paragraph | Directing question | What to express |
+Study a reference as a coherent visual impression. Understand how the person, styling, camera
+relationship and setting create its appeal, and which details carry that character. Keep useful
+observations in the reference notes; distill them into the direction the new picture needs. Careful
+observation makes concise prompting possible: the detail earns its place through what it contributes.
+
+For social-video portraits, look first to Gen-Z styling and social aesthetics for language that
+captures the closest-fitting look and presence. Soft goth, clean girl, coquette or relaxed streetwear
+can organize hair, makeup, clothing and attitude together. Use the words that express this particular
+person, combining descriptions where that is more accurate than a named style. The user's goal and
+references still determine identity, age and defining features; the styling language articulates
+the appearance to preserve or adapt. Where the brief invites new casting, it also offers a useful
+starting point for an original look.
+
+Make the choices that determine the picture explicit. Cultural background, age, intended complexion,
+strength of attraction, a recognizable styling idea and the setting's palette can each change the
+whole direction. Select them deliberately instead of leaving the important decisions at “a young
+woman in a nice room.” A supplied reference can establish the facts it already contains.
+
+A muscular Barbie-like woman, a nerdy student or a sailor-style outfit can also organize visible
+qualities in a few words. “Creator” and “founder” describe a role; alone they give little direction
+about appearance. Keep a role when it belongs to the work, and still cast and style the person.
+
+Use specific detail where it preserves identity, expresses a useful relationship or settles a key
+structure: exact product features, HYPIT lettering, a face's placement, or broad shoulders and
+excellent head-to-shoulder proportions. A concise direction can be forceful and precise. Give the
+image model the visual intention and decisive anchors, and let it realize the picture and organize
+incidental detail within that direction.
+
+When adapting a reference, recover its visual idea before rewriting its details. Relaxed collegiate
+styling may carry what matters about a shirt; the exact fraction of forearm exposed by its rolled
+sleeve may be incidental. If the sleeve exposes a featured watch, preserve that relationship. Keep
+the details the new work depends on, and let its new person and camera realize the shared style.
+More painstaking description does not by itself make the direction more faithful or attractive.
+
+## Four paragraphs, four responsibilities
+
+For a complete character-and-scene camera image, use this order:
+
+| Paragraph | Question | Direction |
| --- | --- | --- |
| Capture | What kind of photographed image is this? | The Kit's full fixed iPhone-video paragraph. |
-| Encounter | How is this person relating to the viewer or another person? | Framing, centered or offset placement, standing or seated posture, a useful idle state, speaking or listening action, and a vivid social attitude, chosen for this scene. |
-| Person | Why would the audience want to watch this particular person? | Explicitly strong attraction, cultural or social character, intended proportions, and a few distinctive hair, makeup and wardrobe choices. |
-| Place | What actual place does this camera view belong to? | A recognizable setting, its visible construction, a few signs of use, and colors carried by its materials and objects. |
+| Person | Who should we want to watch? | Cultural background and age, strong attraction and a cultural or aesthetic comparison for that appeal, complexion, defining appearance, proportions and styling. |
+| Shot | How does the camera see them, and what are they doing with whom? | Framing, distance, posture, face direction, placement, speech, gestures and useful relationships with props or other people. |
+| Setting | What place surrounds them? | A recognizable place, overall palette, and a few useful structures, materials or objects. |
-Think about these relationships together. The encounter determines gaze; the person's clothing
-participates in the place's palette; a side table can explain the composition. Fluent paragraphs
-carry those choices. The Kit assembles the encounter as `shot` and the person and place as
-`direction`; reference responsibilities remain in its optional `references` text.
+These are paragraph responsibilities, not a form requiring a sentence for every possible attribute.
+A hairstyle name can settle the hair; a styling idea and a memorable item can settle the outfit.
+Related information can share a sentence. Keep each paragraph focused while considering the picture
+together: wardrobe participates in the palette, and a table can explain an offset composition.
-## Make the person immediately compelling
+The [balcony example](examples/image-direction.md#japanese-sailor-on-an-apartment-balcony) preserves
+a successful complete prompt. Use it to see the amount and kind of direction that
+produced the result, rather than turning every incidental detail in the image into another requirement.
-For an invented social-video presenter, actively cast someone the audience will want to watch.
-State the strength and kind of attraction directly: exceptionally beautiful, stunningly handsome,
-Korean girl-group idol beauty, a commanding mob wife, or a round orange cat with a boss's authority.
-Heritage, subculture, profession, and attitude can quickly give that appeal a particular character.
-Choose the ones that make this person come to mind.
+## Use a concrete capture direction
-For an attractive invented human presenter in this style, explicitly direct a small face,
-**broad shoulders and excellent head-to-shoulder proportions**. These choices carry across seated,
-standing, centered and offset views. Narrow, pinched shoulders under a large-looking head can make
-the portrait feel doll-like. Preserve the physique of a supplied person or a deliberately different
-character, developing the appeal of that person.
+The Kit preserves the chosen capture paragraph verbatim. It establishes a frame from actual iPhone
+video, the photographic surface, a clearly visible background, natural fine skin texture, natural
+lighting and a coherent image. Carry this wording intact into phone-footage images; later paragraphs
+direct this picture's content. Adding sensor grain, exposure falloff or elaborate lighting language
+is an additional aesthetic choice, not a necessary completion of the capture paragraph.
-A few visible choices can carry the look: blunt bangs and long straight black hair, feline eye
-makeup, a spiked cap, and a goth T-shirt. Describe the clothing vibe and its memorable details.
-Exceptional beauty and strong styling fit naturally inside real phone footage.
+GPT Image 2 is the usual choice. Favor `2K` for a full-screen picture or video reference; `1K` often
+suffices for an inset or small graphic. Set aspect ratio and resolution on the model's parameters.
+The prompt concentrates on composition inside that shape. For an illustration, product photograph
+or other explicitly different visual form, select capture language that serves that goal.
-Keep the language forceful when revising. "As beautiful as a top Korean girl-group idol" directs
-appearance; "idol-level polish" directs grooming. Preserve the intended beauty, attitude and cultural
-meaning when carrying a successful direction into a new prompt.
+The Kit assembles `person`, `shot` and `setting` after the fixed capture paragraph:
-Vibe is useful when it brings a person or scene immediately to mind: "Stay out of my husband's
-business," "a cat with born-to-be-the-boss authority," or "she recognizes a record from its first
-two seconds." Write that feeling into the image. Production reasoning belongs in the Treatment;
-the prompt gives the model a person, an action, and a world it can picture.
+```svml
+<import as="text" from="@hypit/text@1"/>
+<import as="ugc" source="@hypit/gpt-image-kits/phone-ugc-v1"/>
-## Establish the place with a few details
+<text:Render id="portrait-prompt" template={ugc.phone-ugc-v1}>
+ <text:Set name="person" text={portrait-person}/>
+ <text:Set name="shot" text={portrait-shot}/>
+ <text:Set name="setting" text={portrait-setting}/>
+</text:Render>
+```
-Give the camera a view into a recognizable, constructed place. A wooden café shopfront with a
-recessed entrance and tables along the pavement implies surfaces, depth and a way the place is used.
-In a record shop, shelves turning into an aisle and a nearby browsing bin can do the same work.
-Choose the few features that make this particular view come to mind.
-
-Surface texture and spatial structure contribute different things. Wood grain or plaster texture can
-give a surface variation while the whole background still reads as one flat plane. A door recess,
-window bay, shelving return or relationship between furniture and the room can establish how that
-space continues. Describe the useful feature directly. A broad wall can belong naturally to this
-view, with its material, edges and relationship to the surrounding architecture.
-
-The setting can support, contrast with, or simply coexist with the person's style. A striking goth
-girl can sit in an ordinary classroom. A glamorous woman can sit in a restaurant. Ordinary places
-often make a distinctive person more striking. Choose surroundings that suit the intended encounter;
-styling alone does not determine a person's profession, room, or lifestyle.
-
-A useful detail can do several jobs: a table on the left balances a person on the right and gives
-the scene a nearby surface; a half-eaten dessert suggests a visit already underway and carries a
-color accent; an undersized cat tree makes a cat CEO funny.
-
-## Give the frame a clear color relationship
-
-Consider the person and background together: skin, hair, clothing, furniture, and the surfaces around
-them. Choose which colors occupy the larger areas, which support them, and which appear as small
-accents. Give those colors real places in the image. Coral plaster walls, an oak desk, a group of blue
-posters, and a small green plant describe both a palette and a room.
-
-Consider the relative areas of color, the variety in detail, and where attention goes in the whole
-image. A wooden shopfront, a red awning and leafy plants give colors both a role and a physical form.
-Larger, quieter surfaces can sit beside smaller areas of detail. Color contrast and material changes
-can give the person prominence. Express that balance through the chosen clothing, construction and
-objects.
-
-Name hues without adding brightness instructions to them. Purple, coral red, blue, and green already
-give direction; casual additions such as "bright," "deep," "dark," or "light" can change the overall
-image when only a hue was intended. The Kit supplies the natural-lighting premise. Add a particular
-time, weather condition, or light event when the intended scene actually needs it.
-
-For clean phone-footage realism, build atmosphere through the scene's colors, materials and character.
-Wood grain, fabric weave and foliage give actual objects tangible detail. Added "fine sensor grain"
-asks for noise across the image; "natural falloff" and "dim" or "dark" lighting directions can push
-the picture toward gray, muddy shadows. These choices can undermine the clarity of this photographic
-style. Let the fixed capture paragraph establish its natural lighting and video texture, and put the
-scene's distinctive detail into the people, place and objects.
-
-For this photographic style, prefer colored, textured materials over broad white or flat gray areas:
-colored plaster, wood, fabric, and foliage can supply surface variation and a more deliberate palette.
-Choose what complements the subject. A texture adjective alone does little for a weak color pairing.
-Cream clothing against an extensive white background once washed the person into the setting; navy
-clothing gave her definition. Treat wardrobe and surroundings as a combination.
-
-When exact white, gray, or black belongs to a product, uniform, or chosen outfit, preserve that fact
-and let the surrounding colors and materials support it. A concise material cue can give a large
-surface variation. Ordinary surfaces can otherwise keep their naturally rendered detail.
+These slots take ordinary paragraphs, not individual appearance parameters. Pass `{portrait-prompt}`
+to the image Surface and connect reference images there. For a derived view, a reference may already
+establish person or setting; the Kit permits those blocks to be omitted. State the needed inheritance
+and change in the relevant paragraph. Its package README owns the exact assembly interface.
-## Frame the image for what it will become
+## Make the person immediately compelling
-Choose placement, stance and framing from what this scene needs to show. A centered person can
-address the viewer directly; an offset person can share the picture with a product, a partner or MG.
-Standing or sitting describes posture. A standing presenter can still have a close half-body view,
-and a seated person can appear in a wider composition. Choose each relationship deliberately.
+Begin an invented person's paragraph with the cultural background and age range you intend. State
+the intended complexion when choosing it; a broad identity label need not settle it. A supplied
+person's image and the user's description establish their identity and appearance.
+
+For an invented human presenter in this photographic style, **explicitly state strong beauty or
+handsomeness, with an appropriate visual comparison, and explicitly include broad shoulders and
+excellent head-to-shoulder proportions**. Preserve their force when revising. “Exceptionally,
+strikingly beautiful, with the looks of a top Korean girl-group idol” directs both strength and kind
+of appeal. “Idol-level polish” describes grooming and cannot replace that beauty direction.
+
+The shoulder and proportion wording is a deliberate structural anchor for these half-body portraits:
+pinched shoulders beneath a large-looking head can undermine the intended human presence. Keep the
+anchor across centered, offset, seated and standing views. Develop the appeal of a supplied person
+while preserving their recognizable identity and physique. Animal and established character designs
+use the appealing proportions and defining features that belong to them.
+
+Choose a few appearance anchors under the whole-person idea. Long straight black hair with bangs,
+alluring goth eye makeup and a sweet-but-edgy beret can establish a person without an inventory of
+facial measurements or garment construction. Overall makeup style, a strong eye-makeup emphasis and
+the intended allure can work together. Keep the strength of these choices rather than replacing them
+with general praise such as “well presented.”
+
+Attraction is a positive casting goal. Photographic credibility does not call for automatically
+adding fatigue, rough skin or an unflattering pose. Select a particular appearance because this
+person and work need it. A cultural or styling anchor has a scope: goth can describe the person
+while an ordinary classroom, balcony or café supplies their surroundings.
-For a face-led speaking image, a close half-body view is a useful starting point: make the face and
-gestures easy to read while retaining the shoulders and upper body. Keep a comfortable camera
-distance and natural-looking proportions. Use a wider view when the work needs an outfit, a bodily
-action or more of the surrounding encounter. Describe the intended visible area of the body so the
-crop supports that purpose.
+## Frame the image for what it will become
-Off-center placement can feel natural through the scene itself: put a desk to the person's left,
-or let their thighs angle toward it. Seat and leg details can help establish posture without asking
-for a wider full-body view. Keep the intended visible body within the frame.
+The Shot paragraph makes the encounter visible. For a reusable, direct-to-viewer UGC image, explicitly
+settle these relationships. The wording in the successful example provides a useful baseline:
-Compose for the later video. A Ranking icon on the left may call for a person on the right, with
-their face near the upper right. Describe that placement and the actual room in the image prompt.
-Author the later icon or MG in its own component. Faces, hands, microphones, and products need room
-to remain readable when those layers appear. [Compositing](compositing.md) owns the layer relationship.
+| Relationship | Useful direction and its purpose |
+| --- | --- |
+| Framing and distance | A medium or close half-body shot, fairly close to the camera, makes the face, shoulders and gestures readable. Framing decides visible extent; distance decides the camera relationship. |
+| Posture | Sitting on a chair or standing gives the body a clear arrangement. Choose the posture the scene needs. |
+| Address and activity | Holding a handheld microphone and speaking directly to the viewer establishes an ongoing exchange. Choose a microphone or other prop when it belongs to that encounter. |
+| Accompanying movement | With the free hand gesturing gives the speaking state visible life. Adapt this to the hands and objects involved; an exact finger pose usually adds little. |
+| Frontal face | “Her face points straight toward the lens, with no head tilt or rotation.” Preserve this explicit wording for the reusable frontal speaking view. |
+| Placement | State where the face belongs: for example, toward the upper-right part of the frame. Choose the position for this picture and its later composition. |
+| Scene relationship | A table entering the lower-left corner can balance a person on the right and explain their placement. A seat, partner or body orientation can do this in another encounter. |
+
+Centered, leftward, rightward, seated and standing images all retain deliberate framing and body
+proportions. Plan shared space with later MG through the source camera view. Describe the actual
+person and scene; author the later graphic in its component. An image made for a half-screen or
+circular presentation needs a camera view that serves that crop. [Compositing](compositing.md)
+owns those presentation relationships.
+
+### Choose an idle state for the encounter
+
+Idle is a readable exchange that can continue into different passages. Speaking, gesturing and an
+engaging gaze belong in it. Give the person a useful state from which to perform. An elaborate
+“about to say something outrageous” expression chooses a particular dramatic beat; use such a beat
+when the passage earns it. Natural speaking gestures need no finger-by-finger choreography.
+
+Choose direction from whom the person addresses:
+
+| Encounter | Face and body relationship |
+| --- | --- |
+| Directly addressing the viewer | Keep the reusable frontal view explicit, including no head tilt or rotation, with an engaged speaking state. |
+| Podcast partners | Look toward the partner, keeping the face usefully visible to this camera; preserve complementary positions and microphone relationships. |
+| Interviewer and guest | Preserve the shared encounter and its gaze axis in the two-person image and derived close views. |
-## Choose an idle state for the encounter
+A deliberate story image can show a particular laugh or action. Opening and closing performance
+beats belong in the relevant video passage. The reference supplies visual facts; the selected model
+mode determines whether it is a literal first frame. See [video direction](video-direction.md) and
+[conversation images](examples/conversation-images.md).
-For a camera image reused across speaking Takes, idle means a comfortable, readable state from which
-the person can keep performing different passages. Give them a recognizable attitude, an engaged
-speaking or listening presence, and a natural relationship to their microphone or prop. "Holding a
-handheld microphone and speaking, with the free hand gesturing naturally" already brings the image
-to life. Save a particular punchline reaction for the passage that earns it.
+## Establish the place with a few details
-Choose face and body direction from whom the person is addressing:
+Begin the Setting paragraph with a recognizable place and give it an overall palette. Add the few
+structures and objects that make this camera view specific. A Korean street-side café, navy blue and
+yellow, with stairs, an entrance, textured walls, plants and railings gives the model a coherent
+world to arrange. The prose can place the palette at the end while it guides the choice of details.
-| Intended camera relationship | Useful reusable image direction |
-| --- | --- |
-| Speaking directly to the viewer | Face the lens straight ahead, with no head tilt or rotation. This keeps the face readable and gives later passages a settled frontal view to grow from. |
-| Talking with a podcast partner | Turn and look toward the partner, keeping the face usefully visible from this camera. Complementary views preserve the pair's screen positions, gaze and microphone relationship. |
-| Interviewer and guest attending to each other | Preserve their shared spatial relationship in the two-person view and its close views. Each person's attention follows the exchange. |
-
-The intended relationship decides the pose, including when a format contains more than one kind of
-shot. Idle leaves room for expression and motion in the generated performance. An image meant to
-establish a particular laugh, glance down or other story moment can show that moment directly.
-For reusable views, direct opening and closing actions in the relevant video passage: a guest can
-look down before noticing the interviewer, then turn toward them. A character-and-scene reference
-supplies visual facts; the selected generation mode determines whether it is a literal first frame.
-See [video direction](video-direction.md) and the
-[conversation images](examples/conversation-images.md) for those relationships.
+Texture and spatial construction contribute different things. Wood grain varies a surface; a doorway,
+balcony railing, shelving aisle or view beyond a window helps establish how a place continues. A wall
+covered with many small notes can remain one flat plane. Choose actual spatial anchors rather than
+adding more decoration merely to make the description detailed. A broad wall can naturally belong to
+a well-chosen view. The person, setting and objects need not all express the same subculture or job.
-## Use a concrete capture direction
+A useful object can carry several relationships: a half-eaten bingsu suggests a café visit underway
+and adds color; the nearby table belongs in Shot when it explains the person's framing or interaction.
+Keep the setting concise enough for the model to complete the incidental architecture and arrangement.
-For this photographic style, use the full iPhone-video capture paragraph from
-`@hypit/gpt-image-kits/phone-ugc-v1`. It describes a frame from video actually shot on an iPhone,
-with natural lighting, fine skin texture, and a clearly readable background. This wording comes
-from successful production use with GPT Image 2. Keep it intact when selecting this Kit; put the
-current picture's framing in `shot` and its character, setting, colors, and vibe in `direction`.
+### Give the frame a clear color relationship
-```svml
-<import as="text" from="@hypit/text@1"/>
-<import as="ugc" source="@hypit/gpt-image-kits/phone-ugc-v1"/>
+Choose the overall palette explicitly and consider it with the person's skin, hair and clothes.
+Give the larger areas a useful relationship and let smaller colors support or accent them. Navy
+clothing, terracotta architecture and green plants can separate the person from the setting while
+belonging to one image. A place and a few material names can carry this direction without assigning
+a separate color to every object.
-<text:Render id="portrait-prompt" template={ugc.phone-ugc-v1}>
- <text:Set name="shot" text={portrait-shot}/>
- <text:Set name="direction" text={portrait-direction}/>
- <text:Set name="references" text={portrait-references}/>
-</text:Render>
-```
+For clean phone-footage images, prefer considered colors and material variation to an unchosen white
+or flat-gray background. Avoid defaulting to broad white clothing against a broad white wall: that
+combination can wash the person into the setting. Wood, tile, fabric and foliage can provide tangible
+variation. The useful choice is the combination; a texture adjective cannot repair a weak palette.
-`shot` and `direction` are required. Omit the optional `references` Set when no reference is used.
-The slots assemble ordinary prompt text. Group related choices into short paragraphs: the café
-example gathers framing and action in `shot`, then casting and styling, followed by setting and
-color, in `direction`. This keeps each relationship easy to see and revise. Pass the resulting
-`{portrait-prompt}` to the image Surface and connect any actual reference images there.
+Keep lighting at the natural premise supplied by Capture unless the scene calls for a particular
+light event. Avoid casually adding dim lighting, dark shadows, exposure falloff or sensor grain to
+make the picture “more real.” Name hues directly when hue is what you intend. A local dark textured
+wall or white trim is a surface choice, not a request to darken or brighten the whole image. Preserve
+exact product and wardrobe colors, and choose their surroundings to support them.
-GPT Image 2 is the usual choice. Favor `2K` for a full-screen picture or video reference; `1K` often
-suffices for an inset or small graphic. Set aspect ratio and resolution on the model's parameters.
-The prompt can concentrate on composition inside that shape. Illustration, animation, product
-photography, and other visual forms can use the capture language appropriate to their own goal.
+## Let references supply the facts they own
-A fantastical subject can still look physically present in this photographed world. For a mermaid,
-wet hair, shells, a coral reef, and a few fish can establish the scene. "She looks as though she is
-really sitting underwater" reaffirms the intended feeling. The [complete examples](examples/image-direction.md)
-show the capture paragraph with the goth café, record shop, mob wife, cat CEO, and mermaid.
+Look at a supplied person before directing them. Their reference establishes identity; the Person
+paragraph develops the styling and appeal this work needs. A parent view can establish the shared
+place; an exact product reference establishes the product. Put these responsibilities in the paragraph
+they affect, with the needed preservation and change. Connect actual reference Resources on the model
+Surface in the stated order; mentioning an image in prose does not supply it.
-## Let references supply the facts they own
+Keep a derived view as short as its inherited facts allow. A close view can change attention and crop;
+a product-holding view can change hands and prop state while keeping the camera and setting. Preserve
+the capture language. Another angle may reveal different surroundings within the same place.
+[Reference relationships](generated-dependencies.md) and [conversation examples](examples/conversation-images.md)
+own these branches; [Transformations](../../creation/transformations.md) owns a swap across the video.
-Look at a supplied face or character before writing. "Use the person in reference image 1" establishes
-identity; direct their attraction, clothing, setting, colors, and camera view with the same care as an
-invented subject. Develop the features and presence that make this particular person compelling while
-keeping them recognizable. Their identity comes from the reference and the user's description.
+## Keep successful direction useful
-A product reference can supply the exact product; a parent camera image can supply a person and room.
-Explain what carries forward and what changes. Another view of the same place may show different
-objects because the camera faces a different direction. Preserve the capture language on that view.
-An image used for identity does not automatically fix the next video's literal first frame.
+Judge a picture first by the intended person, appeal and feeling, then by the framing, color and
+interactions the video needs. Use that understanding when directing another image. Preserve useful
+results with their exact prompts, references and known model settings. Distinguish a proposed rewrite
+from the wording that actually produced an image. A successful rendering of individual leaves,
+tiles or reflections does not make their exact arrangement necessary in the next prompt.
-Choose references for the facts the work needs: a user's face, an exact product, a logo, a particular
-room or a continuing character. Keep the dependency tree shallow where one parent already supplies
-the needed facts. Familiar things can be described directly when recognizable meaning is enough.
-Connect reference images as real model inputs; prose alone does not create those edges.
-[Reference relationships](generated-dependencies.md) and [conversation examples](examples/conversation-images.md)
-cover multi-view work. [Transformations](../../creation/transformations.md) covers what a face or product
-swap changes across the video.
-
-## Judge the picture by its intended appeal
-
-Look first at whether this is the person and feeling the work needs. Then check framing, proportions,
-color relationships, and the visible interactions the video will depend on. Read the background as
-a place: does its construction and use come through, and does the person feel situated in it? When
-a view feels like a flat backdrop, reconsider the view and its spatial anchors before adding surface
-adjectives or more props. A well-described room
-does not compensate for weak casting. When revising, change the choice responsible for the weak
-result; retain the choices that already work. Keep produced images that remain useful to the work
-with their exact prompts and model settings so the next direction begins from actual production
-evidence.
+Carry accepted material into the production. When a new request calls for a different image, revise
+the choice responsible for that difference while retaining the direction that still serves the work.
diff --git a/skills/hypit/references/playbooks/index.md b/skills/hypit/references/playbooks/index.md
index 553fff0a..762106a8 100644
--- a/skills/hypit/references/playbooks/index.md
+++ b/skills/hypit/references/playbooks/index.md
@@ -22,7 +22,7 @@ not need a matching playbook before it can be understood or made.
| Read | When the current question concerns |
| --- | --- |
-| [Directing generated images](craft/image-direction.md) | phone-video realism, character presence, image prompts, materials and palette, or preparing a useful visual reference |
+| [Directing generated images](craft/image-direction.md) | phone-video capture, compelling people, camera relationships and settings; four-paragraph image direction and reference use |
| [Directing a voice](craft/voice-direction.md) | casting an appealing voice, choosing its vocal character, writing Voice Design direction or selecting a sample line |
| [Voice and performance](craft/voice-and-performance.md) | how Segment performances create semantic time, who is A-roll, recurring voice identity, visual A-roll presentations, or genuinely independent narration |
| [Directing generated video](craft/video-direction.md) | choosing a generation relationship, directing visible performance or silent action, camera behavior, cuts, or request duration |
@@ -44,7 +44,7 @@ unless the decision has changed.
| Read | When a concrete directing example would help |
| --- | --- |
-| [Image direction examples](craft/examples/image-direction.md) | complete phone-video prompts, their visible results, and the aesthetic relationships behind them |
+| [Image direction examples](craft/examples/image-direction.md) | four-paragraph production prompts for the balcony portrait and five reviewed character portraits |
| [Conversation image examples](craft/examples/conversation-images.md) | complementary podcast and interview views, product references, lifestyle B-roll, and changing camera relationships |
## Formats
diff --git a/skills/hypit/references/production/prompt-kits.md b/skills/hypit/references/production/prompt-kits.md
index a67b9831..0e98d0db 100644
--- a/skills/hypit/references/production/prompt-kits.md
+++ b/skills/hypit/references/production/prompt-kits.md
@@ -17,7 +17,7 @@ Choose a Kit whose fixed assumptions serve that intention:
| Direction needed | Where to look |
| --- | --- |
-| The fixed phone-video capture wording, with this picture's framing, subject and environment | The installed `@hypit/gpt-image-kits` README and its `phone-ugc-v1` template |
+| The fixed phone-video capture wording, followed by Person, Shot and Setting | The installed `@hypit/gpt-image-kits` README and its `phone-ugc-v1` template |
| A speaking performer, podcast, street encounter, video call, silent B-roll, or motion/camera transfer | The installed `@hypit/seedance-kits` README and its corresponding exported template |
These are packaged Sources, so their README and exported `.svs` files supply the template names,
@@ -26,9 +26,9 @@ Text slots, defaults and Recipe choices. A package's Surface vocabulary serves a
describes generation. Read the chosen template when its fixed text or a choice's exact wording matters.
For an ordinary speaking passage, connect the Segment's `.dialogue` and this passage's action;
-the selected Recipe supplies recurring performance choices. For an image, bring the framing and
-subject/environment direction from Craft. Keep the template's reference roles consistent with the
-actual media connected to the model. The owning package documents that order.
+the selected Recipe supplies recurring performance choices. For an image, bring Person, Shot and
+Setting direction from Craft. Keep reference responsibilities consistent with the actual media
+connected to the model. The owning package documents that order.
Fit includes assumptions as well as available knobs. A two-view conversation and a combined
split-screen opening may need different prompt structures. Keep the useful Kit for passages it