Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

docs(x-posts): draft the 3D AR Studio launch copy and its video script

nirholas13bb3bc06778
diff --git a/docs/x-posts/3d-ar-studio-launch.md b/docs/x-posts/3d-ar-studio-launch.md
new file mode 100644
index 000000000..80890ab71
--- /dev/null
+++ b/docs/x-posts/3d-ar-studio-launch.md
@@ -0,0 +1,141 @@
+# 3D AR Studio launch
+
+The multi-model AR surface behind `three.ws/ar/studio`, extracted, generalized and published
+as an open-source package anyone can drop into their own site.
+
+**Live:**
+
+- Demo: <https://nirholas.github.io/3D-AR-Studio/>
+- Repo: <https://github.com/nirholas/3D-AR-Studio>
+- npm: `3d-ar-studio`, `3d-ar-studio-mcp`
+- MCP registry: `io.github.nirholas/3d-ar-studio`
+
+**Media plan.** Lead with the room screenshots, not the UI. The whole argument is that an
+object is standing on a real floor, so the first frame has to show a real floor.
+
+1. The car model standing in the living room, shot through the phone
+2. The armoured character standing in the same hallway
+3. The studio itself with the prompt box visible, so the reader sees where it came from
+4. Optional fourth: the desktop QR handoff
+
+---
+
+## Angle A: the one-line pitch (recommended)
+
+Post this as a single tweet with images 1, 2 and 3.
+
+```
+Your iPhone has shipped with AR hardware for years. Almost no website uses it.
+
+So I made it one line:
+
+<script src="unpkg.com/3d-ar-studio"></script>
+<ar-studio></ar-studio>
+
+Type a prompt, get a 3D model, stand it on your actual floor at actual size. Add ten more. Send the whole room as a link.
+
+Free. Open source. Works right now.
+
+nirholas.github.io/3D-AR-Studio
+```
+
+### Shorter alternate
+
+```
+Every iPhone has had ARKit for years and almost no website uses it.
+
+3D AR Studio makes it one script tag: describe an object, watch it appear, put it on your real floor. Add more. Share the whole scene as a link.
+
+Open source, free, no key.
+
+nirholas.github.io/3D-AR-Studio
+```
+
+---
+
+## Angle B: the fix (use if the launch post underperforms, or as a follow-up a day later)
+
+This is the sharper hook and it is true: the studio used to fall back to compositing over the
+camera feed on iOS, which photographs like AR and behaves nothing like it.
+
+```
+Spent the day finding out my own AR was fake on iPhone.
+
+No WebXR on iOS, so the model was just composited over the camera feed. It floats. It has no real size. Nothing occludes it.
+
+Every iPhone already ships ARKit through Quick Look. Now it uses it, and the model is converted on the device in about a second.
+
+nirholas.github.io/3D-AR-Studio
+```
+
+---
+
+## Thread
+
+Post Angle A as 1/, then:
+
+**2/**
+
+```
+Most web AR drop-ins place exactly one model and hand you off to a native viewer. That ends the session.
+
+This keeps the whole scene in your page. Place as many as you want, drag them, pinch to resize, twist to rotate. The arrangement is yours.
+```
+
+**3/**
+
+```
+Every device gets its real path, not an approximation.
+
+Android and headsets: WebXR with one anchor per model, real light estimation, and depth occlusion so things hide behind your couch.
+
+iPhone: Apple's AR Quick Look. The model is converted to USDZ on the device in about a second.
+```
+
+**4/**
+
+```
+Scenes are links. Compose on a laptop, scan the QR, it reopens exactly on your phone.
+
+Open a room and someone else can build in it with you, live.
+
+There's an MCP server too, so an agent can generate a model, arrange a scene, and hand a person one link that opens it in their room.
+
+npm i 3d-ar-studio
+github.com/nirholas/3D-AR-Studio
+```
+
+---
+
+## Reply and quote-tweet lines
+
+Use these under the main post rather than stuffing them into it.
+
+- The models are free too. A few hundred public-domain props, and text-to-3D with no key and
+ no account.
+- Point it at your own catalogue with one option: `assets: 'https://your.cdn/models.json'`.
+ Five common manifest shapes are read without reshaping anything.
+- `npx 3d-ar-studio create my-ar-site` then `npx 3d-ar-studio deploy` puts a live AR page on
+ GitHub Pages. Two commands.
+- Characters do not T-pose. Any humanoid with no baked animation gets an idle clip retargeted
+ onto its own skeleton, with no rig allow-list.
+- Apache-2.0. Take it and put it in your product.
+
+## Replies to expect
+
+- **"Does this need an app?"** No. It is a web page. The AR opens in the viewer your phone
+ already has.
+- **"What does the generation cost?"** Nothing, and it needs no account or key.
+- **"Can I use my own models?"** Yes, one option. Or write a source object with a `list()`
+ and it can come from anywhere.
+- **"Why not just model-viewer?"** model-viewer places one model and hands off. This composes
+ a whole scene, keeps it in your page, generates new models into it, and shares the
+ arrangement as a link.
+- **"iOS support?"** Real ARKit through Quick Look, with the USDZ built on the device.
+
+## Do not claim
+
+- Do not say WebXR works on iPhone. It does not. iOS gets Quick Look and that is the point.
+- Do not promise multi-model placement inside Quick Look or Scene Viewer. Those native
+ viewers take one model at a time; the multi-model scene lives in the page.
+- Do not put a number on generation time. It depends on queue depth.
diff --git a/docs/x-posts/3d-ar-studio-video-script.md b/docs/x-posts/3d-ar-studio-video-script.md
new file mode 100644
index 000000000..6368fd0f8
--- /dev/null
+++ b/docs/x-posts/3d-ar-studio-video-script.md
@@ -0,0 +1,118 @@
+# 3D AR Studio launch video
+
+Narration script for the launch video. Written for ElevenLabs text to video, with a three.ws
+3D avatar render used as the visual reference for Veo.
+
+**Voice direction.** Even, unhurried, slightly dry. No rising enthusiasm, no smile in the
+voice. The footage is doing the selling; the read is a person telling you something they find
+mildly outrageous. Leave the pauses in.
+
+**Pacing.** Roughly 155 words a minute. Each script below is timed at that rate, so the 30
+second cut is about 80 words.
+
+---
+
+## 30 seconds (primary)
+
+> Your phone has had augmented reality hardware in it for years.
+>
+> Almost no website has ever used it.
+>
+> This is 3D AR Studio.
+>
+> Describe something. Watch it appear.
+>
+> Then put it on your actual floor, at actual size.
+>
+> Add another one. Move them around. Walk through them.
+>
+> When it looks right, send the whole room to someone as a link.
+>
+> One script tag. No key, no account, no app.
+>
+> It's free, it's open source, and it already works on the phone in your pocket.
+
+---
+
+## 15 seconds (cutdown for the reply slot)
+
+> Your phone has had AR hardware in it for years, and almost no website uses it.
+>
+> 3D AR Studio is one script tag.
+>
+> Describe something, watch it appear, stand it on your real floor.
+>
+> Free. Open source. Works right now.
+
+---
+
+## 60 seconds (extended, for YouTube and the site)
+
+> Your phone has had augmented reality hardware in it for years.
+>
+> Almost no website has ever used it.
+>
+> This is 3D AR Studio. It's one script tag, and it turns any web page into this.
+>
+> Describe something you want. It gets generated, and it lands in the room in front of you.
+>
+> Not floating in a camera filter. On the floor. At the size it would really be.
+>
+> Add another. And another. Drag them around until the room looks right.
+>
+> Then send it. The whole arrangement travels as a link, so what you built on a laptop opens
+> exactly the same on someone else's phone.
+>
+> Or open a room, share the code, and build it together while you're both standing in it.
+>
+> There are hundreds of free models in it already. Generating a new one costs nothing and
+> needs no account.
+>
+> It's open source. Take it and put it in your own product.
+
+---
+
+## Shot prompts for Veo
+
+Use the three.ws avatar render as the character reference on shots 2 and 5. Keep every shot
+handheld with slight natural motion: the entire point is that the objects stay put while the
+camera moves. No drone moves, no gimbal glide, no lens flare.
+
+1. Close handheld shot of a person holding a phone in a bright living room, screen facing
+ camera, empty hardwood floor visible behind them, soft daylight from a window
+2. Same room, camera pushes past the phone: a stylized 3D character now stands on the floor at
+ human height, with a correct contact shadow and window light falling on it from the same
+ direction as the room
+3. Handheld orbit at chest height: two or three objects standing on the real floor at
+ different distances, strong parallax as the camera moves around them
+4. Close on the phone screen: a text field, a few words typed, then an object materializing
+ into the room visible behind the screen
+5. Wide shot, two people in the same room both holding phones, both looking at the same
+ character standing on the floor between them
+6. End card, matte black, product name and URL only, no motion
+
+### Negative direction
+
+Give these to Veo as things to avoid, because they are exactly what makes AR footage read as
+fake:
+
+- objects floating above the floor or intersecting furniture
+- an object that slides or swims when the camera moves
+- shadows that do not match the room's light direction
+- a phone screen showing a different scene from the one behind it
+- lens flares, glossy product-ad lighting, sped-up camera moves
+
+---
+
+## Captions
+
+Burn in short captions rather than relying on autoplay sound. Keep them to one line each and
+match them to the narration beats:
+
+1. Your phone has had AR hardware for years
+2. Almost no website uses it
+3. Describe it. Watch it appear.
+4. On your actual floor, at actual size
+5. Send the whole room as a link
+6. One script tag. Free. Open source.
+7. nirholas.github.io/3D-AR-Studio
diff --git a/docs/event-x-posts.md b/docs/x-posts/event-x-posts.md
similarity index 100%
rename from docs/event-x-posts.md
rename to docs/x-posts/event-x-posts.md
diff --git a/docs/fomo-verified-x-post.md b/docs/x-posts/fomo-verified-x-post.md
similarity index 100%
rename from docs/fomo-verified-x-post.md
rename to docs/x-posts/fomo-verified-x-post.md
diff --git a/docs/ibm-skillsbuild-x-post.md b/docs/x-posts/ibm-skillsbuild-x-post.md
similarity index 100%
rename from docs/ibm-skillsbuild-x-post.md
rename to docs/x-posts/ibm-skillsbuild-x-post.md
diff --git a/docs/metaplex-agent-deployer-x-post.md b/docs/x-posts/metaplex-agent-deployer-x-post.md
similarity index 100%
rename from docs/metaplex-agent-deployer-x-post.md
rename to docs/x-posts/metaplex-agent-deployer-x-post.md
diff --git a/docs/metaplex-agent-mcp-x-post.md b/docs/x-posts/metaplex-agent-mcp-x-post.md
similarity index 100%
rename from docs/metaplex-agent-mcp-x-post.md
rename to docs/x-posts/metaplex-agent-mcp-x-post.md
diff --git a/docs/news-archive-x-post.md b/docs/x-posts/news-archive-x-post.md
similarity index 100%
rename from docs/news-archive-x-post.md
rename to docs/x-posts/news-archive-x-post.md
diff --git a/docs/onchain-agent-wallets-x-post.md b/docs/x-posts/onchain-agent-wallets-x-post.md
similarity index 100%
rename from docs/onchain-agent-wallets-x-post.md
rename to docs/x-posts/onchain-agent-wallets-x-post.md
diff --git a/docs/oracle-trading-mcp-x-post.md b/docs/x-posts/oracle-trading-mcp-x-post.md
similarity index 100%
rename from docs/oracle-trading-mcp-x-post.md
rename to docs/x-posts/oracle-trading-mcp-x-post.md
diff --git a/docs/pumpfun-article-x-post.md b/docs/x-posts/pumpfun-article-x-post.md
similarity index 100%
rename from docs/pumpfun-article-x-post.md
rename to docs/x-posts/pumpfun-article-x-post.md
diff --git a/docs/x-posts/threews-avatar-video-script.md b/docs/x-posts/threews-avatar-video-script.md
new file mode 100644
index 000000000..d312a77b1
--- /dev/null
+++ b/docs/x-posts/threews-avatar-video-script.md
@@ -0,0 +1,188 @@
+# Text-to-video script: the three.ws avatar
+
+Prompts for a video model (Veo, run through ElevenLabs), using a three.ws 3D avatar render as the
+reference image. This file is the **picture**, not the words. The spoken lines live in
+[metaplex-agent-deployer-video-script.md](./metaplex-agent-deployer-video-script.md), and the
+deployer-specific shot list lives in [metaplex-agent-mcp-x-post.md](./metaplex-agent-mcp-x-post.md).
+
+The idea being sold here is the tagline and nothing else: **give your AI a body.** A model with no
+form gets one, the form moves, the form owns a wallet, the form walks out into the real world. Four
+claims, all true today.
+
+---
+
+## Before you generate
+
+**Reference image.** Open any avatar on [three.ws](https://three.ws), frame it front-facing in a
+neutral pose against the dark stage, and screenshot at the largest size you can. Use the **same
+file on every shot** or the character drifts and the cut stops reading as one being.
+
+**Clip length.** Veo generates in 8-second beats. Every prompt below is written to resolve inside
+one beat, with one idea and one camera move each. A prompt that asks for two things at 8 seconds
+returns mush.
+
+**Text.** Video models mangle letters. Nothing legible is described in any prompt, and the negative
+prompt bans it outright. Every real string gets composited afterwards in an editor.
+
+**Palette, held across all shots.** Near-black void, violet as the light of the machine, teal as the
+light of the network, one warm gold reserved for money in motion. No fourth color.
+
+---
+
+## Hero clip, 8 seconds
+
+Use this if you only render one thing.
+
+> Cinematic product shot on a near-black void stage. A soft cloud of violet light particles hovers
+> at center frame, formless and breathing. The particles rush inward and collapse onto a figure: a
+> stylized 3D humanoid avatar matching the reference character exactly, materializing from the feet
+> upward in one continuous sweep as surfaces and textures resolve and catch the light. As the last
+> surface lands the figure lifts its head and looks directly into the lens, alive and calm. A single
+> point of warm gold light ignites in its open palm and settles, pulsing slowly. Camera: slow push
+> in from wide to medium, ending at chest height, locked and steady. Lighting: soft key from the
+> upper left, violet rim from behind, gold bounce from the palm onto the jaw. Shallow depth of
+> field, gentle volumetric haze, fine film grain. Audio: low sub-bass swell, a soft rush of
+> particles, one crystalline chime as the palm ignites, then quiet room tone. No text, no logos, no
+> other characters.
+
+---
+
+## Five-shot sequence, 8 seconds each, about 40 seconds cut
+
+### Shot 1 of 5: formless
+
+> Cinematic 3D render on a near-black void stage. No character. A shifting cloud of violet light
+> particles hangs in the center of frame, pulsing gently in and out like breathing, with faint
+> threads of light passing through it as if something is thinking inside. The cloud has no shape and
+> keeps failing to hold one. Camera: very slow push in, wide to wide-medium. Lighting: the cloud is
+> the only light source, spilling violet into the haze around it. Volumetric fog, fine film grain.
+> Audio: low sub-bass hum, faint electrical whisper, no music. No text, no characters.
+
+### Shot 2 of 5: the body
+
+> Same near-black void, same violet cloud. The particles pull sharply inward and collapse onto a
+> figure: a stylized 3D humanoid avatar matching the reference character exactly. Materials resolve
+> across the body from the feet upward in one smooth continuous sweep, each surface catching the key
+> light as it lands. A faint wireframe grid maps the silhouette for an instant and fades. The figure
+> stands still, newly solid. Camera: slow dolly in, wide to medium. Lighting: soft key upper left,
+> violet rim from behind, haze in the beam. Audio: rising sub-bass, a soft mechanical lock as the
+> final surface lands, then silence. No text.
+
+### Shot 3 of 5: alive
+
+> Same character, fully rendered, standing center frame in the near-black void. It lifts its head,
+> turns slightly, and looks directly into the lens. Its chest rises with a breath. It shifts its
+> weight, flexes one hand open and closed, and the beginning of an expression crosses its face.
+> Small, human, unhurried movements. Camera: slow arc from the right, settling to a medium close-up
+> on the face and shoulders. Lighting: soft key upper left, violet rim, a warm fill lifting the
+> eyes. Shallow depth of field with the eyes in sharp focus. Audio: one quiet breath, low room tone,
+> a single soft note. No text.
+
+### Shot 4 of 5: it holds its own money
+
+> Same character in the same void. It turns its right hand palm-up. A small hexagonal token of warm
+> gold light ignites an inch above the palm, rotates slowly, and settles there pulsing like a
+> heartbeat. Gold light spills up onto the face and chest. Thin teal threads reach in from the
+> darkness toward the token and connect, and the token pulses brighter with each connection. Camera:
+> slow push to a medium close on the hand and face together. Lighting: gold from the token as the
+> key, violet rim from behind, teal accents on the threads. Shallow depth of field, the token in
+> sharp focus. Audio: crystalline chime, quiet electrical hum, a soft ascending tick with each
+> connection. No text, no coins, no currency symbols.
+
+### Shot 5 of 5: out into the world
+
+> Same character in the void. A tall rectangle of soft teal light opens in the darkness in front of
+> it, like a doorway cut into the black, and through it we glimpse a sunlit ordinary room with plain
+> furniture. The avatar walks forward and steps through the rectangle. On the far side it stands in
+> the real room, lit by real daylight, still glowing faintly violet at the rim, gold still in its
+> palm, looking around the space. Camera: track behind the character through the doorway, then swing
+> around to a wide of it standing in the room. Lighting: violet and teal on the void side, natural
+> warm daylight on the room side, a clean transition at the threshold. Audio: the hum drops away as
+> it crosses, replaced by quiet natural room ambience and birdsong outside. No text.
+
+---
+
+## Negative prompt
+
+Paste into the negative field on every shot.
+
+> text, letters, numbers, words, subtitles, captions, watermark, logo, UI, user interface, HUD,
+> charts, coins, dollar signs, currency symbols, crowds, extra characters, duplicate character,
+> warped hands, extra fingers, face morphing, character drift, shaky camera, jump cuts, lens flare
+> spam, rainbow colors, oversaturation, low resolution, cartoon shading, distorted anatomy
+
+---
+
+## If you want it to speak
+
+Veo can generate its own audio. If you use that, put the sentence in double quotes inside the
+prompt, name who says it, and append **"no subtitles, no captions, no on-screen text"** or captions
+get burned into the frame.
+
+If ElevenLabs speaks it instead, generate the voice first and cut the picture to the read. Never the
+reverse: matching a render to an existing voice track works, matching a voice to an existing render
+never does.
+
+Lines that fit these five shots, one per beat, counted to sit inside 8 seconds:
+
+1. *(11 words)* "Your model can write, reason, and speak. It has no body."
+2. *(12 words)* "Type a sentence. It gets one. Rigged, animated, ready in a minute."
+3. *(9 words)* "Now it can look at you when it answers."
+4. *(12 words)* "It holds its own wallet on Solana. It can be paid directly."
+5. *(13 words)* "Then it walks off our site and onto yours. One tag, any page."
+
+Keep the pause after "It has no body." That silence is the whole premise.
+
+---
+
+## Post-production
+
+- Composite every real string in an editor, never in the model: the URL, any address, any label.
+- End card, 2 seconds, static: `three.ws` on black, and the tagline under it. Nothing else.
+- Cut so the muted autoplay frame is shot 2 or shot 3. A body assembling reads instantly with no
+ sound; a violet cloud does not.
+- Post as native video, not a link to a video. X will not show a link preview and a video at once,
+ and the video wins.
+
+---
+
+## X post
+
+### Recommended
+
+> Give your AI a body.
+>
+> Type a prompt, get a rigged 3D avatar that talks, walks, and holds its own Solana wallet. Drop it
+> on your site with one tag.
+>
+> three.ws
+
+Line 1 is the product in five words. Line 2 is the three things that make it different from a
+picture, in the order people care about. The wallet lands last because it is the part nobody expects
+from an avatar company.
+
+### Alternate: builder-facing
+
+> Your agent has a voice, a model, and a name. It still has no body.
+>
+> three.ws gives it one: a rigged 3D avatar you can talk to, embed anywhere with a single element,
+> and fund with its own on-chain wallet.
+>
+> ```html
+> <script type="module" src="https://three.ws/agent-3d/latest/agent-3d.js"></script>
+> <agent-3d src="agent://base/42"></agent-3d>
+> ```
+
+### Alternate: $THREE-first
+
+> Every 3D agent built on three.ws ships with its own Solana wallet, an on-chain identity, and
+> pay-per-call built in.
+>
+> $THREE is the coin of that economy.
+>
+> Give your AI a body: three.ws
+
+### Reply slot, post under the main one
+
+> The avatar in the video was generated from a text prompt on three.ws, rigged automatically, and
+> rendered in the browser. No modeling, no rigging software, no download.
diff --git a/docs/x-meetup-posts.md b/docs/x-posts/x-meetup-posts.md
similarity index 100%
rename from docs/x-meetup-posts.md
rename to docs/x-posts/x-meetup-posts.md
diff --git a/scripts/audit-docs.mjs b/scripts/audit-docs.mjs
index 661ae73e1..6f121963d 100644
--- a/scripts/audit-docs.mjs
+++ b/scripts/audit-docs.mjs
@@ -295,10 +295,13 @@ const UNPUBLISHED_DOCS = new Map([
['pumpfun-article-x-post', 'internal: ready-to-paste X post drafts; posting is owner-gated'],
['memetic-launcher', 'owner-gated: names a crypto project other than $THREE'],
['meta-allocator', 'owner-gated: names a crypto project other than $THREE'],
+ ['metaplex-agent-deployer-video-script', 'internal: unposted spoken-script drafts for a marketing video'],
+ ['metaplex-agent-deployer-x-post', 'internal: unposted X announcement drafts, not a reader-facing doc'],
['metaplex-agent-mcp-x-post', 'internal: unposted launch copy and video script, not a reader-facing doc'],
['money-map', 'internal: revenue-share and treasury routing'],
['native-launchpad', 'owner-gated: names a crypto project other than $THREE'],
['news-archive-x-post', 'internal: paste-ready X copy for the news archive, drafted for the owner to post'],
+ ['onchain-agent-wallets-x-post', 'internal: unposted X announcement drafts, not a reader-facing doc'],
['oracle-trading-mcp-plan', 'internal: build order for the Oracle and trading-agent MCP servers, not a reader-facing product doc'],
['oracle-trading-mcp-x-post', 'internal: paste-ready X copy for the MCP launch, drafted for the owner to post'],
['nvidia-apps-catalog-listing', 'internal: paste-ready listing copy and portal steps for the NVIDIA Inception catalog'],
@@ -315,6 +318,7 @@ const UNPUBLISHED_DOCS = new Map([
['robinhood-chain-markets', 'owner-gated: names a crypto project other than $THREE'],
['solana-pumpfun', 'owner-gated: names a crypto project other than $THREE'],
['syndication', 'internal: distribution mechanics for the announcements feed'],
+ ['threews-avatar-veo-script', 'internal: unposted text-to-video prompts and X drafts for a marketing video'],
['trading-experiment', 'owner-gated: names a crypto project other than $THREE'],
['trading-hub', 'in-flight: landed mid-session, register with its /trading page'],
['event-x-posts', 'internal: ready-to-paste X post drafts; posting is owner-gated'],