Skip to content

armgit / kokon

Public Imported from GitHub
0

Edit photos by describing the change, on your own GPU. A free, open-source alternative to Nano Banana and ChatGPT image editing: photos stay on your computer, no daily limit. One installer for Windows, Mac and Linux sets up the best open model (Qwen-Image 2.1). MCP server for AI agents.

video: Prompt+ line as two short sentences, read like the rest of the take

Gabriele Lozupone83febbdafd95
diff --git a/marketing/audio/voiceover.md b/marketing/audio/voiceover.md
index 62ff9e0..e77ae61 100644
--- a/marketing/audio/voiceover.md
+++ b/marketing/audio/voiceover.md
@@ -8,7 +8,7 @@ scene windows in the current cut; the redesign times the scenes to these lines.
|---|---|---|---|
| 1 | Intro (the mark) | 0:00–0:03.4 | Introducing Kokon. |
| 2 | Hook (wall of edits) | 0:03.4–0:07.8 | Another image editor? Nope. This one is yours. |
-| 3 | Create | 0:07.8–0:23.6 | Kokon runs on your PC, and everything stays there. Describe an image in a few words: Prompt Plus writes the full prompt, and your graphics card makes it. No cloud. No account. No subscription. |
+| 3 | Create | 0:07.8–0:23.6 | Kokon runs on your PC, and everything stays there. Describe an image in a few words. Prompt Plus writes the full prompt, and your graphics card makes it. No cloud. No account. No subscription. |
| 4 | Region | 0:23.6–0:39.5 | Want to change just one thing? Paint over it, and say what you want. Remove the scooter. Add a cat. Everything else stays exactly as it was. |
| 5 | Versions | 0:39.5–0:44.6 | Changed your mind? Every version is saved. |
| 6 | Montage | 0:44.0–1:00.9 | Change the season. Fix the sign. Restyle a portrait. Swap the furniture. Expand the frame. Shoot a product ad. One sentence each. |
@@ -21,7 +21,9 @@ line are in FINAL there); files land in voice/final/, one per line, mastered to
0.10 update (2026-10-20): Prompt+ in Create, Linux in Hardware. Only the changed sentences were
read again, with `gen_voice.py redo` / `splice` (the rest is the approved take, cut from
-voice/final/raw): line 3 sentence 1 and line 8 sentence 0. Line 6's "Shoot a product ad." is
+voice/final/raw): line 3 sentences 1-2 and line 8 sentence 0. Like the approved take, the new
+lines are short sentences read one at a time with a pause between them; one long sentence
+sounded rushed next to the rest. Line 6's "Shoot a product ad." is
no longer played (the montage has five edits, so the Agent scene still lands on the drop); the
line keeps its text so the file and its sentence numbers stay as approved. "Prompt Plus" is
spelled out for the voice; on screen it is "Prompt+".
diff --git a/marketing/video/src/app/timeline.ts b/marketing/video/src/app/timeline.ts
index f9d8941..036e8b4 100644
--- a/marketing/video/src/app/timeline.ts
+++ b/marketing/video/src/app/timeline.ts
@@ -79,7 +79,7 @@ const all = (scene: SceneName, line: number, local: number): Cue => cue(scene, l
// Create with Prompt+: a few words typed, Prompt+ on, Send on beat 10 and a cut to the card; it
// writes the prompt and fades to the percentage (a closer shot on beat 17); the logo forms, and
// from it the image hatches full frame.
-export const CREATE = { ask: 100, pill: 150, send: 172, job: 173, write: 180, written: 244, leave: 256, form: 278, closer: 279, done: 294, hatch: 300, full: 355, promises: 360 };
+export const CREATE = { ask: 100, pill: 150, send: 172, writes: 178, job: 173, write: 180, written: 244, leave: 256, form: 278, closer: 279, done: 294, hatch: 300, full: 355, promises: 360 };
// Shot changes (removeCut, removeShot, addCut, addShot) and the scene's end sit on beats.
export const REGION = { paint: 46, paint1: 130, removeCut: 138, remove: 150, removeSend: 177, removeShot: 185, addCut: 246, add: 256, addSend: 298, addShot: 307, same: 344 };
export const MONTAGE_EDIT = 91.5; // 1.5 bars
@@ -93,7 +93,8 @@ export const CUES: Cue[] = [
all("hook", 2, 15),
cue("create", 3, 12, 0),
cue("create", 3, CREATE.ask, 1),
- cue("create", 3, CREATE.promises, 2, 4),
+ cue("create", 3, CREATE.writes, 2),
+ cue("create", 3, CREATE.promises, 3, 5),
cue("region", 4, 6, 0),
cue("region", 4, 58, 1),
cue("region", 4, REGION.remove, 2),
diff --git a/marketing/video/src/app/voice.json b/marketing/video/src/app/voice.json
index 144235c..be622d3 100644
--- a/marketing/video/src/app/voice.json
+++ b/marketing/video/src/app/voice.json
@@ -26,7 +26,7 @@
]
},
"3": {
- "dur": 12.24,
+ "dur": 12.64,
"sentences": [
[
0.0,
@@ -34,19 +34,23 @@
],
[
2.92,
- 8.12
+ 5.02
],
[
- 8.52,
- 9.56
+ 5.42,
+ 8.52
],
[
- 9.96,
- 10.76
+ 8.92,
+ 9.96
],
[
- 11.16,
- 12.24
+ 10.36,
+ 11.16
+ ],
+ [
+ 11.56,
+ 12.64
]
]
},
diff --git a/marketing/video/src/scenes/AppScenes.tsx b/marketing/video/src/scenes/AppScenes.tsx
index ed1980e..7fb46d4 100644
--- a/marketing/video/src/scenes/AppScenes.tsx
+++ b/marketing/video/src/scenes/AppScenes.tsx
@@ -158,7 +158,7 @@ export function Create() {
// Once the card has formed the logo: a hard cut to the same logo filling the frame, and the
// image hatches out of it full frame (the app's hatching, on the frame instead of the stage).
const full = f >= C.hatch ? 1 : 0;
- const words = cues.find((c) => c.from === 2)!;
+ const words = cues.find((c) => c.from === 3)!;
return (
<AbsoluteFill>
<Backdrop />
@@ -178,9 +178,9 @@ export function Create() {
<Caption title="Made on your own GPU." sub="Qwen-Image 2.1 Turbo · Prompt+ ~2 s, then 2752×1536 in 38 s on an RTX 5070 Ti (sped up here)" start={C.leave + 14} end={C.hatch} size={48} />
<Words
items={[
- { text: "No cloud.", at: sentenceAt(words, 2) },
- { text: "No account.", at: sentenceAt(words, 3) },
- { text: "No subscription.", at: sentenceAt(words, 4) },
+ { text: "No cloud.", at: sentenceAt(words, 3) },
+ { text: "No account.", at: sentenceAt(words, 4) },
+ { text: "No subscription.", at: sentenceAt(words, 5) },
]}
end={LENGTH.create - 4}
/>