Skip to content

The VS Code of video editing, built for agents.

TypeScript

Update README and captions documentation to enhance clarity and provide usage examples

konstantin-paulus5fa6ef426537
diff --git a/README.md b/README.md
index 43f49061..639cefe8 100644
--- a/README.md
+++ b/README.md
@@ -1,8 +1,6 @@
# Diffusion Studio
-**FFmpeg for agents:** a multimedia editor you drive from the command line, with generative AI and multimodal understanding built in.
-
-What sets Diffusion Studio apart is `dapi`: a CLI that makes every part of the editor (projects, scene graphs, assets, AI generation, and rendering) scriptable over a local socket.
+Diffusion Studio lets your agent make videos. The agent writes a composition in TSX. The `dapi` CLI mounts it into the editor. Every element stays editable. Think of it as **FFmpeg for agents**, with generative AI and multimodal understanding built in.
```sh
dapi open # use -b to run the editor headless
@@ -10,6 +8,50 @@ dapi mount hero.tsx # render a composition into it
dapi node render -o hero.mp4 # encode the scene to disk
```
+## Getting started
+
+Use with Claude Code, Codex, Cursor, Copilot, or Gemini CLI. Install the skill once, globally:
+
+```sh
+npx skills add diffusionstudio/skills -g
+```
+
+`/editor` is the main skill you'll use. Ask for what you want in plain language.
+
+## Prompt examples
+
+**Video understanding and reasoning**
+
+```text
+/editor In three bullets, explain what starts the conflict. Include timestamps. https://youtu.be/aqz-KE-bpKQ
+
+/editor Name three recurring locations and give one visual cue that distinguishes each. https://youtu.be/dQw4w9WgXcQ
+```
+
+**Motion graphics**
+
+```text
+/editor Recreate the 3blue1brown animation from https://youtu.be/HEfHFsfGXjs, closely matching its visual style, pacing, framing, colors, labels, and transitions. Use the exact collision mathematics from Gregory Galperin's original paper, do not approximate the physics.
+
+/editor Create a ~20-second promo for vercel-labs/native in Vercel's presentation style. Research its official website, GitHub, and brand guidelines; use authentic assets and verified product features, with crisp typography, polished motion, and a strong final CTA.
+```
+
+**Video editing**
+
+```text
+/editor edit the footage in /path/to/folder
+
+/editor turn this footage into a polished YouTube video. Add readable captions and an attention-grabbing graphic in the opening to give viewers a strong visual hook.
+```
+
+**Clipping**
+
+```text
+/editor Can you pull the best 30-second moment from https://youtu.be/MtQ0qxyf-Ds and make a vertical version for social?
+
+/editor Make a 15-second version of this launch video. https://x.com/claudeai/status/2045156267690213649
+```
+
## Compositions as code
Compositions are [SolidJS](https://www.solidjs.com) modules. `dapi mount` compiles them and renders each element into an editable node in the document; re-mounting rebuilds in place, like reloading a webpage. Solid's control flow (`<For>`, `<Show>`, …) and any npm package are available to compose the tree, and generative assets are declared as values and produced on mount:
diff --git a/reference/jsx/captions.md b/reference/jsx/captions.md
index ae9f3f53..78ddbf0d 100644
--- a/reference/jsx/captions.md
+++ b/reference/jsx/captions.md
@@ -8,7 +8,7 @@
Transcription is **asynchronous and non-blocking**: the caption node is inserted at commit and its transcript attaches once ready. Because it reads the scene's audio, it runs **after** any generated assets in the scene have landed and after [audio sync](./audio-sync.md) has resolved; captioning a generated `voice`/`audio` track transcribes the finished audio at its final placement. The scene must contain an unmuted audio or video source; otherwise the caption node is left empty.
-Transcripts are **cached**: every transcript asset records a fingerprint of the scene's audible mix (source content, placement, source offset, playback rate, gain). When a scene's fingerprint matches an existing transcript, that asset is reused instead of transcribing again; re-mounting a project with unchanged audio consumes no credits. Any change to the scene's audio invalidates the fingerprint, and so does a new `seed`.
+Transcripts are **cached**: every transcript asset records a fingerprint of the scene's audible mix (source content, placement, source offset, playback rate, gain). When a scene's fingerprint matches an existing transcript, that asset is reused instead of transcribing again; re-mounting a project with unchanged audio consumes no credits. Any change to the scene's audio invalidates the fingerprint.
## Props