Roadmap: webprompts — a forkable AI prompt→media→video canvas on your pod

[melvincarvalho]

A pod-native node canvas for AI prompts, media assets, and video assembly. You hand it a JSON-LD doc (inline data island or ?src=<pod-url>) and it renders a canvas of prompt / image / video nodes wired by lineage. Your data on your pod → fork = copy the doc. Built in small phases, each shipping something useful.

Inspired by the prompt-canvas pattern (promptsref, ComfyUI, Krea, Flora). Video side aligns with OpenTimeline Lite (https://github.com/inartes/otio). Always intended for prompt sharing (webprompts.org).

Phases

P1 — Canvas reader

• Load a webprompts JSON-LD from a <script type="application/ld+json"> data island OR ?src=<url> (pod file, via xlogin.authFetch / conneg)
• Render nodes: prompt, image (schema:ImageObject), video (schema:VideoObject); lineage edges (wp:derivedFrom)
• Pan + zoom infinite canvas (vanilla pointer/SVG, build-free)
• Read-only. Useful: view & share any prompt canvas.

P2 — Edit + save to pod

• Add / move / connect / edit nodes (prompt text, model + params, media URL)
• Save to /public/webprompts/<name>.jsonld; fork = copy to your pod
• Useful: author your own canvases.

P3 — Timeline (OTIO-lite)

• A timeline strip: arrange image/video clips into tracks; preview the sequence in order
• Import / export OTIO-lite (Timeline → Track → Clip with source_url, source_range, record_range)
• Useful: assemble a video sequence from the assets.

P4 — Generation

• Per-node Generate (prompt + model + params)
• Backends: BYO endpoint/key (stored in /private/) and/or NIP-90 nostr DVM (publish a job to a relay, provider fulfills, result back over nostr) — the decentralized, no-central-credits path
• Output saved to the pod, becomes a new node/clip with a wp:derivedFrom edge
• Useful: actually generate image/video on the canvas.

P5 — Prompt library / sharing

• Browse / tag / fork prompts (the webprompts.org purpose); shareable canvas/prompt URLs
• Useful: share prompts.

Data model

One JSON-LD doc, two views (canvas + timeline):

• schema.org — ImageObject / VideoObject for media assets.
• wp: vocab — wp:Prompt (text, wp:model, wp:params, wp:aspect), canvas wp:x / wp:y, wp:derivedFrom (lineage edges).
• OTIO-lite — tracks[] -> clips[] referencing asset @ids for the sequence.

A prompt canvas and a video timeline are the same document seen two ways.

Key decisions / constraints

• Build-free core (P1-P3): canvas + edit + timeline are pure client, no backend.
• Generation needs a backend (P4) — most model APIs block direct browser calls (CORS), hence BYO-endpoint or NIP-90 DVM (decentralized, reuses our nostr plumbing from yap/taproot).
• Fork = copy the pod resource — decentralized remix, no SaaS.

TODO / references

• Align P1/P3 data model with open-cinema/AI-VIDEO-WORKFLOW-TIPS.md (shot/prompt fields, continuity) — need the file (not yet located).
• OTIO-lite spec: https://github.com/inartes/otio

[melvincarvalho]

P1 shipped ✅ — the canvas reader is live: https://solid-apps.github.io/webprompts/

Loads a webprompts JSON-LD (inline data island, ?src=<url>, or the built-in sample.jsonld) and renders the node graph on a pan/zoom canvas — CharacterSheet → Storyboard → Clip → Stitch → Score → Export, wired by wp:derivedFrom, with the Bible panel, per-type colours, status dots, and clip chips (model · duration). Verified live: 7 nodes, 9 lineage edges, no JS errors.

Next: P2 — edit on the canvas + save to your pod (/public/webprompts/<name>.jsonld; fork = copy).