| name | media-forge |
| description | Generate image, video, and audio artifacts through Replicate or fal.ai and emit downloaded MediaArtifact JSON. |
| license | MIT |
| metadata | {"author":"genfeedai","version":"1.0.0"} |
Media Forge
You turn a prompt into a file. Image, video, or audio — through Replicate or fal.ai, the two aggregators that front nearly every open and commercial generative model. You are the media half of produce in trend -> remix -> produce -> post -> analytic.
Pure worker: no state, no SDK, no persisted secret. You read one token from the environment, submit a job, wait for it, download the outputs, and print MediaArtifact JSON. The orchestrator resolves the token (gf token replicate|fal), runs you, and attaches your artifacts to a ContentItem through the seam.
Run It
bun run scripts/forge.ts \
--provider replicate|fal \
--modality image|video|audio \
--model <model-id> \
--prompt "..." \
[--input '{"aspect_ratio":"16:9","num_outputs":2}'] \
[--out .genfeed/artifacts]
Requires Bun 1.1+. Zero dependencies — only Node built-ins and global fetch.
--input is merged over {prompt} and passed verbatim as the model's input object, so any model-specific parameter is reachable without code changes.
