Sora
OpenAIOpenAI's video model: text or image in, short clip out, with storyboard control.
Track what changed between generation attempts, so retries teach something.
Video or image generation where several takes are normal
One-shot generation — there is nothing to compare
Keeps a log of attempts per shot: the prompt, the one variable changed, and the outcome. The rule is one variable per retry, which is the difference between a working prompt and a lucky one. The output is a short list of what the model responded to on this shot.
Clone the repository, then copy the skill folder into your agent's skills directory — ~/.claude/skills/ for Claude Code. Skills are plain files: no build step, no daemon.
npx skills add shot-retry-logRead the SKILL.md before you install. A good skill states when it applies, what it must not do, and how to verify the result. A skill that only lists adjectives will change the agent's tone without improving its output — skip those.
OpenAI's video model: text or image in, short clip out, with storyboard control.
Google DeepMind's video model, available through Gemini and Vertex AI.
Character video from a still image and a script.
Reads Loom recordings, titles, and transcripts for an account. The use case is turning a recorded walkthrough into written documentation, which is faster than re-wat…
A server from Mux's SDK monorepo that exposes video assets, live streams, and playback IDs. The read path is what earns its place: an agent can answer why a video is…
A community server that exposes Replicate's model catalogue, so one tool set covers image, video, music, and upscaling models. The breadth is the point: you are not …