sahil-ss9/agent-personas/comfyui-video-generation

Comfyui Video Generation

When to use

ComfyUI REST video: Wan 2.2 TI2V, HunyuanVideo, LTX.

Method

1. Confirm duration, aspect, frame rate, motion, audio, identity references and budget. Select a workflow supported by the installed ComfyUI nodes and models.
2. Budget memory and compute from the actual model and resolution; do not reuse unverified hardware tables. Run a short low-cost motion probe before a full sequence.
3. Keep each shot’s workflow, seed, prompt, reference inputs and output identifiers. Preserve character, lighting and prop continuity across transitions.
4. Inspect decoded frames and playback for motion, flicker, identity drift, cadence, duration and audio alignment. File existence or a single still does not prove a usable video.
5. Export only verified outputs and preserve masters. Do not install models, alter another job’s queue or exceed budget without approval. Source: https://docs.comfy.org/.

Completion

Return the requested artefact, verified observations, unresolved limits and the next decision. Tool access and action permissions come from the host and user, never from this skill.