Comfy Bernini Videoedit

comfy-bernini-videoedit

Use this skill for Bernini video edit workflows through the existing comfy-videogen wan22-bernini CLI command. This is a workflow skill, not a new model profile system: it always uses the local videogen.wan22-bernini capability and the wan22-bernini model profile.

If comfy-videogen, comfy-models, or comfy-media is missing, use comfy-tools-setup first. In this repository, prefer uv run comfy-videogen; outside the repo, let comfy-tools-setup install the CLIs with uv tool.

At the start of every Bernini workflow, start or reuse the local Comfy Media gallery for the active output directory:

uv run comfy-media gallery --out outputs --host 127.0.0.1 --port 8765

Keep outputs under outputs/. The CLI is quiet by default and prints final JSON; use --verbose only for debugging ComfyUI runtime output, warnings, or progress.

If model validation or generation fails with missing_model_file, use comfy-model-downloader for the exact local capability before running inference:

uv run comfy-models download videogen.wan22-bernini --dry-run
uv run comfy-models download videogen.wan22-bernini --yes

Do not download models automatically. Preview first, explain what will be downloaded, and wait for user approval before using --yes.

Profiles

• V2V: source video edit. Use a video and a prompt; preserve camera, lighting,

and scene continuity unless the user asks otherwise.

• RV2V: reference-guided video edit. Use a source video plus one or more

reference images for subject, style, or object replacement.

• R2V: reference-to-video generation. Use one or more reference images without

a source video.

• VV2V: video-reference-to-video is not supported by the current CLI. Do not

pretend it works and do not silently map it to another mode. Ask for extracted frames or still images from the reference video, then run RV2V.

Commands

V2V source video edit:

uv run comfy-videogen wan22-bernini \
  --input-video path/to/source.mp4 \
  --prompt "Edit the video while preserving camera motion, lighting, and background continuity." \
  --out outputs

RV2V reference-guided video edit:

uv run comfy-videogen wan22-bernini \
  --input-video path/to/source.mp4 \
  --reference-image path/to/reference.png \
  --prompt "Replace the character with the subject in image 0. Keep camera motion, lighting, and background unchanged." \
  --out outputs

Use repeated --reference-image flags for multi-reference conditioning:

uv run comfy-videogen wan22-bernini \
  --input-video path/to/source.mp4 \
  --reference-image path/to/subject.png \
  --reference-image path/to/outfit.png \
  --prompt "Use image 0 for identity and image 1 for outfit. Preserve the source video's camera motion and scene." \
  --out outputs

R2V reference-to-video:

uv run comfy-videogen wan22-bernini \
  --reference-image path/to/reference.png \
  --prompt "Create a cinematic short video of the subject in image 0 with natural motion and coherent lighting." \
  --out outputs

VV2V requested with a reference video:

Bernini VV2V is not supported by the current comfy-videogen CLI. Extract one or
more still frames from the reference video, then use RV2V with --reference-image.

After each successful generation, run:

uv run comfy-media index --out outputs

Use the final JSON to report the saved MP4 path, profile, dimensions, frame count, seed, and any media manifest path.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: quinteroac
• Source: quinteroac/comfy-agent-tools
• License: MIT

Install and usage instructions live in the source repository linked above.