Adjustment Frame Workflow
A Director-style AI workflow that generates a shot, refines camera angle, and produces multiple video takes across top models so you can quickly choose the best result.
概要
The Adjustment Frame Workflow is a director-style pipeline for turning a written shot idea into multiple polished video takes you can compare side-by-side. It stages your work like a real set: first you block the shot as a still, then you refine the camera, and finally you roll multiple takes across different model backends. A Text Multiline node captures your prompt (what the image is and the action you want), and the 1A) GENERATE SHOT STILL group produces a reference frame you can review with PreviewImage/PreviewAny before spending any credits.
入力するもの
- ワークフローを読み込み中
ComfyUI で実行できます。 このワークフローはこのページでは実行できません。すべてが読み込まれた状態でComfy Cloudで開くか、ご自身のComfyUI用にダウンロードしてください。
Cloudで試す例を試す
よくある質問
The API partner nodes—GrokVideoNode, KlingVideoNode, and GeminiNode—consume credits. ByteDance2ReferenceNode typically pairs with an external generation call as well. Work only inside 1A/1B first (local and free nodes like PreviewImage, PreviewAny, Image Resize, ComfyMathExpression, SetNode/GetNode) until your still frame and camera angle are locked. Then enable a single video node and render.
Use QwenMultiangleCameraNode in 1B) CONTROL CAMERA ANGLE. Add clear cues such as shot size (e.g., "medium shot"), angle ("low angle"), lens ("35mm"), and motion intent ("slow dolly-in"). RegexExtract helps parse these from your text, and the node produces consistent camera-aware variations you can preview before video generation.
Yes. PrimitiveInt and ComfyMathExpression nodes set timing values like duration/fps. Use Image Resize to match model-specific resolution requirements. Note that supported lengths vary by model—some allow longer clips than others—so check each node’s limits and adjust accordingly.
Provide a clean, well-lit still as a reference and route it through ByteDance2ReferenceNode before your chosen video backend. Keep the composition similar to your approved still, and avoid large mid-scene camera changes. Small prompt changes can shift identity—lock your description once the still is approved.