cinematic_annotate_video
Cinematic Annotate turns one clean image into adirector-style camera-path markup, converts it into a Seedance 2.0 motion prompt, and generates a clean cinematic video from the original frame.
Try an example
About this workflow
cinematic_annotate_video turns a single clean image into a short, smooth, cinematic shot by auto-writing a director-style camera path and translating it into a Seedance 2.0 motion prompt. The workflow starts with LoadImage and passes your frame into GeminiImage2Node, which analyzes the scene and generates camera-path markup using the "Nano Banana 2" annotation style. That markup is then refined/formatted by GeminiNode into a Seedance 2.0 motion specification.
The Seedance 2.0 prompt and your original image are fed into ByteDance2ReferenceNode to synthesize the motion while preserving the look of the source frame. PreviewImage and PreviewAny help you inspect the input frame and the auto-generated text prompts before you commit to a render. Finally, SaveVideo writes the result to disk. Groups are organized for clarity: Annotate Image (image analysis and camera notes), Prompt (LLM formatting to Seedance 2.0), and Seedance 2.0 (reference-driven video generation).
Questions & answers
Seedance 2.0 is a motion-prompt format describing camera moves (e.g., path, timing, easing). GeminiNode formats the camera-path annotation into this schema, which ByteDance2ReferenceNode uses to drive the reference-based video generation.
If your GeminiImage2Node/GeminiNode are configured to call a cloud Gemini API, you will need valid credentials and an internet connection. Check those nodes’ settings in ComfyUI and add your key or switch to a local provider if supported.
Start with a clean, high-resolution image with clear subject/background separation. Keep camera moves moderate (avoid extreme zooms/rolls), and adjust duration/easing in the Seedance 2.0 prompt for smoother acceleration. Use PreviewAny to validate the generated motion text before rendering.
This workflow is designed for a single reference frame and a short cinematic move. For multi-shot sequences or longer durations, replicate the pipeline per shot or extend the Seedance 2.0 prompt and ByteDance2ReferenceNode settings if the node exposes longer frame counts.
Related workflows
- LTX-2.3
Turn an image into a video
- LTX-2.3
Add camera motion first last frame
- Wan Animate
Copy movement from a video
- LTX-2.3
Make your character talk
- Wan2.1 InfiniteTalk
Make your character sing
- LTX-2.3
Animate a scene from a reference sheet
- Seedream, MiniMax
Turn a product photo into a video
- wan2.2 Animate
3.2 Advanced: Control Video Generation
- Seedance 2.0
4K Seedance 2.0 - Reference to Video
- SD1.5
AnimateDiff - IPAdapters Looping Animation
- Wan2.1 InfiniteTalk
Audio-Driven Character Lip Sync Video
FastVideo FastH3: Image to Video