Seedance Lite Reference Images

Character consistency is the eternal video-model problem, and this node is Seedance Lite's answer: give it two to four reference images and it generates a clip that keeps their look. Unlike the first-frame and first+last nodes, these references aren't literally the opening frames - they're references, input images the model uses to pin down who or what your subject is, how it's styled, its palette. Then the prompt describes the action and the model animates while staying faithful to the references. It's the closest thing this pack offers to "character sheet → video," which is exactly why you'd reach for it over the plain image-to-video node when your subject needs to be recognizable.

Seedance is ByteDance's API-only video line (the KB's ByteDance panel walks through the whole "flagship products closed" posture), so this is a cloud task: your reference images are base64-encoded and sent as reference_image content parts, an async task is created, and the node polls until the video URL is ready. Same Lite cost-and-speed trade as the other Lite nodes - good enough quality for a lot of uses, not the top Pro tier.

Inputs and outputs that matter

• reference_image_1 and reference_image_2 - required. Give it consistent, well-lit shots of the same subject for best results.
• reference_image_3 / reference_image_4 - optional; more references help, up to four.
• prompt - the action and scene. This node's prompt is about what happens, not what things look like - the images carry the look.
• resolution - 480p or 720p only (no 1080p here).
• ratio - six fixed aspects; notably no adaptive on this node, so match it to your references.
• duration - 3–12 seconds. seed - reproducibility, -1 for random.

Output: one STRING with the hosted video URL. Preview/save it, or pull frames with Video URL to Frames.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/FloyoAI/ComfyUI-Seed-API
cd ComfyUI-Seed-API
pip install -r requirements.txt

Restart, set your BytePlus API key in config.ini or BYTEPLUS_API_KEY. Manager: search "ComfyUI Seed API Integration".

Where people get burned

References are guidance, not puppets. If your two reference images disagree - different outfits, different angles, different lighting - the model will pick a compromise and your "consistent" character will wobble. Feed it a tight set of images that already agree. Also, this is not an animation-from-still node in disguise; the references don't dictate the opening shot the way a first frame does, so if you need the clip to start on a specific image, use the first-frame or first+last node instead. Empty-string output means the error is in the console (key, quota, or model permission, as usual for this pack), and every render is a metered cloud task. Reference-based consistency is the killer use here - just set the references up to win.