ByteDance Seedance 2.0 Reference To Video (Channel Edition)

Reference-to-video is the mode that makes a video model feel like it actually understands your intent: instead of a single start frame, you hand it images, other videos, and even audio as reference material, plus a text prompt, and it generates a clip informed by all of them. This is the Channel Edition wrapper for ByteDance's Seedance 2.0 reference endpoint - the API-only video model from the folks behind Seedream. In ComfyUI terms, it's the node that lets you do "here's the character, here's the vibe, here's the B-roll I shot - make me a sequence" without running a local video model.

How it works

Cloud execution through BizyAir, like every node in this pack: your reference media uploads to api.bizyair.ai, the request hits seedance-2-0-base/reference-to-video, and the finished clip downloads back as a native ComfyUI VIDEO. No local model, no VRAM spike, no checkpoint downloads. The trade: metered billing on BizyAir credits and your reference media passing through SiliconFlow's servers.

Inputs that matter

• prompt - the text steering, up to a generous 1–20480 characters. Describe motion and intent, not just a still image.
• images / videos / audios - the autogrowing reference slots. This is the heart of the node: images for look/character, a reference video for motion or style, audio for mood or sync. You don't need all three - one well-chosen reference beats three conflicting ones.
• resolution - defaults to 2k. Native outputs are 480p/720p/Native1080p; the higher tiers (1080p/2K/4K) are upscales of a 720p base.
• duration - seconds, default 8, range 4–15.
• ratio - aspect ratio, default 16:9.
• seed - integer, default 0.
• generate_audio - default on, for clips with sound.
• return_last_frame - default off; turn on to get the final frame as an image for chaining.
• api_config / skip_error - the pack standard: per-request key/base override and placeholder-on-error.

Outputs: video (into preview/save), request_id, urls.

Install and first run

ComfyUI Manager → "BizyAirPlus", or:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart, register at bizyair.ai, set the key in Settings > BizyAirPlus > API Key (or BIZYAIR_API_KEY), and flip the BizyAirPlus action-bar toggle ON.

Gotchas

The usual setup trio - missing key, toggle off, wrong Python (python -m pip show bizyair-cloudberry). The R2V-specific lesson is that references fight each other: feed it a reference video in landscape and a reference image in portrait and the model will compromise awkwardly. Keep references consistent in aspect ratio and lighting, and remember every retry is a billed job - spend the extra minute curating the references before you queue.