FAL Seedance 2 Reference To Video (Soze)

The multi-reference sibling of FAL Seedance 2 Image To Video. Instead of one starting image, you hand this node up to nine reference images, three reference videos, and three reference audio clips - and then direct how they combine using plain text in the prompt, citing each one by name (@Image1, @Video2, @Audio3, and so on). It's a genuinely different mode from straight image-to-video: less "animate this picture," more "here's a pile of reference material, now describe the video I want made from it."

How it works

Every reference you connect gets uploaded to fal.ai (images directly; video and audio references go up via fal's upload_file, per their tooltips), and your prompt is expected to point at them explicitly by their @ tag rather than relying on the model to guess which reference does what. Everything else - the speed/resolution tradeoff, duration, aspect ratio, audio generation - works the same as the Image-To-Video sibling, because it's the same underlying Seedance model, just a different input mode.

The inputs and outputs that matter

• prompt (multiline, required) - this is where the real work happens. Its own tooltip spells out the citation convention: cite uploaded inputs as @Image1..@Image9, @Video1..@Video3, @Audio1..@Audio3. A prompt that never mentions @Image2 is effectively ignoring whatever you plugged into that slot.
• image_1..image_9 (IMAGE, optional) - up to nine reference images, each tagged individually in its tooltip as @ImageN.
• video_1..video_3 (VIDEO, optional) and audio_1..audio_3 (AUDIO, optional) - reference clips, same @ citation pattern.
• speed, resolution, duration, aspect_ratio, generate_audio, seed - identical to the Image-To-Video node: fast disallows 1080p, standard is required for it; seed 0 means fal picks.
• Optional end_user_id - usage attribution, generally leave blank.
• Outputs: video, video_url, config, status - same shape as the sibling node.

How to install it

Via ComfyUI Manager: search "Quality of Life Nodes for ComfyUI", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/SozeInc/ComfyUI_Soze.git
pip install -r ComfyUI_Soze/requirements.txt

Needs FAL_KEY in your environment, same as its sibling - and again, every run is a metered fal.ai call, not local compute.

Common issues & troubleshooting

A reference you wired in doesn't seem to affect the output. Check your prompt actually cites it. This node doesn't automatically blend everything you connect - the model only knows to use image_3 if your prompt says @Image3 somewhere.

Video/audio references fail to upload or time out. These go up separately via fal's file-upload path rather than inline like images - larger video/audio files will naturally take longer to upload before generation can even start, so a failure here can look like the whole node hung.

You only have one image and want simple image-to-video. This is the wrong node for that - use FAL Seedance 2 Image To Video instead, it's built for exactly that single-image case and doesn't require you to learn the @ citation syntax.

1080p/speed conflicts, billing surprises. Same as the sibling node: fast caps below 1080p, and every call spends real fal.ai credit tied to your FAL_KEY.