Concurrent Submit | Zhenzhen Seedance 2.0 Ref2Video Fal

Seedance 2.0's reference-to-video mode, served through fal's queue. The "Ref2Video" name is the point: this node can take up to nine reference images, three reference videos, and three reference audio clips in a single generation, and you address them from the prompt with the pack's @ syntax - @Video1 in your prompt means "use reference video #1," @Audio1 means "use reference audio #1." That's the tool you want when you're trying to keep a character, a motion style, or a voice consistent across generated clips.

It's a video node, so: shared video pool (10 workers), COMFLY_VIDEO_FUTURE task output, and Concurrent Collect Videos (ComflyConcurrent_Video_Await) is your mandatory partner - up to 10 tasks, returned as video_1…video_10 in order. The submit node itself never blocks; the waiting happens in the collector, in parallel.

The inputs

• prompt - your direction, including the @Video1 / @Audio1-style references. The tooltips on the reference URL fields spell out the syntax.
• image1 … image9 - still references (frames, character shots, style boards).
• video_url1 … video_url3 - reference clips as hosted URLs (MP4/MOV).
• audio_url1 … audio_url3 - reference audio (MP3/WAV), e.g. for a voice or sound to carry over.
• resolution - 480p / 720p / 1080p.
• duration - auto or 4–14 seconds.
• aspect_ratio - auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16.
• generate_audio - on by default; Seedance will synthesize synchronized sound (sfx, ambient, even lip-sync) for the output.
• seed - fal max 65535, 0 = random.
• image_way - image_url or base64 for the stills.
• poll_interval / max_poll_attempts - default 600 × 6s = 1 hour.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Or ComfyUI Manager → search Comfyui-zhenzhen → restart, token into api_key (ai.t8star.org overseas; api.seedance.nz for the domestic discount).

Gotchas

• The @ syntax is mandatory, not decorative. Feeding reference slots and not referencing them in the prompt is a common way to get a generation that ignores your refs. Address them explicitly.
• Reference videos cost like output. Seedance counts reference clip seconds toward the price - the pack's own changelog calls this out. Trim refs to what you actually need.
• Timeouts with heavy references are realistic. Nine images + three clips + audio is a slow job. If it times out at the 1-hour default, check the async-task page before resubmitting - and remember fal lanes are pre-deducted and settled later.