Zhenzhen Seedance 2.0 Ref2Video Fal

Seedance 2.0's reference-to-video mode is the "give me a whole scene and remix it" version of video generation: feed it a pile of source material and it makes a coherent new clip out of it. Comfly_seedance2_fal is the zhenzhen pack's FAL route to exactly that - up to 9 reference images, 3 reference videos, and 3 audio clips in, one generated video out, through the fal.ai queue. If the low-price Comfly_seedance2_low_price node is the cheap domestic path, this is the overseas one with more per-input flexibility on the audio/video reference side.

It's a FAL node, so the key goes in its own api_key field and the traffic runs through zhenzhen's proxy (https://ai.t8star.org/fal standing in for queue.fal.run). Submit, poll, download - the usual queue dance, prepaid credits and refunds included.

The inputs that matter

• prompt - required, and the reference tags live here. The tooltips are explicit: reference a video with @Video1 / @Video2 / @Video3 and audio with @Audio1 / @Audio2 / @Audio3 inside the prompt text. Images are referenced positionally - check the shipped workflow examples for the exact convention, because getting this wrong silently produces a clip that ignores half your inputs.
• image1–image9 - reference frames wired in as IMAGE tensors.
• video_url1–video_url3 - reference videos as public URLs (MP4/MOV). Note these are URL strings, not video tensors - the node doesn't upload video tensors the way the image inputs work.
• audio_url1–audio_url3 - reference audio as public URLs (MP3/WAV).
• resolution - 480p/720p/1080p, 720p default.
• duration - auto or 4–15 seconds.
• aspect_ratio - auto or 21:9 down to 9:16.
• generate_audio - on by default; produces synced sound (effects, ambient, even lip-sync).

Outputs: video (a VIDEO tensor - straight into Save Video), video_url, and response. Seed is capped at 65535 like all FAL seeds, and 0 means random.

Installing it

No models, no VRAM - standard API-pack install:

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

or ComfyUI Manager, search "Comfyui-zhenzhen", then restart. Paste a zhenzhen token (overseas https://ai.t8star.org) into api_key.

Common issues

• Clip ignores your references - almost always a prompt-tag mismatch. The @Video1/@Audio1 tags have to match the slot, and the order matters. Re-read the tooltips on the URL inputs.
• Reference URLs expire - the shop notes that uploaded/linked media URLs can be time-limited. Re-upload or re-point before a long queue wait.
• Timeouts - default poll is 6s × 600 = one hour. Seedance multi-reference jobs run long; if it times out, the job may still complete on fal's side, so check before re-paying.

This is the node to grab when your video needs to be assembled from stuff - a character across several shots, a scene with existing footage, a voiceover you already have - rather than generated cold from a single prompt.