Nodes/ComfyUI_Fill-Nodes/FL Fal Seedance 2.0 Reference to Video
ComfyUI Node

FL Fal Seedance 2.0 Reference to Video

Seedance 2.0 in your ComfyUI graph — via fal, pay per call

By filliptm·Created 3 years ago·Updated about 24 hours ago· 633
FL Fal Seedance 2.0 Reference to Video
  • reference_images
  • reference_videos
  • reference_audios
  • video
  • video_url
  • seed
  • request_id
prompt
resolution720p
durationauto
aspect_ratioauto
generate_audiotrue
bitrate_modestandard
end_user_id
fal_api_key

Straight talk first: ByteDance's Seedance video models have no open weights, so there is no local option at any VRAM. If you want Seedance 2.0, you call it. This node is one way to do that from inside ComfyUI - it wraps fal.ai's bytedance/seedance-2.0/reference-to-video endpoint, uploads your references, submits the job, polls until it's done, and downloads the result back into a native VIDEO output. From the canvas it looks like a sampler. Underneath it's an HTTP client with your fal key baked in.

"Reference to video" is the useful mode here. You're not just text-to-video: you can feed it up to 9 reference images, up to 3 reference videos (2–15 seconds combined), and up to 3 audio references (≤15 seconds combined), and the model composes a new clip from all of them plus your prompt. In the prompt you refer to connected assets by their socket order - @Image1, @Video1, @Audio1, and so on. That's the part people fumble first, so: the tag numbering is tied to the order of the connected autogrow sockets, not the order you wrote them.

Inputs that matter

  • prompt - describe the video, using the @Image1/@Video1/@Audio1 tags. Can't be empty.
  • reference_images / reference_videos / reference_audios - autogrow sockets; connect as many as you have. JPEG/PNG/WebP for images. Note: reference audio requires at least one image or video reference - you can't drive it from audio alone.
  • resolution - 480p to 4k (4k shows up in fal's current schema), default 720p. duration - auto or 4–15s. aspect_ratio - auto or 21:9 down to 9:16. generate_audio - on by default, gives you synchronized sound effects, ambience, and speech.
  • fal_api_key - leave blank to use the FAL_KEY environment variable; fill it in to override per node.

Outputs: video (the downloaded clip as a ComfyUI VIDEO tensor), video_url, seed (the fal-provided seed, so you can reproduce), and request_id (for debugging and billing).

Install and key

Same pack as everything else - Manager search "ComfyUI_Fill-Nodes", or clone the repo into custom_nodes/ and restart. fal-client comes with the requirements, so the real setup step is the key:

export FAL_KEY="your-fal-key"

Get one at fal.ai. Without it the node throws immediately.

The honest cost conversation

This is a metered cloud call, and the community has strong feelings about that. Seedance 2.0 arriving in ComfyUI as an API node drew a famously blunt "Go kick rocks" from the local-first crowd, and the objection is real: a 10-second 1080p Seedance clip is priced per job, and video jobs are where a session gets expensive fast. Your prompt and references also leave your machine for fal's servers, subject to their moderation - you can't abliterate a closed model, so anything ByteDance refuses, the node refuses. The pro-API case is just as real: no GPU required, no weights to babysit, and this route (pay-per-call via fal) sidesteps Comfy's own prepaid-credit Partner-Node storefront if you'd rather bill per call. Just don't treat it as a free local replacement, because it's neither free nor local.

Category🏵️Fill Nodes/AI

Inputs (11)

NameTypeDefaultDescription
promptSTRINGDescribe the video. Refer to connected assets in order as @Image1, @Video1, @Audio1, and so on.
reference_imagesCOMFY_AUTOGROW_V3Up to 9 reference images. JPEG, PNG, and WebP are supported by Fal.
reference_videosCOMFY_AUTOGROW_V3Up to 3 native VIDEO references with 2 to 15 seconds combined duration.
reference_audiosCOMFY_AUTOGROW_V3Up to 3 audio references with at most 15 seconds combined duration. Audio requires at least one image or video reference.
resolutionCOMBO720pOutput resolution. 4k is present in Fal's current machine-readable schema.
durationCOMBOautoOutput duration in seconds, or auto to let Seedance decide.
aspect_ratioCOMBOauto7 options: auto, 21:9, 16:9, 4:3, 1:1, 3:4, +1
generate_audioBOOLEANtrueGenerate synchronized sound effects, ambience, and speech.
bitrate_modeCOMBOstandardHigh requests a larger, higher-bitrate output file.
end_user_idoptSTRINGOptional end-user identifier sent to Fal.
fal_api_keyoptSTRINGOptional Fal key override. Leave blank to use the FAL_KEY environment variable.

Outputs (4)

NameTypeDescription
videoVIDEO
video_urlSTRING
seedINT
request_idSTRING