FL Fal Seedance i2v
FL_Fal_Seedance_i2v — image-to-video with ByteDance Seedance via the Fal.ai API
- image
- frames
- video_url
- status_message
Feed it a starting image and a prompt, get back a video. FL_Fal_Seedance_i2v calls ByteDance's Seedance v1 Pro image-to-video model through Fal.ai's hosted API, so the generation happens on Fal's servers, not your GPU. It's the paid-cloud alternative to running Wan locally - you trade money and an API dependency for not needing a 24GB card and a 20-minute render.
Seedance is one of the closed, API-only video models ByteDance ships (the same company hands the local ecosystem a lot of free components, but the flagship generation products stay behind the paywall). It's genuinely good, and if your local hardware can't run Wan 2.2 at a resolution you're happy with, renting Seedance by the clip is a reasonable call.
The one feature worth knowing about
Beyond the usual start-image-plus-prompt, this node supports multi-segment prompting. The cut_prompt_1 through cut_prompt_4 inputs let you describe successive shots - the node stitches them with [cut] markers - so instead of one continuous action you can script a little sequence: first this, then that. It's a nice way to get a multi-beat clip out of a single call rather than generating pieces and editing them together.
The inputs and outputs
api_key- your Fal.ai key.image- the starting frame.prompt(multiline) - the main motion/scene description.cut_prompt_1–cut_prompt_4- optional per-segment prompts for a multi-shot clip.resolution- 480p / 720p / 1080p.duration(3–12 seconds) - clip length.camera_fixed(default on) - lock the camera, or let it move.seed- −1 for random.
Outputs: frames (the video as an image batch you can save or post-process), video_url (the hosted result on Fal), and status_message (what happened - read this when it fails).
How to install it
Part of filliptm's Fill-Nodes pack. ComfyUI Manager → search ComfyUI_Fill-Nodes → Install → restart, or:
cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI_Fill-Nodes
pip install -r ComfyUI_Fill-Nodes/requirements.txt
No local model. You need a Fal.ai account with credits and an API key.
Common issues & troubleshooting
It costs money per clip, and cost scales. Higher resolution and longer duration bill more on Fal. 1080p at 12 seconds is not the setting to leave on while you iterate - prototype at 480p and short durations, then do your final at full quality.
Auth / job failures. Check status_message first. A bad key, exhausted credits, or a rejected input all report there rather than crashing the node. If the job just never returns, it's usually a Fal-side queue or a content rejection.
Motion is too stiff or too wild. camera_fixed on gives you a locked, stable shot - good for product-style clips, boring for action. Turn it off for movement, but expect more drift. Seed helps you reproduce a take you liked.
Local vs cloud, honestly. If you have the hardware, Wan 2.2 locally is free per clip and uncensored. Seedance-via-Fal is what you reach for when you don't, or when you specifically want Seedance's look. Don't wire this into a pipeline you plan to run thousands of times without doing the math on the bill first.
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| api_key | STRING | — | |
| image | IMAGE | — | |
| prompt | STRING | — | |
| cut_prompt_1 | STRING | — | |
| cut_prompt_2 | STRING | — | |
| cut_prompt_3 | STRING | — | |
| cut_prompt_4 | STRING | — | |
| resolution | COMBO | 720p | 3 options: 480p, 720p, 1080p |
| duration | INT | 63–12 | — |
| camera_fixed | BOOLEAN | true | — |
| seed | INT | -1-1–999999999 | — |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| frames | IMAGE | — |
| video_url | STRING | — |
| status_message | STRING | — |