Seedance Text-to-Video (fal)

Seedance is ByteDance's text-to-video model - the same lab behind SeedVR2, the upscaler half the community quietly runs, and PuLID. It's one of the stronger closed video models going, and this node lets you prompt it from a ComfyUI graph through fal.ai. Type a scene, get a clip. No image needed - this is the pure text-to-video path.

As with the rest of this pack, it's a cloud call: no weights locally, no GPU on your end, billed per generation. Seedance has no open weights, so fal is the way in.

How it works

You describe a scene in the prompt and Seedance generates a video from scratch - no starting frame, so the model invents the composition, the subject, and the motion all at once. That makes text-to-video harder to control than image-to-video (you're not anchoring anything), so prompt clarity matters more here than almost anywhere else in the pack.

The inputs and outputs that matter

• prompt (required, multiline) - the whole brief. Describe subject, setting, and motion, because there's no input image doing any of that work for you.
• resolution - 480p or 720p. 720p for the final, 480p for cheaper test runs.
• duration - 5 or 10 seconds. Ten gives motion room to breathe and costs more.
• aspect_ratio - 16:9 (default), 4:3, 1:1, or 9:21. Frame for where the clip will live (landscape for YouTube, 9:21 for a tall phone crop).
• camera_fixed (default off) - turn this on to lock the camera and get only subject motion; leave it off and Seedance will move the camera too. This is the single most useful knob for controlling the feel of the shot.

Optional seed (default -1 = random) if you want to reproduce a result.

Output is a STRING - a URL to the video on fal, not a previewable VIDEO. Send it through the pack's Fal Save Media from URL node to download it, or into a downstream fal node's *_direct_url input to keep chaining. The URLs expire; save what you like.

How to install it

Part of gokayfem's ComfyUI-fal-API pack. ComfyUI Manager: search ComfyUI-fal-API, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI-fal-API.git
pip install -r ComfyUI-fal-API/requirements.txt

Add your fal key (copy config.ini.example → config.ini, paste from fal.ai/dashboard/keys, or export FAL_KEY=your_key), restart. Nothing downloads locally.

Common issues & troubleshooting

The camera won't sit still. Set camera_fixed to on. Off is a moving camera by default, which is often not what people expect from a "text to video" node.

Composition is a mess. Text-to-video gives the model total freedom, so a thin prompt gets a chaotic result. Spell out the shot: who or what, where, doing what, from what angle. If you want tight control over the framing, consider an image-to-video node instead where a still anchors the scene.

Output won't preview. It's a URL string. Download it with Fal Save Media from URL, then preview the file.

Insufficient funds / credit lag. fal charges per call and is a legit platform, but its billing has occasionally been slow to reflect top-ups. Let credits settle rather than repaying. fal's own error text (policy, quota, validation) shows up in ComfyUI - read it.