Superside Seedance Lite Image-to-Video

If you want ByteDance's Seedance video model in your ComfyUI graph, you're calling an API - there are no open weights to download, full stop. This node wraps Seedance Lite over fal.ai, and the reason to pick Lite over its Pro sibling is right in the name: it's the cost-efficient tier. When your deliverable is a 720p social cut or a motion test and nobody's checking pixels at 4K, Lite is the version your budget wants you to use.

What it does

Feed it a starting image (up to four reference images, actually) and a prompt, and it animates from there. The one-image case is the simple one: still photo → short video. The multi-reference case is where it gets interesting - up to four reference images let you give the model more context about subject, environment, and style than a single frame can carry, which matters when the video needs to stay true to a specific product or character across shots.

The inputs that matter

• reference_image_1 (required) plus reference_image_2…reference_image_4 (optional) - the stills to animate from.
• prompt - what happens in the video. The default is a girl catching and hugging a puppy; you'll want to replace it.
• resolution - 480p or 720p (default). This is where Lite's cost savings live.
• duration - 3–12 seconds, default 5. Longer clips cost more; start at 5 and extend only when you need it.
• aspect_ratio - auto or 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16. auto follows the input image.
• camera_fixed - locks the camera position so the scene doesn't drift around the subject. Good for product pans, bad if you want dynamic motion.
• seed - set a fixed seed for reproducible takes, or leave it random.
• enable_safety_checker - on by default, the API's content filter.

The output you have to handle

Here's the one thing beginners bounce off: the output is a single STRING - a video URL, not a ComfyUI VIDEO tensor. The node gives you a direct fal.ai link to the rendered clip. You'll want a node that downloads a URL to a file (or use the URL directly) rather than expecting a native VIDEO wire into SaveVideo. The pack's other video nodes (like SeedVR Upscale Video) do return native VIDEO - this image-to-video family just hands you a URL and lets you decide what to do with it.

Installing it

Part of comfyui-superside-nodes - clone, install requirements, restart:

cd ComfyUI/custom_nodes
git clone https://github.com/Superside/comfyui-superside-nodes.git
cd comfyui-superside-nodes
pip install -r requirements.txt

Paste your fal api_key into the node. No model downloads - Seedance runs entirely on fal's servers.

The honest take

Video is where API-node bills go from "coffee money" to "oh no" - the community's loudest objection to Seedance-in-ComfyUI is exactly this metering. Lite mitigates it, but you should still treat every render as a metered expense: iterate on stills and prompts before committing to a full-duration render, keep resolution at what the deliverable needs, and check the result URL before you burn a redo. It's the right tool for cheap, fast motion exploration with Seedance quality; it is not the tool for final 4K broadcast.