Seedance 2.0 Mini runs as a selectable model on the existing ByteDance Seedance 2.0 video nodes — Text to Video, First-Last-Frame to Video, and Reference-to-Video — and ships with dedicated templates.
Where Mini fits
Mini gives you the Seedance 2.0 cinematic toolkit at a lower cost per run. It runs on the same Seedance 2.0 video nodes and takes the same prompts. The one tradeoff is resolution: Mini outputs 480p and 720p, while native 1080p and 4K stay on full Seedance 2.0 and 2.0 Fast. For drafting, iterating, and generating at volume, Mini gets you the same look for less.
What Mini does
Mini reads a prompt as a full visual narrative rather than a single isolated clip, generating structured multi-shot sequences with coherent pacing across shots. Character identity stays stable across those shots, so a character remains visually aligned through an entire sequence — perfect for storytelling and branded content. It maps natural language directly to scene composition and visual style, producing multi-scene output from plain text.
Native 4K output
Native 4K is a full Seedance 2.0 feature — not Mini, which tops out at 720p. Full Seedance 2.0 generates native 4K with 10-bit color. Fine detail resolves cleanly at 4K — hair strands, fabric textures, architectural lines, and material surfaces — and holds through complex motion, not just on static frames. The 10-bit color is native rather than upconverted or wrapped at export, which means real grading headroom, smoother gradients, and cleaner output to work from downstream. This is the tier to finish on when a shot needs to ship at high resolution.
Getting Started with Mini
Update ComfyUI to the latest version, or open Comfy Cloud.
Load a Seedance 2.0 Mini template from the Templates panel, or add a ByteDance Seedance 2.0 node and select Seedance 2.0 Mini as the model.
Set resolution to 480p or 720p, drop in your prompt and run.
The three input modes described here solve different planning problems. For a first/last-frame shot, I would check whether the transition respects both endpoint compositions; for a reference-led shot, I would check the subject through the changing views. Keeping those review goals separate makes a small test more useful than judging the model name or output size alone. Disclosure: I represent the independent WAN30 website (https://wan30.io/), an AI video creation platform.
I can't see, on cloud, the 4k option, only 1080.