Seedance 2.0 Mini

ByteDance - TikTok's parent - runs the biggest open-component shop in this field: everything it makes for generation stays API-only, and Seedance is its video model line. Seedance 2.0 Mini (bytedance:[email protected]) is the compact, cheaper rung of that ladder, and this node drops it into ComfyUI as a single text-to-video block with native VIDEO output. No ByteDance account, no separate API, no signup dance - one Runware key covers it.

The "Mini" is the honest part. Seedance 2.0's full-size model is the flagship; Mini trades some fidelity for speed and price, which makes it the draft-and-iterate node. If you're storyboarding, testing camera language, or generating placeholder motion to composite over, Mini is the rung you want. When a take has to be final, you swap up - but most takes don't need to be final.

How it works

Standard Runware mechanics: the node sends a videoInference request with the bytedance:[email protected] AIR, your reference images and frame images ride up as base64, and the finished video downloads back into ComfyUI as a VIDEO object. The interesting part is what Mini can ingest. Beyond plain referenceImages and frameImages, there's a referenceVideos socket - a builder node that lets you feed an existing clip as a motion or style reference, which is how you get consistency across a series of shots. Want audio in the result? settings.audio defaults to on, generating synchronized sound rather than silent footage.

The inputs that matter

• positivePrompt (required) - what you want to see happen.
• size - a dropdown of 15 presets from 480p up through 720p, in 16:9, 4:3, 1:1, 3:4, 9:16, 21:9. This is the "only one of resolution or dimensions is sent" control; pick a preset rather than fighting with dimensions.
• duration - 4 to 15 seconds, default 5. Total frames = duration × model fps, and every second costs you.
• referenceImages / frameImages / referenceVideos - the character-consistency and motion toolset.
• numberResults - up to 4 variations, different seed each.
• advanced_json - the escape hatch, documented for inputs.referenceAudios, i.e. feeding reference audio into the generation.

Installing it

The pack, not the model. ComfyUI Manager → search Runware → install → restart. Manual:

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

Add a key from runware.ai/api-keys in ComfyUI Settings → Runware API key (or RUNWARE_API_KEY). Nothing downloads; Seedance is cloud-only, which is the point.

Where people get burned

• Mini is Mini. Don't expect the full 2.0's fidelity at Mini's price. It's a drafting tool; if you judge it against flagship output you'll be disappointed.
• size and duration are the cost knobs. 720p at 15 seconds is a different bill than 480p at 4. Run the graph once to see the title-bar cost before committing to a batch.
• Reference videos need the builder. The referenceVideos socket only fills from the pack's video-reference builder node - you can't drop a file path into it directly.
• Audio is on by default. If you want silent footage for sound design later, flip settings.audio off; a generated soundtrack you don't want is still paying for itself.

Seedance 2.0 Mini is the cheap seat on a good video model. For iteration volume, it's hard to beat.