DIGIT MU Seedance 2 Character

Character consistency across videos is the holy grail of AI filmmaking, and the standard answer is "reference images, prayed over." MUAPI's seedance-2-character endpoint is a more direct answer: you give it 1–3 reference photos of a character, it builds a reusable character sheet on the server, and you can then use that registered character in Seedance video generations instead of crossing your fingers that the model keeps the face straight. This node is the ComfyUI front door to that endpoint.

Where it slots in: you're generating character-led Seedance clips through the DIGIT video nodes (or MUAPI directly), and you want the same character to survive across shots. Build the sheet once with this node, and the character lives on MUAPI's side for reuse. The output is also a literal IMAGE - the generated character sheet - so you can eyeball it, save it, or wire it into the graph.

How it works

The node uploads your reference image(s) to MUAPI's character endpoint, submits the request with your outfit/costume prompt, polls until the job finishes, downloads the resulting sheet, and hands you both the image and the metadata. Auth comes from the MUAPIAPP_API_KEY environment variable - the same key the DIGIT Seedance and MiniMax nodes use when provider is set to muapi.

The inputs:

• prompt - describes the character's outfit or costume. The default is "A red leather jacket with black jeans and white sneakers", which tells you the shape of the thing: this is a character look, described the way you'd brief a costume designer.
• character_name - an optional label stored with the MUAPI character. Name it something memorable; you'll reference it later.
• image1 - required. The primary reference. The tooltip is worth trusting: clear frontal or 3/4 photos work best.
• image2, image3 - optional extra references. Different angles of the same person help the sheet lock onto the right identity.

Outputs:

• character_sheet - the generated sheet as an IMAGE tensor.
• sheet_url - where the sheet lives (MUAPI's storage).
• request_id - the MUAPI job ID, useful if you need to open a support ticket or debug.
• status - a human-readable summary of what happened.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/thedepartmentofexternalservices/comfyui-digit.git
cd comfyui-digit
pip install -r requirements.txt
export MUAPIAPP_API_KEY=your_muapi_key

(Or ComfyUI Manager → search comfyui-digit → install.) Restart ComfyUI, look under DIGIT.

Common issues

The dominant failure mode is the key: no MUAPIAPP_API_KEY exported and the node can't authenticate. Set it before launching ComfyUI. After that, quality is about your references - a blurry phone photo or a group shot where the model can't tell who the character is will produce a sheet that doesn't lock identity. Follow the tooltip: clear frontal or 3/4 shots, one person per reference.

Worth knowing about the ecosystem this lives in: MUAPI is a unified async API layer over several Chinese video providers, and its whole selling point is lighter content filtering than the stock endpoints - people pass through where fal's Seedance blocks likenesses of real people. That's the reason artists route here, and it's also the reason you should read MUAPI's own terms before you put a real person's face through it. The character sheet is stored on MUAPI's servers, so your references leave the machine and persist there.