Midjourney 图像与视频（16 合 1）

This is the node that puts Midjourney - the most famous closed image service in the world - inside your ComfyUI graph, and it does it with a 16-operation dropdown: Imagine, Blend, Describe, edits, upscale, variation, reroll, zoom, pan, inpaint, remix, and image-to-video. Pick an operation and the node collapses to just the fields that action needs; the rest of the ~40 inputs hide themselves. If you've ever wished your MJ work lived next to your local masking and upscaling instead of in Discord, this is the closest you'll get without Discord.

Same pack deal as everything here: T8mars/ComfyUI_Seedance, API wrapper, key from api.seedance.nz, requests only, no local models.

How it works

Each operation maps to a documented Midjourney action on the gateway. The node maintains an explicit action spec with a per-action field whitelist, so hidden fields never leak into requests. Two mechanisms matter for chaining:

• task_id - connect the task_id output of one Midjourney node into the next one's task_id input to run Upscale/Variation/Reroll/Zoom/Pan/Inpaint/Remix on the previous result. That's how you build a full pipeline.
• buttons_json - the server returns the available buttons and their custom IDs; you can paste one into custom_id instead of fiddling with index.

Local images (image1..image4) get uploaded automatically; you can also paste public URLs into the matching image_urlN fields (don't set both for the same slot - the tooltip warns).

The inputs that matter

• operation - the steering wheel, with the Chinese use-note in the dropdown label.
• prompt - generation or edit instruction; supports external STRING nodes.
• speed (default relax - the cheap tier), size (common ratios, custom opens custom_size for any w:h), quality (default 1), version (default 8.2), style, seed.
• task_id / index / custom_id - the chaining trio for follow-up operations. index is 1–4 for image actions, 0–3 for video - that off-by-one is a classic trip.
• modal_mode / mask - for Inpaint: region mode wants a ComfyUI MASK (white is repainted). The documented no-mask outpaint is rejected upstream right now, so plan around region.
• The MJ parameter zoo (stylize, chaos, weird, tile, niji, iw, cw, sw, cref, sref, dref, dw, repeat, raw, draft, hd, stop, extra) - leave them alone until you know what you're doing; extra even lets you pass native --flags.

Outputs are the full spread: image1..4, grid_image, video1..4, text (Describe results), primary_url/result_urls, primary_path/result_paths, task_id, buttons_json, response.

Installing it

ComfyUI Manager → search "ComfyUI Seedance", or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
cd ../..
python -m pip install -r custom_nodes/ComfyUI_Seedance/requirements.txt

Restart, key from api.seedance.nz/console, Seedance API Config into api_config.

Common issues

• Follow-up action ignores your image - chained operations (upscale, variation…) run off task_id, not re-uploaded images. Connect the upstream task_id.
• Wrong image chosen - index starts at 1 for image actions. If your variation picks the wrong candidate, check it's in 1–4.
• Inpaint stuck in "MODAL" - that's a legal intermediate state where the API waits for the mask/region. Run Inpaint, let it reach MODAL, then feed the mask into a Modal/region step. Don't fight it.
• Outpaint refused - known upstream issue; use region mode with a mask instead.

Midjourney is the most opinionated model in this pack - its whole culture is Discord buttons and --parameters, and this node faithfully reproduces both. If you just want clean image generation without the ceremony, the pack's Qwen Image or Seedream nodes are the no-Drama option.