1. MJ 生图提交 (Imagine)

Midjourney is a closed service, full stop - no weights, no local path, the community's poster child for "you can only call it." This node is the call, on the Wujiai channel, and it's the "1." in a numbered four-node MJ flow: 1. Imagine → 2. Query → 3. Action → 4. Get Result. The author has essentially recreated the MJ app inside ComfyUI, including the parts MJ users expect: style reference, character reference, and the U/V/Reroll buttons you normally click in Discord.

MJImagineSubmit is the submission front door. It takes a prompt, a model version, an aspect ratio, and up to three flavors of reference image, and it fires an Imagine - the "make me a 4-up grid" step - into the Wujiai service. Output is a single report STRING: the receipt identifying the task. That task is what the rest of the flow chases.

The inputs that matter

• prompt (multiline) - your actual MJ prompt. Don't put --ar or --sref flags in here; those are separate inputs.
• model_version - v7 (default), v6.1, v6.0, v5.2, or the anime line niji 7 / niji 6. This is the MJ version selector, straight from the app.
• aspect_ratio - 1:1 (default), 16:9, 9:16, 3:4, 4:3, 2:3, 3:2.
• extra_args (optional) - the tooltip is the author's own description: "可选附加参数，例如: --iw 1.5 --cw 50". So this is where MJ flags like image weight go.
• image_prompts (optional IMAGE) - "常规垫图 (图生图)": plain image-as-input.
• sref_images (optional IMAGE) - "风格参考 (--sref)": the style-reference socket.
• character_refs (optional IMAGE) - "角色/物体参考 (自动适配 --cref / --oref)": character or object reference, auto-mapped to --cref/--oref.
• api_key (optional) - falls back to the Wujiai key.

How it fits the wider workflow

The clever part is downstream. The Wujiai side of this pack adds frontend button interaction - the node panel exposes U/V/Reroll buttons you click directly, feeding MJActionSubmit. So your workflow becomes: Imagine here, check the grid with MJGetResult, then act on it (upscale a tile, make a variation, reroll) without leaving ComfyUI or typing a single slash command. If you've ever managed an MJ pipeline through Discord and copy-pasted URLs, this is the version where the graph does it.

Install & prerequisites

Same pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/jieg9341-lab/ComfyUI-nkxx

restart (or ComfyUI Manager, search "ComfyUI-nkxx"). First launch auto-installs requests, pandas, openpyxl, yt-dlp, opencv-python, aiohttp. No local models - this is a pure API wrapper, so the only hardware requirement is a network card. You need a Wujiai key from wujiai.org, and MJ calls are metered.

Common issues

• Report but no grid: normal - Imagine only submits. You fetch the grid in step 4.
• Ignored flags: put MJ flags in extra_args, not the prompt, or they may be treated as literal text.
• Key missing: the Wujiai key has to be reachable - node field, WUJIAI_KEY env var, or pack-level config. A key that works for Grsai nodes won't satisfy this one.
• Trust note: the Wujiai-side nodes ship obfuscated in the pack, so the Imagine call's exact behavior is behavior-only verifiable. It's a paid channel - test one cheap prompt first.