4. MJ 获取出图 (Get Result)
Pull the finished MJ grid and auto-split it into four
- image
- report
The last node in the MJ quartet (1. Imagine → 2. Query → 3. Action → 4. Get Result) and the one that pays out. Midjourney renders a four-up grid; this node fetches it and hands it to your graph as an IMAGE - and the genuinely useful part, it can auto-split that grid back into the four individual tiles. The other MJ steps are about commanding the service; this one is about getting pixels back into ComfyUI where your local upscaler, mask, or outpainting nodes can take over.
And that's the killer workflow combo this pack is built around: the closed model makes the grid, the local graph does the polish. It's the same "closed frontier + open post-processing" pattern the KB tracks - MJ's strength is the generation, and everything after it belongs to your stack.
The inputs that matter
Only two, both optional:
specific_task_id- the tooltip is the author's own words: "留空则自动拿最老的一个未下载成功任务" (leave empty to automatically take the oldest task not yet downloaded). So by default it drains the Wujiai queue in order; set an explicit task ID to fetch a specific render.split_grid-Yes (自动切分四宫格)(default) orNo (保持原图). Yes cuts the four-up into its four tiles; No keeps the whole grid as one image.
Outputs are image (IMAGE - your result, single or four tiles) and report (STRING - status/details).
Using it
Default behavior is the beginner-friendly path: after your Imagine and action complete, just run this node and it takes the oldest undownloaded successful task and delivers it. If you're juggling multiple jobs, set specific_task_id to be explicit. Keep split_grid on Yes when you intend to feed individual tiles to U/V actions or local processing; switch to No only if you want the raw grid for comparison.
Install
One pack install:
cd ComfyUI/custom_nodes
git clone https://github.com/jieg9341-lab/ComfyUI-nkxx
restart (or ComfyUI Manager, "ComfyUI-nkxx"). First launch auto-installs requests, pandas, openpyxl, yt-dlp, opencv-python, aiohttp. No models, no VRAM. Needs a Wujiai key from wujiai.org - though fetching a result that's already rendered shouldn't itself bill a generation, the render it's fetching absolutely did.
Common issues
- Nothing comes out: no "succeeded, not yet downloaded" task exists on this key. Run MJQueryTasks to confirm a task actually finished before expecting this node to deliver.
- The "oldest" surprise: with multiple completed jobs, default mode hands you the oldest, which may not be the one you want - use
specific_task_id. - Grid not splitting:
split_gridoff, or the channel returned a single image rather than a four-up (variations/upscales can come back as one image - in which case there's nothing to split, which is correct). - Trust note: the Wujiai-side nodes ship obfuscated in this pack. For a downloader that's lower-stakes, but the whole channel runs on a single author's reseller setup, so verify behavior on a cheap task first.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| specific_task_idopt | STRING | 留空则自动拿最老的一个未下载成功任务 | |
| split_gridopt | COMBO | Yes (自动切分四宫格) | 2 options: Yes (自动切分四宫格), No (保持原图) |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| report | STRING | — |