局部细节呈现

Here's a fact about diffusion that never stops being annoying: a face that's 70 pixels wide in a 1024px image comes out a smear, no matter how good the model is - the latent just doesn't have budget to spend there. The local fix is the whole "detailer" category: detect a region, crop it, upscale it, re-diffuse it, stitch it back. This node is that loop in one cloud call. You hand it an image and a mask, and the model re-renders exactly the region you masked at high detail. No ADetailer, no SEGS plumbing, no VRAM - just a mask and a queue.

How it works

The mechanism is the classic detailer loop, split across the network. On your side the node base64-encodes the image and its mask and POSTs both with the model id detail-photo. On MojieAI's side, the region under the mask gets cropped, upscaled, re-generated with the surrounding context as conditioning, and composited back. You get the "crop, upscale, re-render, paste" sequence that local detailers run - the same pipeline the whole masking-and-detailing category is built on - except you supply the mask yourself instead of a detector finding it. That's actually a feature: no detector misfiring on a rock formation it decided was a face.

The inputs that matter

• input_image (IMAGE) - your source.
• mask (MASK) - the whole point. Paint the region you want detailed. White = detail me, black = leave it alone. Anything with an edge is fine since the server blurs and feathers the boundary, but a soft mask edge makes the seam far less visible.
• seed (INT, default 0) - reproducibility.

One output, output (IMAGE). Save it, or feed it onward.

Installing it

Part of the comfyui-MJAPI-party pack. ComfyUI Manager → search "mojieapi_party" → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/MoJIeAIGC/comfyui-MJAPI-party.git

Restart, register at mojieaigc.com, and set your key in config.ini (leave the BASE_URL line as-is):

[API]
KEY = your_api_key
BASE_URL = https://www.mojieaigc.com/v1/completions

Restart again to load the key. No model downloads - the heavy lifting is a server call.

Common issues

• The masked region looks "added" rather than "enhanced." If the detail pass re-invents the area instead of refining it, your mask probably included too much context or the source region was too small to work from. Mask tighter around the actual feature you care about.
• You're not sure how to make a MASK in ComfyUI. Drag your image into the canvas, right-click → "Open in MaskEditor," paint, and out comes a MASK tensor. Any mask-painting node works; the pack doesn't ship one.
• This is a metered API call per run. It's a genuinely nice tool for fixing one hero product shot, but if you're doing this across a hundred images you'll feel it on the balance - the "batch of regions in one graph" economics that make local detailers attractive apply here too.