抠抠图-中阶印花提取功能

KoukoutuImageExtractV2 (抠抠图-中阶印花提取功能) is the sibling of the standard print extractor, and the pitch is one line: same inputs, same outputs, better model. Where the standard node sends your garment photo to the image-extract model, this one sends it to image-extract-v2, which the pack calls the "mid-tier" (中阶) extraction model - its selling point is cleaner results on harder prints.

If you read the standard extractor's article, everything carries over unchanged: it's part of comfyui-koukoutu, an API wrapper around the Koukoutu cloud, so the extraction runs on their servers, costs credits, and sends your images off-box. Same async dance - upload, create a task, poll once a second with a progress bar, download the result. Don't install anything extra; both extractors ship in the same pack under image/koukoutu.

Inputs that matter

• image - the photo containing the print.
• api_key - required; koukoutu.com/user/dev, with credits.
• extract_type - required freeform string telling the model what to look for: 服装 (clothing), 鞋包 (shoes/bags), 配饰 (accessories), or whatever describes your case.
• resolution - 1k or 4k, default 1k. 4k for print files, longer and pricier.
• size - 19 output aspect presets, default 0:0 (original dimensions).
• skip_error - on by default.

Outputs

image plus message. On error with skip_error on, you get the original image back with the error text on message - same fallback as the whole pack.

When to reach for V2 over the standard node

Honestly? Run both on your hardest print and compare - that's the intended workflow, and the reason they ship as separate nodes instead of a dropdown. If the standard extraction comes back with background bleed or ragged edges around a complex pattern, V2 is the next thing to try before you start hand-correcting. The tradeoff is it's described as the heavier model, so expect it to take a little longer per job. For a simple flat print on a clean shirt, the standard node is usually fine and cheaper on time; V2 earns its keep on wrinkles, busy backgrounds, and fine detail.

Install

One install, all nine nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/zhenzi0322/comfyui-koukoutu
pip install -r requirements.txt

Or search "koukoutu" in ComfyUI Manager, then restart. Dependencies are minimal (requests, Pillow, numpy, torch), no model downloads, no VRAM. As with every node in this pack, it's an API wrapper that holds a key and sends your images off-machine - install from the official repo and don't reuse keys you care about elsewhere.

Troubleshooting

Same code family as the standard node: 401 bad key, 403/409 no credits, 406/413 file over 15MB, 407 source too small, 422 bad params (often an unrecognised extract_type), 429 rate limit. If V2 returns worse-than-standard results on a specific pattern, that's a per-image thing - swap back to the standard node rather than fighting it.