【破綻対策あり】 僕の完璧主義 😆

① 最初に生成されたイラスト (Shiitake-Mix v2 SDXL - 1024 x 1024 px)
② アップスケール (Wai-NSFW-Illustrious v15 SDXL - 2048 x 2048 px) とトリミング後
③ 2回目のアップスケール後 (2048 x 2048 px)
④ MS Paintで編集した後 (ブラシツールを使用)
⑤ もう一度SDXLで生成された (Wai-NSFW-Illustrious v15 SDXL、i2i、8ステップ、6%ノイズ除去、アップスケールがない)
⑥ もう一度MS Paintで編集した (今回、少しだけ)
⑦ もう一度SDXLで生成された (Wai-NSFW-Illustrious v15 SDXL、i2i、4ステップ、3%ノイズ除去、アップスケールがない)
⑧ MS Photosアプリで色補正を行った後

終了しました！

よく使うアプリ (すべてが無料！)： MS Paint, MS Photos, GIMP, Inkscape

（残りのテキストについては、DeepLかGoogle翻訳を使用してください。 ごめん！ 🙇‍♂️）

-- English --

[With Corrections] My Perfectionism 😆

① Initial generation (Shiitake-Mix v2 SDXL - 1024 x 1024 px)
② After upscaling (Wai-NSFW-Illustrious v15 SDXL - 2048 x 2048 px) and cropping down
③ After a 2nd upscale (Wai-NSFW-Illustrious v15 SDXL - 2048 x 2048 px)
④ After editing in MS Paint (using the brush tool)
⑤ Generated again using SDXL (Wai-NSFW-Illustrious v15 SDXL, i2i, 8 steps, 6% denoise, no upscaling)
⑥ Edited again in MS Paint (only slightly this time)
⑦ Generated again using SDXL (Wai-NSFW-Illustrious v15 SDXL, i2i, 4 steps, 3% denoise, no upscaling)
⑧ After color correction using the built-in Photos app (Windows)

All done!

Commonly used apps (all free!): Paint (Windows), Photos (Windows), GIMP, Inkscape

On generating the first picture, I decided I wanted to just focus on the part above the couch line, so I cropped the rest out and let that get pushed down to 512 x 512 px by my Comfy UI workflow. The resulting upscale wasn't quite crisp and detailed enough, so I upscaled it again.

Then, I found things I wanted to paint out: the part of the shirt sticking out next to her arm, the portion of her ear inside of the hair that seemed off to me, the extra eyelid fold, the strange artifacts on the bottom of the image, the slightly narrow neck which ended before going behind her hair, etc. Finally, I changed something to fit my personal preference - I made the pupils dark and more rounded.

All of this was done with the dropper tool and the brush tool in Paint. It's fine to leave the edges a bit messy (especially if editing before upscaling to remove logos, fix larger issues, etc.). The subsequent generations smooth out gradients and straighten lines as part of the noise removal process. If you still see color bands in gradients, slightly blur the area in GIMP with the Gaussian Blur tool and then do an i2i generation to bring everything back to uniformity.

For bigger edits, use more advanced software (like GIMP) that supports layers and can do free rotation, background removal, and all kinds of other things. Paint technically supports layers, but it's not great for that. I also use this same method to separate merged fingers, paint out extra fingers, fix deformities, remove weird extra lines, etc.

When I do i2i (Image to Image) generation between edits, I want to smooth out my edits without making too many changes, so I disable upscaling and stick with a small number of steps and denoise percentage. What I start with depends on how rough my edits look and how big they were:

- Pretty obvious/large changes: 10 steps, 8% denoise
- Moderate edits: 8 steps, 6% denoise
- Light edits: 5 steps, 4% denoise
- Very slight adjustments: 3-4 steps, 2-3% denoise

I then make smaller and smaller edits and use less and less steps/denoise% until I'm happy with the result. It's kind of like slowly honing in on a target! 🎯

I hope this helps and is fairly easy to pick up for basic corrections!