AiPlaygroundWM

 In the middle of generate of girl in swimsuit, 2.5x hires I was trying to upscale useing " Soft inpainting " new features in stable diffusion WebUI 1.9v I don't know how to use so start with default settings. By accident Tiled diffusion default setting 2x, not 1.3x as I use. !!!!This is before inpainting.  Almost no need to inpainting. I have to learn and study how to use detailly " Soft inpainting "! PS inpaint and edit done.  Final version. Size: 2560x3200

In the middle of process hires 2.5x  Upscaler : 8x_NMKD-Superscale_150000_G Where is all the detail that I expect? 4x-UltraSharp Stable Diffusion WebUI Update 1.9 There are new upscaler in it DAT x2 Similar quality 8x_NMKD-Superscale_150000_G. Low detail. dat 3x ...What's the difference? DAT 4x Conclusion: Just keep use 4x-UltraSharp

prompt : (full body:1.3), a lady, (solo focus:1.4),  (look at viewer:1.2), long hair, __hair_color__, __hairstyle__,  __color__ (__revealing_swimwear__), (curvy body), (saggy breast:1.22),  (huge breast:1.38), narrow-waist, (huge_pelvic:1.28), (thick thigh:1.37), (highheels:1.2), (__pose__), __WAKA/background__, (best quality, realistic, detail, bokeh, shallow depth of field:1.4), QuickHands, <lora:QuickHandV2:0.6> negative prompt : (fused highheels, fused feet), (multiple view:1.4), (finger_ring, finrger_nails:1.4), extra hand, extra foot, (missing finger, extra fingers, fused fingers:1.4), monochrome, greyscale, (text, signature, watermark:1.4), mutated, (cropped, breast out, head out of frame, worst quality,bad quality:1.4)   Base 1.5 hires (full body:1.3), a lady, (solo focus:1.4), (look at viewer:1.2), long hair, Chestnut hair, gradient hair, (shiny skin, oiled skn:1.4), azure_color (The Cutout Mesh Monokini, Flaunting a monokini with strategic cutout...

 In the middle of generate spider gwen, Background buildings are too messy. So I'm gonna find out what lora is effected to background. 1 remove fisheye lora Still mess. 2. remove  spider_gwen lora Still... 3. Remove pakour lora I gotcha!! 4. Only remove parkour and other lora all mix The fisheye lens lora seems to be affecting the background as well. Now just adjust weight of pakour lora 1. <lora:Parkour:0.7>   Mess. 2. <lora:Parkour:0.6> Mess. 3. <lora:Parkour:0.5> Mess. 4. remove fisheye lens lora and <lora:Parkour:0.5> Looks ok. 5. <lora:Parkour:0.5> <lora:fisheye_slider_v10:1> Barely see fisheye lens effect. And the background is little mess. So test is done. Maybe I have to change process. The character's pose is entirely generated by ControlNet. The background is generated separately and photoshop with character. Then Img 2 img.

Today, the image down here I'll trying to upscale and edit. This is upscale done. Add left hand, fix some error in right hand Now inpaint both hand, face, feet in highheels, breasts and edit her shadow. Stupid stable diffuision!!! Still noticeable edit trail. ??? But I like the top of the foot in pantyhose, in highheel detail!!! Edit again and inpaint again untill I satisfied. Time goes fly~~~(?) Hm....better left hand before? Hard to pick....There are pros and cons, both. In the meantime, flowers in down left inpaint. Still something missing. When do i have quite good quality to my intention?lol

Trying to change photo style to painting style, like oil painting or water color painting...and etc Source image is Generated 23.07.21  Check Point model is  photon_v1 Using i2i Now check point is  meichidark V3.8 add Lora is  Painting Failed.  Changing denoising 0.5 Failed. Fix prompt original to " beautiful form of chaos,  impasto painting, brutalist designed,  vivid color, (contemporary_art, watercolor:1.4), " Changing denoising 0.6 Hm....Something missing. changing lora  MZ Watercolor Style Hm...still....maybe need inpaint face? ? ??? Way worse. Another lora concept is crayon(Deleted on civitai by owner) ? ?? ??? Since I'm already here (crazy smile:1.4), face of girl feat samaritan   Every image has exif info

CheckPoint : MH v2 Inpaint test with this image Make her wear pantyhose. Image to image (i2i), change checkpoint to  ProtoVision XL  Using Lora :  <lora:seamless pantyhose LoHa:1> This in not final upscale result but pretty good.  P.S Upscale in  ProtoVision XL  Juggernaut XL Need adjust setting value. Seems very bad quality. Upscale done (temporary) Texture of pantyhose is low quality. Until now generate about 2200 files.... Maybe this is not the way to go. More try. Final This is the best now.