Posts

Trying to make a new process

Image
 The previous process was giving unsatisfactory results, so we tried a new process. Base image (model : perfectrsb v6321 ) Before upscale, I2I Inpaint first with SD 1.5 (model is : MH v2 ). Upscale 1.4x 3 times and 1.2x once . Applyed detail deamon and ddsd extension The hair looks over-sharpened, but the other parts seem fine. Inpaint again  The hair texture still appears over-sharpened, while the pantyhose texture feels slightly lacking, but overall, it looks fine. Remove blemishes. Add shadow noise, skin texture, sharpen, color and lighting contrast. Final edited result. In this case, new process has much better result. The texture details could probably be improved by using Flux with ComfyUI for inpainting, leading to better results.   All image has exif info. You can find me on  instagram  ,  deviantart   p.s This base image to this

Cultist nun

Image
Got some pretty cool results today! Lora random combination : {2-3$$<lora:cultistexl:__weight__>|<lora:unholy:__weight__>| <lora:ph4nt0m01lXLP:__weight__> |<lora:hinaMixArmor_pony-rev:__weight__>| <lora:Aura_r1:__weight__> } ILXL check point :  Uncanny valley Pony lora  :  <lora:unholy:__weight__>    <lora:cultistexl:__weight__>    <lora:hinaMixArmor_pony-rev:__weight__> Those shinny texture and mysterious mood~ And cute!? Is there anyone who would worship with me? All image has exif info. You can find me on instagram , deviantart  

Check point Model Juggling

Image
As I frequently switch between different models to generate images, I occasionally make mistakes by forgetting to adjust some settings after changing models. For example, when generating images with the Pony model, I use the Pony hand LoRA for inpainting hands. However, after switching to the ilxl model, I sometimes mistakenly continue applying the Pony hand LoRA without making adjustments. Ilxl model ( neonCityBlend_ncbIllustriousXL ).  If you look closely at the hands, you can see subtle differences in the art style. (The checkpoint model is for 2D, but the hands appear semi-realistic) This was created with the Pony Hand LoRA ({lora:pony_good_hands:0.9|lora:detailed_hands:0.9}) applied.  (Since it was set up using a wildcard, it's unclear which one was actually used.) So here is my theory For semi-realistic images generated using Stable Diffusion 1.5, where the depiction of hands is unsatisfactory, wouldn't it be effective to improve the results by inpainting with the ilxl c...

Today's small discovery

Image
 I understood that pony LoRA could not be applied because the structures of the ilxl and Pony models were different. However, today's result was different. lora : she venom Check Point Model : 2dn As you can see, the texture is different. PS And I love she venom!!!

Generate chibi bulleta

Image
Check point Model :  novaUnrealXL_v2 Not a perfect but still pretty fun~ EXIF included

The Future of Style: A Cybernetic Queen in Her Golden Chariot

Image
 🚨🚗 Introducing the Cybernetic Luxe Ultra-Roadster 3000! 🚗🤖 Why settle for ordinary when you can have extraordinary ? ✨ Featuring the ultimate fusion of beauty, brains, and bionics ! 💛 Comes with gold-plated armor for unmatched elegance! 🖤 Wrapped in sleek black tech-skin to resist scratches and haters alike! 🔥 Includes a self-charging AI system —because who needs gas? 👠 Bonus: Built-in stiletto boosters for instant glam-on-the-go ! 💥 Ride like royalty! Lounge like a legend! All while looking like you just stepped out of the future! 📞  CALL NOW  to secure your limited-edition Cyber Luxe package! Operators are standing by… or are they cyborgs? 🤔

Today's simple(?) discovery

Image
 It suddenly felt like the generation speed had slowed down for some reason today. ???? It was taking about 8 minutes for 4 images! !?!? From what I remember, with my settings (euler a, 16steps, 1024 * 1280 resolution), generating using the ilxl or pony model used to take around just over 4 minutes for 4 images. I was testing the 2dn_illustV1 model and wondered if there was an issue with the model, so I tried switching to a different one. The result was the same, with the generation speed taking about 7 to 8 minutes. While retracing the settings I changed around yesterday or the day before and reverting them one by one, I was fortunate to quickly identify the cause on my very first attempt. How relieved! The cause was the 'Use separate width/height' option in the after detailer settings.  Since the day before yesterday, I had been generating with the default setting turned off (checkbox unchecked).  After confirming there were no issues with the generated results, I kept...