Description
I made a different base model from V1.0, and in this update I made checkpoints that make it easier to combine several Loras such as Lora characters, poses and clothing. So at this checkpoint we can maximize the Lora that we use for our design. Good Luck and Have Fun
Usage Guide
(highly recommended) Use a Lora for best results
For face, I use https://civitai.com/models/26124/koreandolllikeness-v20 with weight 0.5
Use character lora to produce maximum character detail
(highly recommended) Upscaling at 2x using 4xUltrasharp works very well (with denoising 0.35-0.45 will work fine). Additionally, generating the pre-upscaled image at 512x768 .
(recommended) Use DPM++ 2M Karras, 20-30 sampling step, with Clip Skip 2
(recommended) use prompts: "Hyper realistic" and "Realistic" for realistic results
This merge is very easy to prompt, and does not require a ton of prompt engineering to get good results, but if you want maximum results, use the prompt from following format will yield maximum results:
Prompt:
((masterpiece)), (best quality),intricate design, (insanely detailed:1.4), (extremely fine details:1.35), hyper-detailed, (detailed),
Negative:
easynegative, (worst quality, low quality:1.4), bad-hands-5, bad-image-v2-39000, (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), bad ???????, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, watermark,
