Discord to try the model with a bot. LCM-Lora available for lightning fast generation.
For Creative and Colorful content. Unleash your imagination ! Fork of ZavyChromaXL
Realism is not the purpose of this model.
NB : I'm not responsible for your bad/cheap prompts. 😏 cf. My Gallery
Still a very first version with many composition errors. Result may vary. 🙃
To use Highres_Fix is the way to use this model, and more generally SDXL.
Eventually Adetailer.
I don't recommend TensorRT, to maintain the maximum level of quality.
Settings:
As usual, it's subjective. SDXL works the same for everyone. It's a story of expectations.
My settings :
Prompt: Your prompt <lora:Details:0.85>
Negative prompt: BadHands, (NSFW), (nude), (naked), undetailed, poorly drawn lines, illustration, 3d render, painting, unrealistic skin, ugly teeth, ugly pupil, (worst quality), (low quality), (normal quality), lowres, signature, watermark, username, cropped
Steps: 30
Size: 768x1280 (1152x1920 after hires. All 4 values are multiple of 64)
Sampler: DPM++ 2M Karras
Lora: Details:0.85
Embeddings in negative prompt: BadHands
CFG scale: 8
- Highres_Fix:
Hires steps: 15
Hires upscale: 1.5
Hires upscaler: 4x-UltraMix_Balanced (You need any custom upscaler when using DPM++)
Denoising strength: 0.4
Cheers ! 🥂