This is a large model based on the official SDXL-base-1.0-Cloud platform rewash three times concept-full volume stylized base——

Training over 2,100,000 steps +!

Training time over 200 hours +!

Repeated training washing 3 times-concepts over!

✨Model Highlights:

✅Better aesthetics: the atlases are manually collected and screened, rejecting miscellaneous or homogenized atlases

✅Better response: SDXL full natural language annotation is used, the model is more obedient.

✅Better base: full-volume three times BD training, without any fusion traces, more suitable as a base model for Lora training & outgrowth, supports better learning rate fitting!

✨Parameter recommendations

👉Comfyui
・CFG: 2-3

・Size: 1024*1024 or above

・Steps: 20-30

・Already built-in VAE, no need for external VAE

・Recommended sampler: dpmpp_2m_sde_gpu-above version

・Recommended scheduler: simple

・HD fix: no need to turn it on

👉Webui
・CFG: 4-5

・Size: 1024*1024 or above

・Steps: 20-30

・Already built-in VAE, no need for external VAE

・Recommended sampler: DPM++ 2M

・Recommended scheduler: SGM Uniform

・HD fix: recommendedo peninga mplitude0 .3,m agnification1 .5t imes