• without hires to gen highly detailed photo

• without default asian face

• without need too much tag weight, better text understanding

• works fine with almost all LORA and TI

[update:0230825]

• version 2 (close to sdxl1.0)

Advices

1. Not require strong prompt weight adjustments. Try removing all prompt weights.

2. Not require Hi-RES. Try big steps (>50steps) and use a DPM++ series sampler.

3. Asian-style version 👉 LOFI-v4

4. Fantasy-style version 👉 EPIC-v1

[update: v1]

What raw-xs ?

• raw: Realistic Artistic Wizardry

• xs: Extra Small

(chatgpt generated)

a little better than sdxl-0.9 （XD

based on LOFI_V3 (with sdxl0.9) fine-tuning

Tips

this model may be include wrong sdxl partten 😅

(for example, in some color modes, human skin will be very oily, You can try different samplers, there will be different effects)

Backup

https://huggingface.co/lenML/raw-xs/tree/main