VAE is NOT include in V2 checkpoint

My first model merges with personal LoRAs and the IllustriousXL base checkpoint.

VAE is include in V1 checkpoint

I get better results with encoders like Euler or DPM2++, between 25 and 40 steps, with a CFG between 5 and 7.5. Don’t hesitate to use Dynamic Threshold for more variations. Adetailer is recommended for faces as usual, but also feel free to slightly upscale your image.