I used a large number of Flux-generated images to create this upgrade file. It makes generating Asian faces easier, reduces the rate of distorted hands, and delivers sharper focus.

this is for all friends who, like me, love realistic Illustrious.

Try it online: Tensor | Seaart | Liblib

Hope you all have fun with it!

Parameters:Steps: 20-40, CFG scale: 4-7

Samplers: Euler a/dpmpp_sde/dpmpp2m_sde

Schedulers: karras/sgm_uniform/exponential/beat

VAE is pre-integrated.

My go-to settings: 25 steps, CFG 6, dpmpp_sde sampler, beat scheduler, 832x1536 resolution.

For the record, I’ve always trained and updated this checkpoint independently, without merging it with any other checkpoints. Please note: reselling this model or selling merged versions of it is strictly prohibited.

The base model used in the merging process is xRicaMix_ILXL_v5.2. For facial restoration, please switch to XLVAE.