Atomix XL V4.0
improved the understanding of prompt.
Atomix XL V3.0
Still a Lightning version,it is base on V2,improved the understanding of prompt and image quality.
Atomix XL V2.0
V2.0 is a Lightning version, its generation speed and image quality are much better than V1.0.
Atomix XL V1.0
I only used the SD1.5 model before,not only because the generation speed of the SDXL model is too slow, but also because the generated image are too realistic(too many film grain).In the past two days, I occasionally tried SDXL Turbo model and was surprised to find that the speed of SDXL has greatly improved, the time spent is completely acceptable. So I started trying on solving the rest problem, and I succeeded. So, this is a SDXL model that balances reality and CG well,which generate clean images(fewer film grain),and supports Turbo acceleration.
Recommended settings
cfg: 1~2
steps: 6~10
sampler: dpmpp_sde_gpu
scheduler: sgm_uniformDescription
FAQ
Comments (6)
What checkpoints did you merge to make your lightning version?
There are no changes compared to version 1.0, except for the addition of a 8step Lightning Lora. ByteDance/SDXL-Lightning at main (huggingface.co)
cfg=1.5 is perfect
first lightning one that worked for 1920x1080 size in 6 secs, nice, and at cfg scale 1, is astonishing, cheers mate, great work
Thank you~I just updated the V3,welcome to try it out.