This model was not trained using any images of celebrities. Please do not generate any images of celebrities using this model.
Steps: 8-20, CFG scale: 0.7-1.5, Sampler: LCM
Scheduler: simle/beat/karras/sgm_uniform/exponential
My usual settings: 8, 1, LCM, simle, 1024x1280
VAE integrated.
Tensor uses lcm sampler, which can only use up to 8 steps...
Realistic style, using a lot of Flux-generated image training, making it easier to generate Asian beauties. Other content is the same as the base model, with the exception of the lack of action and anime, but it can do almost everything. I've worked hard on the fingers, and I'll focus on improving them in the next update. DMD2 acceleration has been added, so I can only use the LCM sampler.
Use a resolution larger than 1024x1024 for the original image, as this will affect hand stabilization. For resolutions above 1280, use an upscaler.
One final piece of advice: Please dress her in more clothes, because I want to see your work.🫠