search

IDKiro / sdxs

Official repo of our paper "SDXS: Real-Time One-Step Latent Diffusion Models with Image Conditions"
Apache License 2.0
558 stars 20 forks source link

cannot get comparable results as original DreamShaper model #20

Open whhamber opened 1 month ago

whhamber commented 1 month ago

Thanks for releasing the model. Really interested in your work. However, using the same prompt, I CANNOT get comparable results as original DreamShaper model. Here are the examples:

sdxs-512-dreamshaper / large vae / fp16 image

dreamshaper / 4 steps / cfg 3 res-6-1421173560-4

dreamshaper / 8 steps / cfg 3 res-5-8530355218

Can you help me with this problem? Maybe should I change more parameters during inference?

IDKiro commented 1 month ago

I apologize for taking so long to respond.

Because SDXS is a highly distilled miniaturized model, it can only generate lower-quality images than the original model, so this phenomenon is normal. If you are interested in training a DreamShaper-based distillation model with higher quality, I recommend DMD2, which contains the training code.