SD3 medium model Generate 512x512 image

huggingface / swift-coreml-diffusers

Swift app demonstrating Core ML Stable Diffusion

Apache License 2.0

2.45k stars 200 forks source link

SD3 medium model Generate 512x512 image #95

Open pandora523 opened 3 weeks ago

pandora523 commented 3 weeks ago

When trying to generate image from diffusers I just notice the image is kinda flop and idid try saved the image and the resolution is 512x512

Labrador_in_the_style_of_Vermeer

the same thing I reported on hugging face respiratory https://github.com/huggingface/diffusers/issues/8502

ZachNagengast commented 1 week ago

What do you mean by "flop"? If you mean bad quality this is definitely a common complaint for this model. You can try reducing the guidance scale to 4.5, which is recommended, but there are fundamental issues with the model itself as well.