File size: 787 Bytes

6d985ca
 
 
 
 
 
 
 
dcb79f6
6d985ca

Inference code: https://github.com/SwayStar123/microdiffusion/blob/39b7046ef09c8bacef7b83133c671ca6ed79e44c/test_model.ipynb
Use no_cfg/microdit_model_epoch_19.pt seems to be the best

example images


![image/png](https://cdn-uploads.huggingface.co/production/uploads/6310dabee87051f3e3e4d061/41Br4qrZBUid9DxBtPVQl.png)

![image/png](https://cdn-uploads.huggingface.co/production/uploads/6310dabee87051f3e3e4d061/kSokAQhyM3rjeDILL7Vzd.png)

![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/6310dabee87051f3e3e4d061/MJydBx8dbcx86AFn0p_U3.jpeg)

It can also do different aspect ratios by default (just make sure the latents size is approx 32*32) 

![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/6310dabee87051f3e3e4d061/AAscvKtgXlTinRtMU6WEr.jpeg)