phalexo

if_I = IFStageI('IF-I-XL-v1.0', device='cuda:1') if_II = IFStageII('IF-II-L-v1.0', device='cuda:2') if_III = StableStageIII('stable-diffusion-x4-upscaler', device='cuda:2') t5 = T5Embedder(device='cuda:0', torch_dtype=torch.float16)

Much better. The aspect ratio was affecting quality. 



Mine are overclocked and get about 9Tflops. That said, I am comparing Floyd with SD V1 running on the same GPU, at the same clock rate. I got some really...

Wouldn't you have to retrain text to image models because the representations are different?

I have found one working combination. If set count = 1, and t5 = T5Embedder(device='cpu'), it generates at least one pic, I only run the if_I.show(result['III'], size=14). Perhaps a blas...

I can see only a single call to this function. Should that call be commented out?

> There are 3, one per stage. There is an embedded comment that says to remove this line Hm. I grepped for it, saw only one. Do you know which...

> No, to try it I have to provide my info to download the checkpoints. If the above is not possible with the current code, then I don't want to...