Can I run the StableSR without CFW?

[Sunlight-zero]

Thanks for your nice work! Now my super-resolution model is trained successfully. However, in my case, I don't need the 2nd stage CFW. I wonder if I could use the fine-tuned StableSR 1st stage model, and the vanilla Encoder-Decoder (VQGAN used in Stable Diffusion, without CFW part), to run the StableSR model?

[IceClear]

Hi. Yes. you can run it w/o CFW.

[Sunlight-zero]

Hi. Yes. you can run it w/o CFW.

Is there any script to run it without CFW? I only find the script with --vqgan_ckpt argument. I don't know how to run the model without CFW.

[IceClear]

If you would like to try the result w/o CFW, you can set --dec_w to 0. But the CFW module is not deleted and still leads to memory cost. You have to delete it manually if you would like to remove it totally. You can also use WebUI directly which does not adopt CFW.