Closed Heathen closed 11 months ago
Yes, it only works with SDXL Base->Refiner. It uses conditioning from Base, otherwise you will have to store entire checkpoint of Refiner in memory, because conditioning from 1.5 is not compatible with Refiner, which is very expensive in terms of memory and performance.
** Error executing callback cfg_denoiser_callback for C:\AIStuff\autowebui\stable-diffusion-webui\extensions\sd-webui-refiner\scripts\refiner.py Traceback (most recent call last): File "C:\AIStuff\autowebui\stable-diffusion-webui\modules\script_callbacks.py", line 195, in cfg_denoiser_callback c.callback(params) File "C:\AIStuff\autowebui\stable-diffusion-webui\extensions\sd-webui-refiner\scripts\refiner.py", line 133, in denoiser_callback params.text_cond['vector'] = params.text_cond['vector'][:, :2560] IndexError: too many indices for tensor of dimension 3
*** Error executing callback cfg_denoised_callback for C:\AIStuff\autowebui\stable-diffusion-webui\extensions\sd-webui-refiner\scripts\refiner.py Traceback (most recent call last): File "C:\AIStuff\autowebui\stable-diffusion-webui\modules\script_callbacks.py", line 203, in cfg_denoised_callback c.callback(params) File "C:\AIStuff\autowebui\stable-diffusion-webui\extensions\sd-webui-refiner\scripts\refiner.py", line 148, in denoised_callback p.sd_model.model = self.base.to(devices.device, torch.float16) AttributeError: 'NoneType' object has no attribute 'to'
Ahh, figures you can't go from sd1.5 to sdxl with this. Only via img2img.