I found it really helpful as it forces us to understand what makes SDXL 'tick'. Note: the part on clipping could be rephrased a bit : I'm still rather confused - given that most tutorials recommend to stick 4x the size of the latent in there , in all 4 fields and leaving crop at 0. Could you perhaps explain why it's recommended by the community and how it functions 'behind the scenes'? this could be super interesting. Thank you! brillant tutorial.
thanks! we will get there. The plan is to first implement SDXL as it was released and get a good sense what it can do with the controls that it comes with and then do more targeted improvements as you suggested.
Great tutorial thanks! Could you possible create another with part 4.
"Part 4 - this may or may not happen, but we intend to add Controlnets, upscaling, LORAs, and other custom additions."
Once again thank you so much for such an amazing tutorial.
I found it really helpful as it forces us to understand what makes SDXL 'tick'. Note: the part on clipping could be rephrased a bit : I'm still rather confused - given that most tutorials recommend to stick 4x the size of the latent in there , in all 4 fields and leaving crop at 0. Could you perhaps explain why it's recommended by the community and how it functions 'behind the scenes'? this could be super interesting. Thank you! brillant tutorial.
Hey, I followed the exact steps but for some reason I get the following error.
Error occurred when executing CLIPTextEncodeSDXL:
list indices must be integers or slices, not str
File "/workspace/ComfyUI/execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "/workspace/ComfyUI/execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
File "/workspace/ComfyUI/execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "/workspace/ComfyUI/comfy_extras/nodes_clip_sdxl.py", line 43, in encode
tokens["l"] = clip.tokenize(text_l)["l"]
can you try downloading this image, dragging it on your canvas and seeing if it works?
also did the part 1 work without a problem for you?
https://huggingface.co/pxovela/sdxl_comfyui_post_series/resolve/main/Part%202/ComfyUI_temp_axjsz_00155_.png
For some reason, loading your image works perfectly but whenever I try to build it myself it fails. It's so weird haha.
For the "next topics" I suggest, improve faces and hands
thanks! we will get there. The plan is to first implement SDXL as it was released and get a good sense what it can do with the controls that it comes with and then do more targeted improvements as you suggested.