ComfyUI for Stable Diffusion Tutorial (Basics, SDXL & Refiner Workflows) 

Thank you!

Thank you! Your comment made my day.

Wow, thanks!

Hi, SDXL Clip Conditioning (prompts) are technically divided into Clip G and L. The G indicates Guidance and L indicates latent. As per some testing done, the results are best when the same prompt is put in both G and L. Normal Clip Conditioning do not have this separation. For base SDXL, sdxl clip is always recommended. I have not come across test results for third party fine tuned models. Here is the result for base sdxl: https : //docs .google .com /document/d/144xqK_QSIbP-yJxQsztgRfm5S80rpK-UUrz0mqLy5gg/edit (without spaces)

Thanks for the answer! I will try it out! Cheers

Color correct is there, but on a widen topic for recolor controlnet. I don’t have Anu tutorials on sharpen.

No, not for this video.

Got it do with the overall hardware. 4090 24gb vram with amd 7590x3d, 64 gb ram. 7.5 gbps nvme.

What is the error ?

Same folder as sdxl one, models/checkpoint

Hi, Thank you for the feedback. In fact, I have got rid of the browser tab in the latest video, which is about area composition (will go live in 2 hours). VAE Decode is explained here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-hTSt0yCY-oE.htmlm40s (at 12min 40 sec approx), but I will explain in short: VAE: Variational Auto Encoders. VAE Encode is responsible for taking data from Pixel Space (image), encoding it in compressed format, and converting it to Latent Space (noise). VAE Decode is responsible for the exact opposite. It takes this noise along with your text prompts and converts the latent noise to pixel space in a reconstructed way (image). set return_with_leftover_noise is set to disabled, which means the sampler does not have any noise from the previous sampler to work with. If you enable it, the second sampler will have some remaining noise to work and build on the image. Many times you may get some messy details (when enabled). But for power users, its advisable to play around with the option enabled or disabled. The final result depends on your workflow. If you have any further questions let me know.

I agree with the OP. The video script for the video itself is excellent, about 33:09 the speeding up the video makes it very hard for a beginner to follow. I found my pausing the video and stepping back through it frame by frame sometime to pick up some little detail that flew by at warp 9. We can skip past parts that we already know about with out the video being sped up. Also the habit of renaming the nodes, while it might be convenient for a established user, is very confusing for a beginner. I'm trying to match the work flow with what I see and all the node names do not match. I have to go back through the video small sections at a time to find out what that node was named. I abandoned the video at 33 minutes in simply because with all the node renaming and fast warding it simply was not worth my time any more to go back and figure out what was missing or what I did wrong. On more bit of advice. Try not to have so many models in your menu so they don't look like a huge mess. Only put the videos in your model menu, and anywhere else you need for that video When I was trying to match model names on Civitai and what you had there, I also had to step through the video one frame at a time to find the model name. Sorry to be such a critic but these are the issues i found frustrating

​Thank you for the feedback. Will take your suggestions into consideration, maybe do a clean sdxl tutorial again. If you require any assistance feel free to ask, would be happy to help.​@@jeffhyche9839

I agree. That was my early days.

Don't have any workflow for this video. Am going to re do the video tutorial as a basic series. So stay tuned.

@@controlaltai just share this image 47:28

Don’t have it….thats what I mean there is no workflow.

@@controlaltai oky sorry :/

This is very easy to replicate. Let me whoop up one right now and link it here…..I will just make a basic sdxl refiner workflow, give me few hours.