Just came here to say 'thank you' given this is the tutorial that got me going on my comfyui journey and led to the opening of my visual communication studios. It 'sparked' the flame of curiosity if you will! Thank you Fox!
Great article, what is not clear is why "Load checkpoint" node should be linked with CLIP nodes. FYI ComfyUI won't load your setup with the astronaut picture, probably because has been transcoded to jpg. Original .png file should be used.
I added a direct link to the image so that you can download it without Substack messing the format, good catch. I'll do the same for future posts in the series.
As for CLIP nodes, I added a few clarifying lines to the text but load checkpoint loads all three parts of the model: text encoder, UNET, and VAE. The text encoder in Stable DIffusion is using CLIP based models and that's why in this UI it is referred as such. So after loading it from the checkpoint, we connect it to the prompt encoding node so that it will use this clip based model to turn your text into the format that can be fed to the model
Thank you for your great article! It's always a pleasure to read your quality content.
I can't wait to see the process for a future sdxl fine-tuning.
thanks for the tutorial and the always useful and thoughtful content you publish!
Just came here to say 'thank you' given this is the tutorial that got me going on my comfyui journey and led to the opening of my visual communication studios. It 'sparked' the flame of curiosity if you will! Thank you Fox!
Thank you! With your help I might understand! Could you set it to save the pictures to a different drive on the PC, in a different directory?
Great article, what is not clear is why "Load checkpoint" node should be linked with CLIP nodes. FYI ComfyUI won't load your setup with the astronaut picture, probably because has been transcoded to jpg. Original .png file should be used.
thanks a lot!
I added a direct link to the image so that you can download it without Substack messing the format, good catch. I'll do the same for future posts in the series.
As for CLIP nodes, I added a few clarifying lines to the text but load checkpoint loads all three parts of the model: text encoder, UNET, and VAE. The text encoder in Stable DIffusion is using CLIP based models and that's why in this UI it is referred as such. So after loading it from the checkpoint, we connect it to the prompt encoding node so that it will use this clip based model to turn your text into the format that can be fed to the model
Thanks for your answer, is probably the more pedagogic article I found on stable difusion.