If you're having trouble running Flux on your PC, this is likely the solution you need. GGUF models are smaller, use less VRAM, and are faster! You can find all download and installation instructions in the workflow notes. Use ComfyUI Manager to install any missing nodes, and enjoy!
Description
-First Release
FAQ
Comments (9)
Thank you very much for this. I too only have 12GB card and this flow is something I can largely understand as I don't use Comfy all that often.
I'm not too sure what you mean in your comment about the Ksampler and "connect to the output reroute", but the Ksampler Simple is working well enough for me to have a play.
I'm glad you could use it! About the "connect to the output reroute", it's not hard. In each Ksampler group you can see an "appendage" leaving the group. In the original way the Image output connects into the reroute, that connects to Preview node and save node. If you turn on the dual Ksampler group, and connect the output of it in this reroue, you will have a slight different result. Not better, not worst. Just different. But to be honest, it doesn't really matter so much. As you said, the simple Ksampler does the job! Hehe.
@Sam_A Thanks for the explanation. I see it now. Takes a couple minutes for me to generate an image with a Lora or 2 which is much better than other workflows I tried that took 10 minutes or more. Part of the reason could be they had other 'features' that I didn't really care about (eg: upscaling) and wasn't sure how to disable them without breaking things.
I do have another question about the workflow. What's the difference between the SDXL Prompt Styler and the Positive and Negative text boxes. Out of curiosity I entered text in the Positive and Negative boxes and left the SDXL Prompt Styler blank. When I ran the workflow, the image was nothing like I prompted. It was as if the Positive and Negative text boxes were ignored - though they do appear to be connected to other nodes. Is setting the SDXL Prompt Styler to 'base' and entering prompts the same as entering them in the other boxes?
@jonk999 You have a a bool box at the left of the prompt boxes (above the lora node) to choose between simple Positive/Negative prompt box or SDXL Prompt Styler. SO you can use one or another. Never both at same time. The SDXL prompt styler is a "lazy" stuff I like to use sometimes. You can just describe your subject. And in the SDXL Prompt Styler Node there is a "style" option that will add text to your prompt. It save a lot of time thinking and writing stuff. You can see the final Prompt output at the right of the SDXL Styler node for positive and negative prompts.
@Sam_A I completely missed that Bool box! Apologies for not looking at the flow properly. I follow it now. Thanks very much for taking the time to answer my beginner questions!
@jonk999 It's alright bro! Workflows are hard at first sight. Even more when you're not used to Comfy. But once you get used to Comfy, it's hard to go back!
@Sam_A Yes! I know I really should learn and get my head around Comfy more as it's so powerful, but I'm just so used to the simplicity of Auto1111/Forge! I have a look at some Comfy workflows and I just have no idea how they even know where to start and what nodes to select. Some I'll try and figure out what they are doing, what models are used and where they should be stored, etc, but sometimes I just give up and look for a simpler one or go back to what I'm used to. So much respect! :-)
@jonk999 Take a look at SwarmUI, it uses ComfyUI as backend but make things much easier.