Saved in ComfyUI using the Checkpoint Save node. Combined the Diffuser Model Flux.1 Dev, T5XXL fp16, CLIP_L and VAE in to a single file.
Description
32GB file is targeted at those with 24GB VRAM or greater
FAQ
Comments (12)
This is interesting because I think most people are leaning toward the opposite approach of wanting to remove the text encoders from the models and just download the UNETs to reduce the amount of disk space they consume.
Personally, I hate managing all the different files and nodes :D
So for me, since I use these files all the time separately anyway, I prefer not only the single Checkpoint Loader Simple node, but also the "normal" KSampler node too.
;)
@gonzalu I agree actually, and I am blessed to have the storage to manage it (at least to a point). I was just noting the trend seems to be going the other way as it increases the size of each checkpoint significantly. Once we start seeing dozens of FLUX (and other) checkpoints, it is going to be interesting to see how it progresses. (Just in case, I wish we would get a few more COMFY nodes that handle UNET better than the common one does now.) Anyway, it wasnt a criticism. I am genuinely interested to see how this shakes out.
@pAInCREAT0R Yeah agreed... smaller smaller smaller is all I see. I am very blessed to have a 4090 GPU. Hope the 5090 goes to 28GB VRAM as rumored ...
@gonzalu Yeah I have a 4090 also. Upgraded last year from a 1080TI just for AI.
@pAInCREAT0R I literally got the 4090 because of AI and second for Hashcat :D
hello could you teach us how to merge all the files together and how to separate them again? i wanna run some tests and got a cluster 4090s and id like to share a grid with every samples, dif steps to see if doing (merge / separation) afects the images and in which way also id like to build a tool to quickly do this since i belive its a great idea but we might have to do it on a per request as not to flood our storageds with 10000 copies of the vae, clip, and t5
To combine them, I simply use the ComfyUI Core Node "Save Checkpoint"
You load the original parts in their own nodes such as the UNET Diffusers Loader, Dual Clip Loader and the VAE loader. Connect them all together to the Save Checkpoint node.
The resulting checkpoint will be saved in your Outputs folder.
To split them, you can use the individual save nodes: Save Model, Save CLIP and Save VAE nodes.
Good luck.
Hi, I've never manged to get Flux working in Automatic111, would this one be any good do you know, as it looks amazing?
Thanks for your work & time :)
really hope this works man, i'm having a real hard time with flux, the results i get in forge are literal black screens
usually it is wrong sampler, what setup and UI do u use?
So sorry you're struggling. TBH, I never tested my merge in Forge :( Been sticking to ComfyUI for the past few months.
Let me know if you get it working. And as @blackrubber said, what are the settings you're using? Stick to euler / beta if you can which is what works for me most of the time.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.