FP8 - All-in-one model here: https://civarchive.com/models/671478 (but, try GGUF first!)
[Note: Unzip the download to get the GGUF. Civit doesn't support it natively, hence this workaround]
A merge of FluxUnchained and FastFlux - converted to GGUF. As a result, it can now generate artistic NSFW images in 4-8 steps while consuming very low VRAM. The Q_4_0 model consumes around 6.5 GB VRAM and takes around 1.5 min to generate a 1024x1024 image with 8 steps. [See https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050 to learn more about Forge UI GGUF support and also where to download the VAE, clip_l and t5xxl models.]
You can also combine it with other LoRAs to get the effect you want.
Which model should I download?
[Current situation: Using the updated Forge UI and Comfy UI (GGUF node) I can run Q8_0 on my 11GB 1080ti.]
Download the one that fits in your VRAM. The additional inference cost is quite small if the model fits in the GPU. Size order is Q4_0 < Q4_1 < Q5_0 < Q5_1 < Q8_0.
Q4_0 and Q4_1 should fit in 8 GB VRAM
Q5_0 and Q5_1 should fit in 11 GB VRAM
Q8_0 if you have more!
Note: With CPU offloading, you will be able to run a model even if doesn't fit in your VRAM.
LoRA usage tips
The model seems to work pretty well with LoRAs (tested in Comfy). But you might need to increase the number of steps a little (8-10).
All license terms associated with Flux.1 Dev and Schnell apply.
Description
Almost identical to unquantized bf16
FAQ
Comments (25)
A little bit longer to load the model than the schnell, but way way faster after already loaded. Great for batches
I love the results but it takes AGES for some reason. I'm on an RTX-4080, the standard Flux-Dev (not schnell) model will generate an image in about 12 seconds, without changing any settings if I switch to this model it takes about 15 MINUTES! Do you have any idea what's going on? I'm using the FP8 version of T5, not FP16.
Which UI are you using? Which quant are you using? Is your VRAM 16 Gigs?
Also, have you tried restarting your UI? Some other people have solved their issues by simply doing just that.
@nakif0968 I'm using Forge through StabilityMatrix. Is quant the version of the file here? If so I'm using Q8_0_v1. Yes I have 16GB of VRAM like all RTX-4080's. I have tried restarting everything as well.
@chollman82141 Okay, I can't tell what's going on. You can try other quants. Also compare with other GGUF models from other creators. Also, if everything fails, try the fp8 version https://civitai.com/models/671478
@nakif0968 Thanks I'll do that. I appreciate the work, the results were great, just something messed up on my end.
@chollman82141 Try updating your UI. A lot of kinks were ironed out recently.
@nakif0968 Hey I actually figured it out, I had to update my GPU drivers so I could disable the "CUDA sysmem fallback policy"... went from taking 10+ minutes to about 30 seconds
Do realistic character lora's trained on Flux Dev work on this checkpoint?
Yes. Just verified- https://civitai.com/posts/5808915
Hello what is the différence Q5_0_v1 and Q5_1-v1 ?
Q5_1 is a more precise quant, but it also consumes a bit more memory
@nakif0968 thank you
I use Q5_0_v1, is perfect for me,
Dumb question, but where do you put GGUF files in Automatic1111?
In Forge UI in the Stable Diffusion directory : https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050
my question looks stupid but which from this has the best quality -realistic- (including your other s merges) since everything is chaos and nothing guarantee .. ?
Read the description, please.
@nakif0968 when you expect you will drop an update of this model here in this page (unchained merge with s) that assume will be above the quality of this by far ?
Yes
@nakif0968 when ? your expectation that will be next week ?
So what is the difference between 4_1 and 4_0?
@Missty Asked the same question but with the Q5_0_v1 & Q5_1_v1 models and the answer they got was. "Q5_1 is a more precise quant, but it also consumes a bit more memory" (@nakif0968)
so why i cant find v2 now
where do i put this in comfy ui, i placed in the unet folder and it kept crashing my pc, i am running this on a gtx1080ti legacy edition