Since this is a dedistilled version of "Flux schnell", Apache 2.0 license applies making it free to use commercially.
This fine tuned checkpoint is based on Flux schnell de-distilled thus requires a special comfyUI workflow and won't work very well with standard Flux dev workflows since it's uisng real CFG.
This checkpoint has been trained on high resolution images that have been processed to enable the fine-tune to train on every single detail of the original image, thus working around the 1024x1204 limitation, enabling the model to produce very fine details during tiled upscales that can hold up even in 32K upscales. The result, extremely detailed and realistic skin and overall realism at an unprecedented scale.
This first alpha version has been trained on male subjects only but elements like skin details will likely partically carry over though not confirmed.
Training for female subjects happening as we speak.
Highest quality slow version:
Turbo version with shamelessly good quality!!!:
Resources:
Lora:
Fast Lora: https://huggingface.co/ostris/OpenFLUX.1/blob/main/openflux1-v0.1.0-fast-lora.safetensors
Turbo Lora: https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha
Text encoders:
ViT-L-14-BEST-smooth-GmP-ft:
t5-v1_1-xxl:
https://huggingface.co/city96/t5-v1_1-xxl-encoder-bf16/resolve/main/model.safetensors?download=true
Clip G:
Recommended settings:
Without Turbo and Fast Lora
CFG: 2-4
Steps: 50-60
With Turbo and Fast Lora
CFG: 2-4
Steps: 8-16
Turbo Lora weight: 1
Fast Lora weight: 0.33
This Flux variant works with negative prompts so use them if you want to. Negative prompt will work with and without the Turbo Lora.
Description
Alpha version, trained on men only in this version. Women to follow shortly.
FAQ
Comments (29)
hey, pls some more hints :)
sampler, scheduler?
cfg 1,2,3,4 ???
steps ?
theres an included comfyui workflow with all the settings its made for along with the models and clip used. it works well with those settings for me. the creator says the model is intended for this workflow.
Yes since it’s Flux dev de-distilled, common Flux workflows won’t work well. It needs real CFG basically instead of the flux guidance scale.
I am gonna compare with base de-distilled model lets see improvements thanks.
Any improvements?
Will Img2img work?
I tried plugging in a realistic image in to the upscale step, and yeah it works. It definitely diluted some anatomical accuracy though.
Of course!
wow thanks for sharing some of the images i have generated with this are almost indistinguishable from real life.
Such a great model. I'm basically clueless about his this works. Would you consider in the future a short explanation?
Thank you! Yes I will record a tutorial soon once I’ve nailed down the formula properly.
hook us up with a Q8 quant, bro
Nice model. I wish it worked with character loras but other than that, it's good :)
Thanks. It does work with regular Flux dev Lora’s. Give it a try.
@tamtamx1332 i did try...doesn't work. the results are very grainy.
@ObsidianDreams are you using my worklfow? This is not a regular Flux model. You can't use it with a regular Flux comfyui workflow.
@tamtamx1332 yes, i'm using your workflow. works good normally, but if i use any character lora, immediately the results are grainy, noisy, and poor. I'm sure it's something I'm doing wrong, but i didn't change any of the settings in your workflow, just added my character lora.
@ObsidianDreams Hmmm yeah works pretty good on my side.
Nice approach to training. I've been working on something similar, but without the ultra high resolution aspect. Good to see someone else is actually experimenting with FLUX and not just whacking it with a proverbial hammer.
Tried in Forge om Mac M3: it works awfully. Someone has a clue?
Make sure to have CFG at around 3.5. This is a Flux de-distilled model so it requires real CFG not the standard flux guidance scale. Without the turbo and fast Lora need to have the steps around 50. With the turbo and fast Lora you can go as low as 8 steps.
@tamtamx1332 thank you very much for your help!
Really great skin detail. Works for women too.
Looks spectacular. I'm waiting with baited breath for the training to complete.
Ditto! The slow workflow is also something else, very interesting with very find details. Just like a good burger, sometimes it's worth the wait.
@pychobj2001741 the turbo version looks pretty close to the slow one but so much faster. 13 steps instead of 60.
female fashion / studio photography model pls?
Any chance to get GGUF models?
They are available now.