Welcome to version 1.2 it is just as fast, but the colors are better, the details are finer and the contrast is better, the overall image is just better.
For detailed portraits as seen in the gallery, use the Flux.1 S Workflow with Lightning upscale and detailing.
!!! There is a better workflow for refining skin, it's called the Flux Refiner workflow and it is also based on SDXL Lightning and works best with the model found in the description !!!
I have a 4070 12GB and with CFG 1 and 4 Steps a batch of 4 images takes less than a minute. It's based on Flux.1 S, also called Schnell, so it's free, easy to use and fast!
Best use is CFG 1 and around 4-6 steps, but feel free to experiment.
It's 100% Flux.1 S based, so the Dev license does not apply on generated images.
Description
First release
FAQ
Comments (28)
this is cool, but have you found any way of making it NOT burn the image? i feel like images that schnell generates are too saturated and usually have a very high contrast.
Not sure just working with it for like 6 hours
if you want really burning and saturated images try my new LoRA lol
https://civitai.com/models/640926/jaynlartify
in comfyUI you can use flux guidance and decrease it lower tan 3.5 or use ModelSamplingFlux with shift values under 1
but does it do only 4steps ? :P
If you use Euler simple, CFG 1 and 2-4 steps a batch of 4 (1024x1024 or equivalent) should take less than 1 minute 💕
what kind of merge is this?
t5xxl fp16 baked version would be nice for us 24gb vram folks :P
Merged with my own LoRA
I would like to BTW, but my 4070 takes ages with that version
@JayNL then u need nf4 :) https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/981
Who's going to really cater for you 24Gb guys? You gotta realize that most of us can't even run fp16 versions. If you already have 24Gb, then just use the best version available.
Nobody needs anything with t5xxl baked in. How many times do you want to download the exact same 10 gigabytes?! Unet only MUST become the new standard, sooner rather than later...
@Oracle3910 ye found that out as well by now 🙈
@tazztone I barely use my own checkpoint anymore, GGUF is da way! 👌
@tazztone my new CCA version is fun though, and 4 steps is still as fast as GGUF
https://civitai.com/models/674433/artfusion-flux-cca-4-steps
Thanks Bro! It works for me in forge on rtx 3060 12gb, + lora realism = generation Time taken: 48.1 sec.
Yeah right, it's just really fast, with nice results 💕
tell me please how it takes using your same h.w and same steps count but with sdxl ? let me compare this
@amazingbeauty you can try one of my Artfusion XL models (real or surreal are good), check my page, then use CFG 2 and Steps 6
Made the frontpage with this model! 💕
And again on the frontpage, now with my Flux Latte LoRA! 💩
I'm gonna change the name of this model to get in line with my other models and I hope to come with a new version soon! 🙏
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.