Fluenlty XL - the best XL-model
Introducing Fluently XL, you are probably ready to argue with the name of the model: “The best XL-model”, but now I will prove to you why it is true.
After the release of the Final version there will be no more updates? - Updates will be made as soon as they become available to improve.
About this model
The model was obtained through training on expensive graphics accelerators, a lot of work was done, now we will show why this XL model is better than others.
Features
Correct anatomy
Art and realism in one
Controling contrast
Great nature
Great faces without AfterDetailer
More info
Why do we say our model is the best? Simply put, most checkpoints are mixes of others; ours is trained, not mixed. If you have challenges at first, it may be because mixed models often respond the same way to the same prompts; freshly trained models have a learning curve. Feedback from amateurs and professionals alike tell us that you WILL love this model once you become accustomed to using it!
Using
Optimal parameters in Automatic1111/ComfyUI:
Sampling steps: 20-35
Sampler method: Euler a/Euler
CFG Scale: 4-6.5
End
Let's remove models that copy each other from the top and put one that is actually developing, thank you)
Description
Optimal setting (only for Lightning):
Sampler: DPM++ SDE
Steps: 4-6
CFG Scale: 1.5-2
FAQ
Comments (3)
Playing with it. Thanks.
If you are into training LoRAs - I highly recommend the V2 and V3 models.
I personally don't think this is that good of a model for generating (My favourite this far is Atrium 2.0) - maybe I'm doing something wrong with prompting for this. I get very basic and clean generations (but between v2 and v3 I think v2 generates the best). HOWEVER! This is absolute gem for training LoRAs on. I have no idea why, but V2 and V3 are the best at training with of all models I have tried. They make very clean results. I suspect that this is just down to the fact the model is fairly clean overall - lacking in random crap. And the LoRAs seem to work in nearly all models (which aren't a merged beyond reason) extremely well.
If you are in to training LoRAs (which I do as I like the puzzle of doing it) I highly recommend giving this model a try specifically V2, V3 is bit less responsive but cleaner. The resulting LoRA models are quite flexible and work in nearly all other models without issues or glitches unlike with many others.
So... Keep doing whatever it is you are doing? Training in a clean manner?
Overly impressed with this model, I've thrown everything at it including the kitchen sink and it has not failed to impress me, good job