Hello, would it be possible for you to share the unpruned version? Even if it's on HF, I've been searching in the repositories but I can't find it (or I just don't know where to look haha). Thank you so much

Is this the turbo model of 4.0? When using sde karras 10step cfg 2, the image is much unfinished and blurry compared to 3.0 turbo.

Many thanks! I’ve been following your work since the beginning. It’s my favorite model

Hi, when using the V4 as main checkpoint it generates only disturbed and confused pixels....

What is the optimal first pass resolution for RealVis XL v3.0 Turbo?

Even though Base SDXL Turbo is derived of Base XL, it prefers a different standard resolution according to the documentation of stability AI (512x512, check links below). I wonder how this affects the optimal resolution of Hybrid Turbo merges such as RealVis XL v3.0 Turbo?

The reason I am asking is that if I use first pass resolutions similar to those for standard XL models (1024x1024 or 896x1152), I get quite a high proportion of images with distorted bodies.

References:

Source: https://huggingface.co/stabilityai/sdxl-turbo
Stability AI Base XL Turbo Model Card:
"Preferably, the model generates images of size 512x512"

Source: https://huggingface.co/docs/diffusers/en/using-diffusers/sdxl_turbo
"By default, SDXL Turbo generates a 512x512 image, and that resolution gives the best results. You can try setting the height and width parameters to 768x768 or 1024x1024, but you should expect quality degradations when doing so.

Make sure to set guidance_scale to 0.0 to disable, as the model was trained without it. A single inference step is enough to generate high quality images. Increasing the number of steps to 2, 3 or 4 should improve image quality."

Thank you!
And please also upload the inpainting version

Thanks for your amazing work!

Do you plan to publish a v4.0 Turbo version in the future?

one of the best models for SDXL turbo hands down fast generations at great quality way better then other models will have to check out v4 soon!

I really appreciate your continued updates on this, and I especially like that your negatives are getting shorter and shorter. You'd be surprised how much constriction a long negative has on a good prompt.

Why does this often put weird patches like a sticker, or a blob of paint or piece of paper on the person? Often they are orange or red.

Your model structure Diagram is not updated, right ?
Because why merging 20% of Jugg V5 when the V8 is out (and a lot better)

Im guessing you must likely have re-trained your latest V4 with all the latest version of the other models your merging ?

I absolutely love this model, it performs beyond imagination.

inpainting model fails to load. Am I doing something wrong?

the best model!!!!

Do you have any plans about
SDXL-Lightning ?

This version is one of the best XL models I've seen at making unusual/non-doll-like faces. Great job.

This is an excellent model for photorealism.

when i use the model my images have a lot of images noices. Any ideas to fix this?

I use the recommended settings but still...