Detailed comparison between standard VAE and the UltraFlux VAE: https://imgsli.com/NDM1MTI2
This vae is not made by me, I am simply re-uploading the vae from this awesome project https://huggingface.co/Owen777/UltraFlux-v1!!!
From my testing it improves the sharpness of any model using the flux vae (flux based models, Zimage, etc).
In rare cases it might look a little over sharpened but from my short testing it looked pretty good.
Links to the project from which this vae originates from:
https://huggingface.co/Owen777/UltraFlux-v1
https://github.com/W2GenAI-Lab/UltraFlux/tree/main
I should have clarified from the start, but basically as vladulidlo has mentioned it's best to use this VAE on a second pass after upscaling. In my opinion it is still worth using in simple 1 pass workflows too (which is why I didn't bother overcomplicating stuff) but if you pixel peep and the oversharpening bothers you then use a 2 stage generation workflow for better results.
Description
This VAE works with all flux and zimage models
FAQ
Comments (10)
Can't you see that it generates strange blur artifacts? It is so bad, it even shifts an image
The only "artifacts" I have noticed is oversharpening on small details in some cases, but in general I would say it's a fair trade off for the overall improved sharpness.
Either way I'm not the creator of this VAE, I'm just sharing it because I thought it was neat.
@jorkingtoncityshallwe You can just apply an unsharp filter to any image and it will give you the same (or better) effect this VAE does. This VAE was seemingly trained on images with unsharp filter applied, you could see ringing artifacts between contrasting colors.
@breezyjelly Yeah there is a billion different ways to do anything, this is just another option.
Yo just a reminder this VAE is optional and no one is forcing you to use it lmao.
Also please don't harass the original creator on huggingface, if you don't like the VAE simply leave a bad review or comment here and move on, no need to harass people.
I believe this is wrong use case and misunderstanding of the VAE. This VAE is for native 4K output not 1K output produced from basic ZImage Turbo (or Flux ) workflow. You will need to use this VAE not on first sampler pass latent, but second pass (upscaled) latent.
Examples showing 1 megapixel image will give you oversharpened 1MP image.
Good use case is to use Ultimate SD Upscale with this UltraFlux VAE as second pass upscaled 4MP (you can even try 2-3MP) and then compare the results of normal VAE and UltraFlux VAE, which will sharpen the details in the right way.
VAE itself will not give you magical 4K details in 1K image.
yes what you're describing is the best way to use it, most people probably dont care about that though since even if used with simple 1mp gens you get a "sharper" image if you're not zooming in/pixel peeping, if you're really anal about it you can zoom in and notice the obvious over sharpening ofc.
Perhaps it would've been better if I did explain the best use case in the description mb.
Why does this VAE affect human body POSE? I can't draw a woman POSE- crossing her legs using it, but the official VAE works perfectly.
How do I use this with SeedVR2? It seems like SeedVR2 can only use the official VAE
Comfy won't even let me try it. The nodes don't recognize it, and Model Manager says it' an "unsupportedModelType."
Details
Files
ultrafluxVAEImproved_v10.safetensors
Mirrors
UltraFluxVae.safetensors
ultrafluxVAEImproved_v10.safetensors
ultra_flux_vae_ae.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEMirroredFrom_ultrafluxV1VAE.safetensors
UltraFlux-v1_model.safetensors
diffusion_pytorch_model.safetensors
diffusion_pytorch_model.safetensors
UltraFlux-v1-vae.safetensors
UltraFlux-v1.safetensors
ae.safetensors
ultra_ae.safetensors
ultraflux.safetensors
UltraFlux_vae_diffusion_pytorch_model.safetensors
diffusion_pytorch_model.safetensors
ultrafluxVAEImproved_v10.safetensors
ultraflux.safetensors
UltraFlux-v1-vae.safetensors
flux_vae.safetensors
UltraFlux-v1.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
zimage_vae.safetensors
ultrafluxVAEImproved_v10.safetensors
UltraFlux-v1_model.safetensors
diffusion_pytorch_model.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
aeultraflux.safetensors
ultrafluxVAEImproved_v10.safetensors
ultra_flux_vae_ae.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
VAE-Owen777-UltraFlux-v1.safetensors
ultra_flux_vae.safetensors
flux_ultravae.safetensors
ultra_vae.safetensors
Ultra Flux 4K VAE.safetensors
ultrafluxVAEImproved_v10 (1).safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
ultra_flux.safetensors
ultrafluxVAEImproved_v10.safetensors
ultraflux_vae.safetensors
ultrafluxVAEImproved_v10.safetensors
UltraFlux-v1_model.safetensors
ultrafluxVAEImproved_v10.safetensors
UltraFlux-v1.safetensors
2.safetensors
ultraFlux.safetensors
UltraFlux-vae_v1.safetensors
ultrafluxVAEImproved_v10.safetensors
UltraFlux-vae.safetensors
ultraFlux.safetensors
ultra_vae.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
diffusion_pytorch_model.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10 (1).safetensors
ultrafluxVAEMirroredFrom_ultrafluxV1VAE.safetensors
ultrafluxVAEImproved_v10 (1).safetensors
ultrafluxVAEImproved_v10 (1).safetensors
ultrafluxVAEImproved_v10.safetensors
ultrafluxVAEImproved_v10 (1) (2).safetensors
ultrafluxVAEImproved_v10 (1).safetensors
kkd1_v10.safetensors





