Z-Image is here, and it's pretty damn great. Let's make it more NSFW!
Recommended Settings
Steps: 12
CFG: 1.0
Sampler: dpmpp_sde
Scheduler: simple
FP8 Download
For some reason, when you upload multiple formats (BF16 + FP8) on one model, which they now support, they hide the FP8 under the BF16 model group in the side menu...
ComfyUI
Models should be placed in diffusion_models, and uses the official workflow with the Load Diffusion Model loader, and loading the VAE/CLIP models separately. Models named "_AIO" should go in the checkpoints folder, and uses the normal checkpoint loaders.
Description
Trained for 46.000 steps on a huge dataset.
Special thanks to Alcaitiff for their awesome work!
Place this version in the diffusion_models folder.
FAQ
Comments (9)
pretty good, but doesn´t work well with LORA ... :-( ... any chance for a fp16 version?
Yeah, it does not match well with other LoRAs, they eat each other up. I'm not sure how to solve this. It may not be possible with the current method we are training ZiT.
FP16 version coming.
I was same via lora
Can't say I dislike this model, its fairly interesting, I would even call it POV ZIT as it specializes in POV view
Didn't consider that, maybe there's too much PoV in the training data if it has a bias towards this. Good catch.
Even after adding LoRA, it still generates speckled images. The LoRA was trained on this website.
Yeah LoRAs are not working well here. Hoping to fix in future versions.
@6tZ Does this model have special requirements for text encoders and VAEs? I directly used the invoice and the UI generated based on the official version of Z-Image that I modified myself, without adding Lora, both is still a black and white speckle image, like a QR code
@leewheel456 hmm, no special requirements. Version 1 should be placed in diffusion_models, and uses the official workflow. Version 2 and forward are checkpoints, so you use the normal checkpoint loaders.