NOTE: Civit has made the download UI confusing. For the v4.0 model, it contains four versions:
BF16 UNET (11GB)
FP8 UNET (6GB)
Q6 GGUF UNET (5GB)
FP8 Checkpoint (16GB)
I uploaded them very descriptively but Civit has stripped away the descriptions so it's now confusing to see what you're downloading. The FP8 checkpoint is an all-in-one version that contains the VAE and Text Encoder. For the UNETs you need to provide those extra models.
I've used 100s of curated images to help add some better color, contrast and detail to the Z-Image Turbo model. It's not a huge difference yet, but it is noticeably better in side by side tests. I'm starting out a bit conservative because I don't want to mess up any of the great compositional features that Z-Image offers.
Use the workflow embedded in the showcase images for the model version to get the best results. Check the About this version section for links to the workflow and recommended settings. Z-Image currently mangles nipples and genitalia so my workflow uses SDXL detailers to refine just those aspects of the image.
P.S. I would appreciate any constructive feedback on how to make this model better.
Description
An incremental update to the first Insta version, hopefully a little more forgiving with character loras but no guarantees. Use the same settings as described in Insta and take the ComfyUI workflow from one of the showcase images.
FAQ
Comments (15)
Really impressive update to the Gonza series..much different from Z PoP...extremely diverse in results, which is refreshing for a Z-Turbo checkpoint...thanks for the efforts!
it does realism quite easy and good, but prompts longer than 100 words already degrade the quality. only good for portrait, but too much randomization. I can only recommend for portrait. Nothing else. Thank you for the checkpoint.
Thanks for the feedback. I hadn't noticed an issue with token length but I'll try to keep an eye out for that. Compositional randomization and facial diversity are kind of the point of the Insta models. The regular ZPop versions are more stable.
I liked ZPop. Insta2 is really solid work. Love it
Can we have some samples with male anatomy? ;)
Sorry, this z-image model is not great for that. z-image in general is not great for nudity but there are some other models that are better focused on that. This model is only ok with female nudity and my workflows are designed to use sdxl-refinement for better realism.
@GBRX awww... that's sad
@iwantobeleaf my sdxl and chroma models are better for that.
where can i download the UltralyticsDetectorProvider bbox.pt ?
@GBRX Thanks!
Thank you for the excellent model! I was wondering regarding the sample image prompts. They are structured more like sdxl prompts - commas, parentheses - while zit usually prefers natural language. Is this intentional? Does it work better without natural language?
I tend to use a lot of the same prompts across many different model types, just to save time. That said, z-image does seem to support both natural language captions and sdxl-style tags.