Z-Image is here, and it's pretty damn great. Let's make it more NSFW!
Still a work in progress. It can output good genitals, but not always. About 3/5 images are good, working on penis stability specifically for future versions.
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.
Recommended Settings
Steps: 12
CFG: 1.0
Sampler: dpmpp_sde
Scheduler: simple
Description
Trained for 246.000 steps on a huge dataset. Special thanks to Alcaitiff for their awesome work! This is only the diffusion model. Does not include CLIP or VAE.
FAQ
Comments (18)
5k too much
No worries! I got you!
Just for you, I will release the model for free in a week! <3
dont worry it will be free in 6d16h40m
maybe a stupid question but, cant one just extract the diffusion model from the combined checkpoint using ComfyUI?
@bewunderteradler112 Yup! That's what I did here.
It's just that for many people that's a harder thing to do.
I'm not sure if it requires a lot of VRAM or RAM either.
congrats on the v2. can you tell us what the differences between v1 and v2? thanks
Read the notes. v2 was trained with more steps
qek is correct, I trained another NSFW LoRA with a new dataset I made for this model, which is then merged into here. It helped a lot. But there are still lots of improvements.
@6tZ sorry, i meant to ask if the new training allowed for more accurate and more variety of NSFW. So far, for example, none of the zimage resources can do side view missionary, etc.
@Melodic_Possible_582589 Hmm, I would say no to that. I haven't gone into detail about poses or positions. It's more about general anatomy and unlocking genitalia overall.
Hi, could you make an FP8 AIO?
So many requests and versions XD
I can do it :)
@6tZ Thanks, I'm using version FP8-AIO (10GB) of the user SeeSeeLP, And it works well for me on a GTX 960, believe it or not.
@syphonfilterargAI It's nothing special, why not using split files? I am sure nobody trains text encoders of models like Z
@qek For me, It's just a convenience thing. Less annoying nodes to deal with in comfy, workflow is cleaner, and fewer files to manage. It's easier, but less optimized file-size-wise. But for some, this is not a problem.
@6tZ Exactly what you say.
@syphonfilterargAI it's up:
https://civitai.com/models/2237711?modelVersionId=2578314
I made it for 2.1
@6tZ Genial, thanks.😄