Z-Image-Turbo NSFW Unlock is live for Generation!
This is the definitive full-NSFW variant of Z-Image-Turbo — purpose-built to completely remove all safety restrictions while perfectly preserving the original model’s signature aesthetic, speed, and quality.
Key Features
Based directly on Z-Image-Turbo (distilled 6B parameter model)
Trained on 6000 high-quality images with rich, illustrious tags to learn many NSFW concepts
Fine-tuned on 350 meticulously curated images to lock in exact style consistency with the base Z-Image-Turbo
Full uncensored NSFW capabilities: seamless handling of nudity, genitals, explicit anatomy, complex adult scenes, fetish concepts, and more
Maintains the exact Z-Image-Turbo look & feel — photorealism, lighting, composition, and artistic coherence are untouched
Works flawlessly with natural language prompts AND booru-style tags
Does everything the original Z-Image-Turbo can do… plus unrestricted NSFW generation
It is a mature model that delivers extremely consistent, high-quality results right out of the box. Tasteful nudes or explicit content, it handles it all while staying true to the Z-Image-Turbo.
⚡️ Fits comfortably in 16 GB VRAM consumer GPUs
⚡️ Identical efficiency and instruction-following as the base model
Description
FAQ
Comments (17)
does this work with loras as zit org does? if yes then need a Q8 gguf version please...
This, we need a q-version for the people without high-end cards
try this tool i made for converting to gguf: https://github.com/qskousen/ggufy it converted this model in 23 seconds for me, and it worked fine. did a quick test with a lora and that seemed to work fine as well. ggufy convert zImageUltimateNSFW_v20.safetensors -d q8_0
Looks nice i cant wait to try it out.
Can you draw real pubic hair?
Im struggling to get this working in comfy UI. DOes anyone happen to have a simple workflow using it they can share?
Actually, update ComfyUI and look in templates. There is a Z-Image Turbo template which can be used for that. Variant with workflow here are less likely to give you promised result because 1st model there is still the same somewhat restricted base zimage. (Though even with full variant it's not quite delivering what's promised here. But the progress is present!)
Are there any suggestions for ideal sample name and step settings for this model?
生成的图片画面都比较黄，没有zit原版亮丽。不过除此之外，真是一个非常出色的模型。
加lora调整啊，例如KOOK大佬的瑶光，应该可以的， 你试试，我还没下载模型
这个模型在双人动态表现上非常完美，在人物和画面上也完整的保留了原本的美感，但是唯一的缺点就是从18岁-38岁的中国女性都是同一张脸，即使用亚洲，日本，韩国这些提示词还是同一张脸，太遗憾了，不知道是什么问题导致的，要是能解决这个问题，这个模型就是神作
I've tried a lot of models in the last month. Out of all of them, this one stands out by far. It's fantastic.
The best model, very few errors. Highly recommended.
The best model, very few errors. Highly recommended. I use Draw Things app mac os