This model is a precision-tuned ZiT merge designed to push the boundaries of fine-grained texture, anatomical accuracy, and atmospheric lighting while maintaining compatibility with LoRAs trained on the base models. I've train my character and style LoRAs on vanilla Z-Image Base. They work fine on Turbo derived models and tend to be more stable during training.
I will upload future models to Hugging Face also, it's a lot faster than dealing with a VPN
https://huggingface.co/PsychoLogic/PsychoMix_ZiT
The Composition
The architecture of this model is a blend of checkpoints and custom refinements:
v1.x
Core Foundation based on Moody Porn Mix v10 for subject pose, with various other models merged for style and consistency (see 'About this version' for specifics for a particular release)
For a more raw/realistic look I recommend using z-image-turbo-flow-dpo with a strength of -0.5 to -0.75 (yes, negative). It was baked into the base model and many (myself included) think it's merely a sloppifier.
v2.x
Core Foundation: A balanced merge of Z-Image Turbo NSFW v7.2 for subject pose, CyberRealistic Z-Image Turbo Catalyst v0.7 for style and technical precision, and selective merge from Famegrid ZIB v1.0, focused on later layers
Feedback & "Show Your Work"
I am actively looking for constructive feedback to improve this and other models. If the model isn't performing as expected for you, please help me fix it!
The "Pics or it didn't happen" Rule: If you encounter issues (artifacts, anatomy failures, or poor prompt adherence), please post your generation with metadata (Seed, Prompt, Sampler). Negative comments without an image and settings typically don't help. If you want a fix, show the failure so I can tune it out in the next merge.
Description
I was tired of the DPO look, so new mix this time:
Z-Image Turbo NSFW
CyberRealistic Z-Image Turbo Catalyst
Famegrid ZIB
FAQ
Comments (21)
I like it. I like to push models with DPM++ 2s a RF and this model handles is quite well. My only two negative comments: for the love of me I couldn't get the model to give me a backhand tennis stroke and generating a "20-year-old woman" made them look very young.
beautiful model! can't wait to try it out!!
FP16
BF16 doesn't work? Some kind of ye-olde hand cranked GPU?
@psychologicau i'm using old CPU , i don't have gpu. fp16 needed.
@amazingbeauty Fair enough. I'll see what I can do for the next one. I'm not happy with the LoRA compatibility of 2.0 and the base model is a lot less stable than Moody leading to some body horror, so it shouldn't be far off
@psychologicau all zit has lora compatibility issues , no zit can work with all loras , it's vary
@amazingbeauty v1.1 provided far better likeness in my testing, and was less likely to overcook images when applying a LoRA at ~1.0 strength. I'll probably revert back to a Moody base, and just undo the DPO
@psychologicau DPO is bad in my experience , FP16 please v1.2
@amazingbeauty I will work on quantization & FP16 conversion tonight, should have something soon
@amazingbeauty FP16 for v1.2 finished uploading after 3 attempts, can you see if this works for you? I modified https://github.com/silveroxides/convert_to_quant and add a new 'convert bf16 to fp16' option
@psychologicau i've made an open source program that is aimed specifically at converting image models, and it can convert to fp16 safetensors or various gguf's, pretty easy to use if you want to check it out https://github.com/qskousen/ggufy
@ferretduck Would you mind sharing the instructions to make a FP16 .safetensors file using ggufy? I couldn't see the option
@psychologicau Yeah, if you are using the GUI, once you've dropped in / selected your file, change the "target format" to safetensors and then select F16 from the data type dropdown.
@ferretduck this conversion is processing heavy like generating images..aka need for cuda and large long Nvidia card ? or it's just fast ?
@amazingbeauty it doesn't use a video card at all. Should take just a few minutes to convert a model, depending on the model size and your CPU.
@ferretduck Hey ferretduck, do you think you could add FP8 for ZIT to Ggufy? It's like most models do come in FP8 but then some don't and I looked into making FP8 myself but it seems very complicated. I'm not using GGUF for ZIT since they originally come in FP8 and I have over 30 models that would need to be converted to GGUF and then whenever you get new ones convert again and again. Thanks
@ferrrett33 ggufy can do fp8, including for ZiT. just select safetensors instead of gguf for the target format
Unfortunately, in version 2.0, when generating NSFW images, issues frequently occur where bodies are distorted or prompts are not properly understood (such as only two women appearing in a sex image of a man and woman, or Westerners appearing when specified as an Asian couple, etc.). This rarely occurred in version 1.1.😢
Yeah different base NSFW model. Would you mind sharing some images / prompts which didn't work and I can see if it can be fixed?
Clean job