last word please about vs scaled zit from kijai.

#1
by pikkaa - opened

this is just more 'clear' compared to similar 'fpX' other ZIT models , no loras added no variant of composition and elements ?

pikkaa changed discussion title from l word please about vs scaled zit from kijai. to last word please about vs scaled zit from kijai.

The Z-Image-Turbo_clear model is a conversion of the original Z-Image-Turbo_FP32 model released by Tongyi-MAI, adapted for use with ComfyUI. We achieved improved detail by adjusting the output of specific layers.

No LoRAs were merged. While the composition might change slightly, there are no major alterations to the elements.

I am not familiar with kijai's scaling method in detail. However, the Z-Image-Turbo_clear_FP8 model was converted using ComfyUI with the --fp8_e4m3fn-unet setting.

Therefore, I believe the outputs from the two models will be different.

The Z-Image-Turbo_clear model is a conversion of the original Z-Image-Turbo_FP32 model released by Tongyi-MAI, adapted for use with ComfyUI. We achieved improved detail by adjusting the output of specific layers.

No LoRAs were merged. While the composition might change slightly, there are no major alterations to the elements.

I am not familiar with kijai's scaling method in detail. However, the Z-Image-Turbo_clear_FP8 model was converted using ComfyUI with the --fp8_e4m3fn-unet setting.

Therefore, I believe the outputs from the two models will be different.

me..as a normal user , i have no enough storage to keep up with testing all models , i have picked your model variant 'fp16' to download and test over all available 'right now' z-image turbo model.

can i use abliterated qwen3 4b text encoder it will work instead of normal one ? please tell me if you sure it will works with your model ?because i'm out of storage i can't waste storage for just trying , hopefully it can just works.

pikkaa changed discussion status to closed
pikkaa changed discussion status to open

abliterated qwen3 4b text encoder

Link please?

https://huggingface.co/IIEleven11/Qwen3-4B-abliterated_dark/tree/main

The second model generated the same illustration as the original Qwen-3-4B. Please try the other models yourself.
image_difference_checker_202512082039

thanks a lottt!!..and i will do test other one by one just i need my time.

easygoing0114 changed discussion status to closed

Sign up or log in to comment