My first Z-Image Turbo model.
Create images with a Kodak Portra style!
for the best result, use a film grain node, and add some grain!
Support my work by bying early access ! I can make the models ready for onsite generation with this buzzz.
Description
another version, little more realism
FAQ
Comments (7)
Your v2 images look good, but what did you add to reach 19GB? Did you include Clip and VAE?
FLUX.1-dev VAE and qwen_3_4b text encoder are included but make little difference and can be removed from the state_dict without issues. The transformer is the juicy part.
How it goes with loras ? Also model can’t handle pussy without workflow fix?
some lora's working fine, some are crap, depends on how they are made. the fix is optional, but creates way better pussies, none of the Z-image Turbo images creates good realistic ones yet imho.
I tested version 2.0 and I think it's a very good AIO checkpoint. I'm using er_sde/beta with 8 steps and the images look very good. I added a little color with the DPO LoRa. The results are very nice. Thank you for the checkpoint.
thanks for your review! really appreciated!
Any chance to share the workflow you are using for this model?