Your ZIT model seems to be listed as Illustrious. Maybe you did it cuz you are so used to uploading Illustrious :P?

Well done my friend! I'm really liking your ZiTurbo models! (NSFW40BF16) Great colors, contrast, pretty faces! I posted a couple images also. Well worth the 1k Buzz to keep the development going! Thanks!!

Dumb question: Can we successfully train loras using this? Something tells me it won't work because it is fp8. Am I right?

Does this Checkpoint work with the character LoRA?

Do i put it in the checkpoint folder or stable diffusion folder

Looking at the photos from 2602 BF16:

it looks promising with both amateur and cinematic styles in terms of realism that doesn't scream "professional studio photos." The realism looks like one that has been taken with a high quality camera that wasn't for a photoshoot, but rather for a place in our favorite "hub." *wink wink

2-25-2026 REVIEW (Version 2602 BF16)

Disclaimer: I am making an honest review based on my personal opinions and what I am think the checkpoint is trying to achieve. I'm not here to bash anyone's work. I am hoping the reviews will help others especially the authors to help them improve their work. I am using Forge Neo Classic with no loras for 2602 BF16.

-res multi with beta at 9 steps

-960 x 1280

Output Quality for Realism: 4.5/5

The output is like that from a professional DSLR shot for your favorite xxx website. Without prompting it in, it doesn't produce that studio/touch up feel we've been seeing with the original z-image turbo. A lot of the generations were like a photo snapshot that one couldn't tell if it was ai or not. Sometimes, images come out oversaturated. Sometimes there are extra limbs, heads, tails, artifacts, etc.

Prompt Adherence: 3.5/5

Prompt adherence in my case was a mixed bag. It followed prompt in some cases and not in others. It also gave huge variations in the output which is good for extra generations, but when one wants to dial in on a concept it gets frustrating that the prompt disobeys. I don't know if prompt adherence is due to certain lora mixing during training or not. Some checkpoints do better than others.

Speed: 4.5/5

I have a 3080 and usually generate one batch of three at 960x1280. Compared to other checkpoints its about the same, but faster than a few others I have tried. I don't remember the time output. Can add it in next time I remember.

Advertised vs User Results: 4/5

I was able to reproduce the quality that the author showcased without doing extra/hidden work. In some showcases from other authors, they run it through extra nodes and steps such as seedvr2 without disclosing it. This will obviously produce different results than advertised. The reason for the 1 point taken off is that the some of the prompts provided in the showcase does not match the photo output. Copy/Pasting the showcased prompts can sometimes produce weird things to happen. Like the photo the blondie was standing by the stairs. A copy/paste of the prompt made her have a long lion's tail and ears. I immediately deleted those photos.

Overall: 4.1/5

With more tweaking/mixing the quality will eventually become the one of the best. I hope the anatomy/artifact issues will be resolved within the next few versions. I hope prompt adherence can improve or perhaps even have two versions down the road. One for fine tuning a concept and one for variations.

Congrats to the author's achievements. I'm excited to see what's next.

am i the only one who's having trouble downloading the latest model? like it's super slow and it cancelled downloading.

2602 BF16 is amazing, my favourite out of Z-Image so far, keep up the great work!

Why is there a trigger word for your most recent model (2602 BF16)? I understand the purpose of using for Loras, but I've never seen that before on checkpoints. Does it trigger NSFW easier or something? Btw, thank you for all your work. You've consistently uploaded awesome models, perhaps more than anyone else has on this site

Question: I cannot figure out how to generate believable futanari, any idea?

doesn't work well with zturbo character lora :/

Very poor compatibility with LORAs.

I just wanted to say that I am loving this model. I've found all your work very useful, but this is the best so far.