Join The Tinkerer on Whop to download this model and many more — all for one monthly payment. Plus, get early releases, private tools and a lot more. 👉 Join on Whop
Yes, there will be many versions. And no, I’m not stopping. I know not everyone loves that, and honestly, it doesn’t keep me up at night. Models don’t improve on their own, and standing still is the same as falling behind. So buckle up and enjoy the ride.CyberRealistic Z-Image Turbo Catalyst is part of the Catalyst experimental line - a dedicated space for targeted experiments that go beyond.
This build focuses on three specific areas:
Character LoRA compatibility: better integration and less drift
Skin & color rendering: refined skin tones and overall color behavior
Bug fixes: corrections for known issues
Why Catalyst and not a new Z-Image Turbo version? These changes are experimental by nature. Rather than folding unproven adjustments into the mainline release, Catalyst exists as a separate track so the standard release stays stable while this build gets real-world testing. Think of it as the skunkworks branch.
⚙️ Personal Settings (Forge Neo)
Required Additional Files
Make sure you also have the following:
16 GB+ VRAM: qwen3_4b.safetensors
8–12 GB VRAM: qwen34bfp8_scaled.safetensors
All VRAM sizes VAE: ae.safetensors
This model is shared as-is, for users who enjoy experimenting, benchmarking, and pushing models outside the comfort zone.
Feedback, findings, and edge cases are welcome - this release exists primarily to learn from real-world usage.
Description
A stable and reliable build with strong LoRA compatibility, making it a solid choice for consistent results across a wide range of setups.
This NSFW version is less versatile and expressive in that area compared to V1.0.
FAQ
Comments (22)
Fantastic checkpoint, artistic yet photorealistic driven, I'm getting amazing results.🤘
Cool! I'll give it a try, thank you!! Are the settings you show for the Z-Base version? 1 CFG and low steps? Thanks!
This is Z-Image Turbo! But the settings are correct. You could increase the steps to 12-14
@Cyberdelia Ah... thank you! The description says Z-image Base for the 0.7 version...
@GenoMachino Yes, noticed that! I changed it.
I've never seen the DPM++ 2s a RF sampler you used in at least one of the sample images. I did a quick Google search but I'm wondering if you can tell me if it's available for Comfy.
I guess it's Forge Neo only.
Nice💕👍🏻
In my opinion, version 0.7 performs better than 1.0. However, you've set it as the base version here.
Ah, good catch - that explains the confusion in some of the comments. I've updated it now.
Version 0.7 also offers more flexibility for custom adjustments. So if you're planning to use a LoRA, it's best to go with v0.7. That’s exactly why I decided to release it as well.
@Cyberdelia I think it has a bit too much color. Perhaps a little less color, that can be easily increased with the DPO LoRA.
@Cyberdelia Ooooooh, I just tested it. If you give DPO LoRA a negative value, it also removes the color :) nice
@Gorean Oh, that would indeed make sense. The DPO LoRA wasn’t used in V1.0 but it will probably have a similar effect if you use it in the negative.
@Cyberdelia Wait I'm still confused. Which is which?
@Miramir it is ok now, v0.7 was at first as Base, it is not Base and he changed it, so problem is solved.
@Gorean @Miramir Yeah.. sorry about the confusion.
I'm testing v1.0 and getting very grainy images. I noticed the file size is only 6GB. Is this an FP8 version? Also, would you mind sharing your exact text encoder? I'm testing different ones to evaluate the results.
There are 2 versions bf16 en fp8
how do you add new sampling methods in forge?
Like: DPM++ 2s a RF?
It's part of Forge Neo.
looks pretty good so far. dpmpp_2s_ancestral/bong tangent, dpmpp_sde/beta, quite a few options sampler&scheduler wise, and no synthetic slop alarm like some other zimg checkpoints which shall be unnamed.
I've been buckled up since 2022, enjoying the ride, and excited to see @Cyberdelia is keeping at it.
Looking forward to an ERNIE version. Ernie is amazing for training. Much faster than ZiB.