Join The Tinkerer on Whop. Membership gets you early releases, private tools and a bunch of extra stuff.
👉 Join on Whop
💬 Join the community for support, free tools and early news on Discord
CyberIllustrious Catalyst is all about experimentation. It’s where creativity runs wild, blending the style of Illustrious-XL with the sharp, detailed look of CyberRealistic.
This isn’t about perfect, predictable results. It’s about trying new things, breaking old patterns, and seeing what unexpected ideas take shape.
If you enjoy exploring new directions and don’t mind a bit of chaos along the way, Catalyst is the perfect place to create something truly unique.
Recommended Settings
Sampling method: DPM++ SDE Karras / DPM++ 2M Karras / Euler a
VAE: Baked In
Sampling steps: 30+ steps
Resolution: 896x1152 / 832x1216
CFG: 4–5
Clip Skip: 2
Description
CyberIllustrious Catalyst V1.0 brings CyberIllustrious back to its roots.
It’s built on V7.0, but redesigned to bring back the rich Illustrious detail and style that had started to fade in earlier versions.
The result is a model with more artistic depth and refined styling, perfect for those who love the signature Illustrious look. It’s slightly less realistic than V7.0, but that small trade-off gives you far better control over complex, creative prompts.
Use my custom Danbooru Prompt Crafter GPT to generate highly detailed, structured prompts: Danbooru Prompt Crafter
FAQ
Comments (20)
Solving missing crotchless pantyhose, eh?
Actually V7 can do it, just didnt have time to upload "proof". :D
haha :) but this is more core Illustrious. V7 is still great.
@Cyberdelia Yea, which means it should solve slightly vanishing knowledge of V7. I will give it a try, think this can be very realistic with few LoRAs on top, while keeping relevant knowledge.
Nice, is this still using Illustrious XL 0.1? Or you move to 1.0 or 2.0? I ask because 2.0 has better understanding of natural language, 0.1 and to some extent 1.1 is mostly still tag base. I asked this also on the previous ones, but it seems no one knew. lol
As a man of culture I can confirm that "crotchless pantyhose" is indeed there. Great job!
Tested for science purposes only, of course. ;)
Finally got around to testing this - feels like a significant improvement over CyberIllustrious
- Overall texture feels more natural, especially when looking at grass, skin, and fur (CyberIllustrious sometimes felt "too sharp")
- More natural lighting
- Prompt adherence is improved with human subjects and massively improved with nonhuman ones - still not up to the same level as the cartoony models, but very good for a realistic model
- Better character knowledge
- No random nudity
Love the checkpoints!
Quick request for future versions to update the file naming to be more clear, as it can be difficult to identify from the dropdown. When downloaded, this one is only called "cyberrealistic_v10", so no indication it is illu or catalyst. Maybe something like cybreal_illu_ctlst_v10 that still keeps it short, but easily identifiable.
Thanks again!
Yep, Civitai changes automatically that name. It's fixed now.
Just rename the file yourself once you’ve downloaded it.
I separate my files into different folders, so instead of 'file name', I see 'folder/file name'
@jamalkingston1964738 Could do, but it was just a request for future variations, and Cyberdelia was kind enough to even change this one.
Even Cyberdelia said its Civitai that changes it, so presumably they usually use different naming structure. At the time I posted this comment, the filename only said cyberrealistic_v10, which didn't identify any of the 12 model variants, and it wasn't clear if it is v1.0 or v10. Nothing wrong with good file naming straight from the get-go. 👍
@Rensik I do separate files into different folders myself, but when using Forge, the folder name doesn't show in the dropdown for Highres or Refiner, just the filename. I think that was where my frustration originated from, as I will usually use a Pony or Illustrious variant for the first pass (more creative, especially with backgrounds), and then switch to a more realistic XL variant for the Highres pass.
personally i just have an innate understanding of the complexities of models and can determine which base model a checkpoint is built upon based solely on taste and smell. maybe try that?
CyberRealistic CyberIllustrious and this model seems very similar. Any notable differences ?
usually the devil lies in the details, best practice is , download booth ,run the XYZ script with same seeds and prompts (i suggest at least 10 images for comparison).
wanna ask something, i saw one of the CyberRealistic Pony picture https://civitai.com/images/106997482
it's the link, i wander if this model could do that? cuz i downloaded many illustrious based lora, dunno if i could use those lora on this checkpoint, it's all anime style, recently i wanna create some realistic style pictures, so if i could use those illustrious based lora, maybe i should keep using illustrious based checkpoint
but i'm not sure which to use, the pictures of pony checkpoint look much better than pictures of this checkpoint
Have you tried: https://civitai.com/models/1125067/cyberrealistic-cyberillustrious ?
You can see the sort of image you can create with this model
https://civitai.com/user/cyberdelia/images?baseModels=Illustrious
@Cyberdelia ah~ thx :D
The prompt adherence is insane compared to Pony models, 10/10 would goon for a Z-Turbo version trained with Danbooru lmao
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.