Join The Tinkerer on Whop to get early releases, private pages, and the Tinkerer Discord role - all in one place. 👉 Join on Whop
After a ton of requests, I’m finally rolling out CyberRealistic Flux (FLUX.1 dev)! It’s designed to make realistic images, both safe-for-work and not-so-safe-for-work. It’s not perfect yet, but it’s a solid start and sets things up for what’s coming next.
Heads up: The (non AIO release) doesn’t include clip models or VAE. You’ll need to grab those separately here.
(clip_lorg.safetensors = orginal clip_l file. clip_l.safetensors is the CLIP-SAE-ViT-L-14 one)
Settings I Use:
Sampling method: DPM++ 2M / Euler
Sampling steps: 20/30 Steps
Distilled CFG Scale: 2–3.5 (Samples were done at 3.5, but lower values often give better results)⚠️ A Quick Warning:
This model can make mature or sensitive stuff. You’re in charge of what you create, so use it wisely (or at least get creative with it).
đź’¬ Want Better Prompts?
There’s a custom ChatGPT that’ll whip up great prompts for CyberRealistic Flux:
đź”— [Try it now on ChatGPT]
Description
Rebuilt almost from scratch and packed with everything learned from earlier versions. It’s smoother, sharper, and delivers way more punch. Quirks are fixed, LoRA compatibility is better, and the output looks cleaner and more detailed. With broader training under the hood, Flux v2.0 pushes creative NSFW limits while still keeping things stable and slick.
Now available in FP8 and FP16.
FAQ
Comments (26)
Skin textures look too AI and unnatural, too smooth and 3D modelish. Is there a way to fix that? Any loras?
Using additional LoRAs to enhance certain aspects may reduce compatibility with Character LoRAs. I’ve worked hard to minimize this effect, but it’s a tricky challenge.
@mygenaiessentials138 also lowering Distilled CFG Scale will remove the 3D plastic skin more (from 3.5 to >3)
Its not easy, cause what you see is basically FLUX itself, which was trained and captioned in most synthetic way possible.
Even extreme modifications like Chroma suffer from this (altho there its more about prompt).
Specific workflows and LoRAs might help with this, but its up to users to figure that out.
And that's exactly why I continue to use the incredible SD1.5 model of Cyberdelia which is far better for realistic skin( especially the classic 3.2). Sadly sdxl, pony and flux suffer from initial censorship in training and it may be not possible to bypass this... but if somebody can, it's probably Cyberdelia !
@coincoin2044365Â Well, running 2nd pass after FLUX is good idea, most use SDXL for that tho.
It seems like V2.0 was a step back in prompt adherance and detail. Check this post:
https://civitai.com/posts/21222092
The first image is V1.5, the second V2.0. The picture is much more blurred out, especially the face. The layer of sand on the skin is completely gone. Its the same seed and settings, using sgm_uniform as sequencer.
Am I doing something wrong here? Please don't get me wrong, both images are great, the change is just...unexpected.
The issue with the previous versions and this one is that in V2 the several sliders have been moved slightly down. The reason for this is that LoRA support suffered a lot before, and now it’s much better. It’s a trade-off, and I also noticed from the feedback that this was important.
I also see that settings like Distilled CFG Scale - when set lower than the default 3.5 - make the result more realistic. Personally, I feel that the latest version follows the prompt much better than, for example, V1.5.
But thanks for your message - it’s definitely something to keep an eye on for a future version.
@Cyberdelia I Really appreciate you responding to my feedback. Would it still be possible to get a FP16 version of V1.5?
A merge of V1.5 and V2 is a good solution. Less blur more details. Loras work fine with it.
@sam345Â care to share the percentage? or was it 50/50?
@bewunderteradler112Â 50/50 is a good start. For close-ups of characters i would choose 60/v1.5 and 40/v2. For landscapes and architecture more V2 is better.
Fantastic model! Always one step ahead.
Exceptional results with V2 and Forge... Thanks!
hi, looks like you are using custom cyberflux_clip_l, where we can get it from ? :D
Hi. I am a big fan of your models. But unfortunately I have to say that V1.5 is really better than V2. Skin detail, overall detail, composition, all of this is better in version 1.5.
A new version doesn’t necessarily mean it’s better in every way. V2.0 is a version that has far fewer issues with LoRAs. Hopefully, it will eventually surpass v1.5 in the areas you mentioned.
And yes, I’m working on it right now.
That's right, version 1.5 is much better but loras are extensions and isn't the quality of a model to use as little as possible?
@barinovrs433 @EMYSTARIEL I’m releasing a new version today. I’ll probably get a lot of comments again about why I’m putting out another one so soon, but I agree that the difference between v1.5 and v2.0 is quite significant.
@Cyberdelia I dont understand why people or Users are mad at your model's ,you been working hard and we all glad we are able to get a free flux model is this detailed. keep up the good work :D
@Dead_Sec I try to see criticism as a form of positive feedback, because I believe most people mean it that way. If you always take criticism as hostility, you’ll end up missing the chance to grow.
@Cyberdelia thats good to hear :D
@Cyberdelia i actually have a dataset with gemma3 captioned images , around 13k images ? if i remember correctly
@Dead_Sec For my part, I'm not angry at all lol! Quite the contrary, this model is wonderful, incredible, magnificent, versatile and above all versatile. Cyberdelia's work is impressive. A review from me is always positive. LOL! I really wish that this base of 1.5 would be perfected with artistic elements and that certain things like anatomy would be more accurate. But this work is close to perfection.
@Dead_SecFriends, I am not angry in any way. I really appreciate and respect the author's work, and of course I sincerely love his works. I just said that for myself I will choose version 1.5 for now, I like it more. But in no way did I want to offend the author or anyone.Â
@EMYSTARIELÂ thats good to hear :D
@barinovrs433Â yes :D its a surprise we are getting a good flux model for free xD
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.