Join The Tinkerer on Whop to get early releases, private pages, and the Tinkerer Discord role - all in one place. 👉 Join on Whop
CyberRealistic Classic is a more unrestricted version of CyberRealistic that offers the same freedom and flexibility. It serves as an experimental platform for testing new features and concepts that haven't been integrated into the original CyberRealistic. Essentially, it functions as a fully operational alternative edition of CyberRealistic.
Recommended Settings
Sampling method: DPM++ SDE Karras / DPM++ 2M Karras
VAE: is already Baked In
Sampling steps: 30 Steps
Resolution: 512x768
CFG: 5-7
Upscale: 2x
Upscaler: 4x_NickelbackFS_72000_G
Denoising strength: 0.3-0.35Example Negative prompts
watermark, text, censored, deformed, bad anatomy, disfigured, poorly drawn face, mutated, extra limb, ugly, poorly drawn hands, missing limb, floating limbs, disconnected limbs, disconnected head, malformed hands, long neck, mutated hands and fingers, bad hands, missing fingers, cropped, worst quality, low quality, mutation, poorly drawn, huge calf, fused hand, missing hand, disappearing arms, disappearing thigh, disappearing calf, disappearing legs, fused fingers, abnormal eye proportion, abnormal hands, abnormal legs, abnormal feet, abnormal fingersless tokens:
lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
Need the perfect prompt? This custom ChatGPT instantly generates high-quality prompts or helps you refine your own.
🔗 [Try it now on ChatGPT]
Description
This version was placed on the operating table again with another round of adjustments and repairs.
It might just be the final version of "CyberRealistic Classic" ... or perhaps not, who knows? ;)
FAQ
Comments (17)
Hi so V2 is new today? At the risk of sounding like a broken record, any chance of an inpainting version?
Personally never use an inpainting model, always use the normal checkpoint. But will upload it today.
Done :)
@Cyberdelia Legend
Ok, I feel like I have to say this at this point. The sheer volume of updates between the two versions has become a bit exhausting. I feel like the versions from four months ago were amazing, and these versions are just a different kind of amazing, but necessary?
Not accusing, but some on here are feeling like certain releasers are spamming updates to stay at the top of ranks on the site. Not sure what benefit that would be, but man, I hardly get a look at these before the next one comes along. That said, *sigh* I will keep getting the updates.
Only for you though, because cyber models are actually the best in any version.
The one uploaded today was an inpainting version of the one released a few days ago. It was requested by a different user, and Cyberdelia was kind enough to comply with the request. If you're not interested in inpainting, you can very easily skip it. It's hardly what I would consider spamming updates.
@hollaman I am really speaking in generalities, and truly not making any accusations. Inpainting versions aside, there are still a lot of updates to these models, I even joked about it months ago on this very model. This is one of those cases where I should have probably just kept my mouth shut, but I didn't. Oh well, just my thought on this, hope it doesn't bother anyone.
Edit: By the way, this is V.2 Proper, and I'll give you that 1.7 came out over 3 weeks ago, but in my time that is a blink. In SD time, it is more substantial though.
@MysteryWrecked I understand what you're saying. It is indeed a lot, but I will try to explain why. As you know, there are two versions: "CyberRealistic" and "CyberRealistic Classic." Both have a different foundation (which may not be entirely clear, but it is to me). Since I am more focused on the technical aspect with Stable Diffusion, I spend hours each day testing new features, and these are integrated into the "CyberRealistic Classic" version by me. Such a version is thoroughly tested, and once it reaches a certain level, it is shared with you all. Hence, the numerous versions. "CyberRealistic," on the other hand, undergoes fewer changes and is only modified when I am 100% satisfied with the adjustments. That's why I often refer to "CyberRealistic Classic" as a beta version and "CyberRealistic" as a stable version. But because the have a different base/foundation they are a also bit different. You could also say that "CyberRealistic Classic" is a more playful or freer version.
And about "stay on the top" I genuinely have no interest in maintaining a top position in the rankings. For me, as I mentioned before, it's just enjoyable to engage with the technical aspect. And again, I completely understand your point as well. It is quite a lot :)
CyberBot out... beep boop ;)
@Cyberdelia I like the idea that classic is a kind of advanced beta, I can say that the current version of classic seems to have an edge over the main model, so you definitely did something right. Running a side by side comparison, the main model is very, very good, but classic is outright amazing in this version, and I'm genuinely not sure if I've ever seen a better model. Someone recently said "Who needs XL?", and this model has the answer: Nobody.
I get the same FOMO vibes with the accelerated pace of updates in the whole SD ecosystem and can very well relate with OP's feelings. Now I've come to appreciate that folks are experimenting and evolving in public and am finding my own pace at which to experiment with new stuff, and where I'd rather let some things go and look at the amazing reviews and experiments done by others to pick up stuff that I've missed out on. I found that the good stuff usually bubbles up after a while.
There's really no real point I am making other than saying thanks @Cyberdelia for sharing work from your process, but also thanks @MysteryWrecked for voicing your thoughts that I find so relatable. I find the community to be quite supportive and truly appreciate that folks share their thoughts :)
@de_lorian This is also why I've separated both models. There's a distinction, but also to avoid overwhelming the main model with beta updates, and as I mentioned earlier, I understand the abundance of updates, and if you're like me, always wanting to have the latest version of something, it can be quite frustrating :)
How to download FP32 model ? is there a list of FP32 model in civitAI ? thanks
I would also like to make a realistic model, and I'm currently trying through Dreambooth spending a lot of time on it, but I can't figure out the settings, the results are disgusting. You can tell me what settings you use, what resolution of images and according to what scheme the training takes place. On the Internet, I could not find lessons on teaching such large-scale models with a package of images of several thousand. I trained 1900 portrait images of 512x512 at 380000 steps tonight (Learning Rate 0.000001), as a result, some kind of crooked nonsense turned out. Save me, brother.
I've never seen the author of one of the big models share their recipes. It's all one big proprietary secret with them. I will give you a little advice, however. Start small. Get proficient with doing fine-tunes on a single concept, a few hundred images, first. Then slowly broaden the range of concepts and images. Also, you can also achieve something close to your desired outcome by training multiple fine-tunes, then merging them. I like to save every 20 steps or so. This way if you overcook by the final steps, you have checkpoints all along the journey you can refer to in order find a workable model that's trained just enough, or give insight as to whether your learning rate or scheduler needs adjustment.
@gurilagardnr true, it's not about sharing secrets - it's just that it's a bit complex to answer.
@Cyberdelia are you using kohya_ss, one trainer or something similar?
@hourglassfcupslimwaist Kohya and some custom tools. I know that Kohya has some flaws but it does the job.