Try it out here! https://mobians.ai/
Join the discord for updates, share generated-images, just want to chat or if you want to contribute to helping perfect the model! https://discord.gg/KfAuP72uCd
USE:
The model uses Danbooru/E621 tags similar to anime models. I would not recommend just generating a character with no follow-up prompts as the results won't look very good. A good negative prompt also helps greatly.
Limitations:
Some characters generate less consistently. This problem will be consistently worked on as new versions progress.
Description
Special thanks to "Mecha from the '80s" and "RinMaru" for helping gather training images!
Base model: https://civitai.com/models/4468/counterfeit-v25
Beta 2 includes more characters and improvements to the generation of existing characters. Beta 2 was also trained at 768 rather than 512. For a full list of characters check here and scroll down https://mobians.ai/
limitations: Similarly to Beta 1, not all characters turn out perfect, more work will be done but I felt I should release this since some progress has been made from the old version and I don't want people to wait too long for an update. Hopefully the full release isnt too far off!
Check out my discord for updates! https://discord.gg/KfAuP72uCd
FAQ
Comments (13)
Upload a SafeTensors version of this.
At this point, people should assume that .CKPT only posts are doing something sketchy/ contain a virus.
Tried to to test on your test area before downloading and it won't generate anything. It just continues forever... But this is a very cool idea! Thanks for this!
I don't suppose you could make a style lora for Sonic X? Combining it with the characters in this checkpoint would be VERY cool.
Also, I am requesting Scourge the Hedgehog and Fiona the Fox be added because that would be great.
It should work, it will just take a while. Traffic has more than doubled these last few days so generating time went from like 30 seconds at peak to ~3 minutes. Currently working on optimizing the site since there's definitely room for improvement.
Also Scourge and Fiona are definitely on the todo list!
@metal079 That's great! Also, I am not quite sure what you are doing to spawn attractive images, but all I am achieving is horrible, terrible, Chernobyl abominations.
@realitythe2ndecflgi690 Yeah it takes some practice to figure out how to work the AI, taking a look at the example prompts in the images should be a good first step.
@metal079 The thing is that i saw it was possible with NSFW and that is what I was interested in. Getting a consistent style is really hard, however, as it was trained on a wide array of styles of drawing. And pairing them with loras should have helped, but it hasn't yet. So far, it's just been a whole abomination on my end. The loras are all for humans, or humans with cat ears, and that is still not really close enough to make it work. I'll browse the images in the gallery and see if anyone posted anything new and see if I learn anything, but you'd gain so crazy popularity if you released some loras for some fun sus poses. (no pressure)
Edit: I don't see the NSFW images that were in the gallery anymore. I suppose that's understandable.
I tried generating something, but got an API error. Because I didn't copy the prompt, as I didn't know this was going to happen, that prompt is lost to time.
So.. how does it actually create sonic characters? I've generated quite and copied the prompts, tried both clip skip 1 and 2 and nothing remotely resembles anything from sonic
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.