Hi there! I decided to have some fun while indulging in a nostalgic cartoon! - Freakazoid!
I hope you remember this character and can recall where you've seen him. I had to modify various versions of the character, there were some unstable options, but this is the only one that turned out better in quality. You can create your own character and make different iterations with him. I hope you'll find it interesting! All the examples and descriptions will come with pictures, so watch and enjoy! If you have any questions, feel free to ask.
Super toon extraordinaire
Freakazoid, Freakazoid
Runs around in underwear
Freakazoid, Freakazoid
Rescues Washington, DC
Freakazoid, Freakazoid
Unless something better's on TV
Freakazoid, Freakazoid
Description
The dataset has been improved, the pictures have been updated and the generations have become more or less clean.
FAQ
Comments (13)
Hi! I just wanted to point out that since this is Lora and the images were taken from a cartoon, the generated pictures might not be 100% accurate. However, they still capture the essence of the character. If you come across any successful matches with tokens, feel free to share examples!
He's here to save the Nation!
"So watch the meme invasion~"
"Check out dexter douglas~! Nerd computer Ace!
He Tried Stable Diffusion-'n got zapped through cyberspace!!
He turned into the Freakazzoid! Strong and super quick!
Gon' drive the ladies crazy~ cuz now he's gotta dick!"
Also yeah, I made "the dick" one... 🥴
"sorry, not sorry" 😉👍
the video compression affected the dataset images which affected the results I suggest upscaling the images then resizing them to get a cleaner dataset
you may also try to use a negative prompt "noise,blur,worst quality,sketch" to get better results :)
I will do so! Thanks for the advice, that's exactly what should have been done
@Kotoshko don't mention it also good job on th new lora :)
@GetAiKOFi I added another work, there is one hero from the same animated 🤣damn
@Kotoshko Good stuff 😂
Ver1.5 is Leaps and bounds above 1.0!
Tips for usage~
I found out negative embeds seem to really hate it! Like wow. 😯
Probably for the simple style and subtle artifacts.
I found removing the negative embeds in a strong checkpoint gave me the best results by far! Only kept BadHands-4 because it mostly played nice and helped hand-gen.
~ I found adding blue_skin (F!) improved the likelihood of getting the logo to show up and the skin to generate the right color.
Creator suggestions~
Definitely keep refining the training data for clean composition and lines where possible. Artifact-ing is probably why every negative embedding I paired it with chewed up the Lora before it even had a chance. 😅
Maybe... uuuhhh... add some rotated figure images in the training data so I can make him cartwheel and handstand, 'ey? 😉
Thank you for taking the time to read my LoRa! I'll try to come up with something in the future! Indeed, I used to add such a tag - f! When I have time, I will sit and make screenshots from the frames of the cartoon. Thanks again for your attention to detail!
@Kotoshko Really appreciating your dedication to improvement. It really showed in that update~
Happy to assist if you need anything~ (merge checkpoints/etc.)
Taking a proper glance at your gallery I can see why you made such a strong and fast turnaround, s'good stuff.