THIS IS THE SDXL VERSION
Based heavily on the 90s TMNT live action films, TMNT 1 & 2.
Trained on around 50 high quality artworks and film frame images at around 1000 steps.
Recommended weights of 0.75 - 1.0 (Lighter weights produce more artistic images and retain less of likeness while heavier weights produce a more film accurate outfit) A word of note: I have not been able to reproduce very high quality depictions of his claw gauntlets, however (with certain models) the generations seem to come out quite nicely for the most part.
For best use case in prompting I recommend:
Depending on what your desire for outfit "a TMNTShredder 1man armored samurai knight with claw gauntlets, wearing a purple/red-purple outfit with pauldrons and metal samurai helmet with mask" (exchange purple or red-purple for whichever style you prefer, or exclude the highlighted text entirely to get more artistic variations of the outfit).
For negative prompt tags I recommend:
Anime/Cartoon/Comic: "cropped, 3d, 3d render, toy, figurine, statue, video game"
Realistic: "cropped, 3d, 3d render, painting, anime, toy, figurine, statue, digital painting, comic book, video game, drawing"
Generations seem good using both Clip Skip 1 and 2
Overall the model is not perfect but is fairly flexible and good for what it does.
Please share your thoughts in the comments, add a review, post your generations, or if you like my work and want to support me further consider tipping Buzz or reaching out via my DeviantArt/Discord/Patreon
Description
SDXL Version
FAQ
Comments (7)
Perfecto! Is that whole phrase the triggerword? And...if this covers the 90s Shredder, does that happen to include a variation for the Super Shredder? :D
TMNTShredder is the triggerword, the whole phrase is just keyword tags I trained into the LoRA to assist in better accuracy, so you might be able to get away with only using "TMNTShredder"... But, I'm overly verbose, which is annoying to some, but believe it or not, my verbosity has seemingly paid off in the way I caption my images for training lol.
As for Super Shredder, TMNT 2 Secret of the Ooze literally only had a few frames of that variation and they weren't clear or useful for AI training, so I couldn't really include it in the dataset. I'm looking into other sources for a movie accurate Super Shredder dataset, but acquiring high quality Super Shredder images is proving to be a daunting task.
@ArchAngelAries Ok, I get it! Yea I've used your TMNT models and learned that being 'overly verbose' is helpful. I tend to write more than necessary, anyways. And I wondered that about Super Shredder. He was awesome in Tmnt2, but barely visible - and the other 'iterations' of him in more recent movies is more like a 3D can-opener. If I dig up any images of SS I'll let you know! **Just out of curiosity, how many images does one typically need to make a LoRa like you do?
@Picky2011670 I know what you mean, it's a shame there wasn't a decent fight with Super Shredder in TMNT 2. That would definitely be helpful for sure, I prefer at least 50 - 100 images for my LoRAs, feels like my training method works best when there's enough variety, if I use less than 50 with my method the model seems to end up being far less flexible.
@ArchAngelAries Go it! I've found a pretty large resource for an accurate Super Shredder....based on photos of a really accurate/lifelike model/collectible of him that I just grabbed a folder of 58 images of - and there's more, I just wanted to look...if that's something you're interested in. Pretty cool to finally get a good look at him!
@Picky2011670 Heck yeah! If you wanna shoot 'em over to me, send me what you've got and I'll start captioning and training asap! [email protected] is my artist work email
@ArchAngelAries Emailinggggg....now
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.