MAJOR UPDATE PLEASE REREAD DESCRIPTION:
Wow, this is a big update folks, I almost uploaded the Illustrious version of this as it's own full release, the pony version of this really just emulated the style, but with everything I've learned, I decided to make this do just about everything the actual artist can.
On top of being more accurate to the style this lora has been trained to produce: Transformation Sequences, Mid-Transformation single images, Post-Transformation single images, Before and After Images, Breast expansion, Ass expansion, Penis Expansion, Muscle Expansion, Full Color images, Partially colored images, Monochrome Images, and a series of transformation types.
Transformation types/ their tags: "Species transformation", "Gender Transformation", "Bimbofication", "Inanimate Transformation", "Living Inflatable" transformation
For all useful tags please check my sample images for all the useful tags, I did my best to use them all. Be warned all images used Hires Fix, several images used Forge Couple for better segmentation of ideas, but it shouldn't be necessary I tested all sequences and transformations without Forge Couple.
ORIGINAL DESCRIPTION:
Yet another style and a request this time. It's Sutibaru an expansion and transformation artist. Pretty happy with how this one came out. Look at showcase images for example tags.
Description
This version is way different than the pony version to the point I almost considered uploading it separately rather than adding it to this one. Please read the main description.
FAQ
Comments (7)
Can this do feral TFs? Ive tried and ive managed to get it KINDA working with a transformation sequence for a feral TF but its very, very hit and miss depending on seed (and presumably, checkpoint. Im going to test that next) Many of the timages are basically just a sequence of ferals, where as another might be a human on their knees and the next one a feral.
I didnt notice any mention of feral in the tags of the lora so im unsure if the artist does feral TFs.
Also what resolution do you suggest for sequences? I usually use the example images as a guide for what resolution a lora works best in especially for sequences, but as you mentioned you used hires fix, i can only guess.
Thanks!
No feral transformations are in the dataset, if Sutibaru has any I didn't use them. For resolution for sequences depending on pose and number of steps you're hoping for I used 1216 x 832, 1344 x 768, or 1536 x 640. Details will be rough at the latter two without some kind of post processing whether through hi-res fix, face fix, or adetailer/udetailer. Hope this helps.
Oh and I recommend some use of something like regional promoter or forge couple to really fine tune your transformation'steps as the model is often unsure what should stay from one stage to the next. Also panel # tags get the number of steps right and keep the background similar such as 3koma or 4koma.
@nemesis2k7 Thanks!
I did actually manage to get feral TF working (kinda) a bit better even without forge couple. so its interesting to hear that there wasnt anything in the dataset to do with feral TFs.
I went through the tags of the lora itself after opening it in notepad and i used some of those and it seemed to do a slightly better job than it did before.
The main thing it seems to struggle with is the size of the person. On the left might be a human person when i specify human, but the next one along it'll be a human that is mid-TF, on all fours like a feral, and yet still be as tall as the human is, so they're just "bigger". That seems to be the biggest hurdle right now.
That being said though, i only tried with dog/wolf feral TFs so far so i have no idea how well it will do with something like dragons or horses in regards to feral TFs.
Thanks for the resolution suggestions! I did guess as to what resolution the "base" images were you provided as an example, guessing that you did a 50% upscale, and it appears i was right because so far i've been doing 1536 x 640 so far, but you're right that some things (especially the humans face) can be very rough, even WITH Adetailer. I'll be sure to try the other resolutions you suggested for the sequences!
And i will certainly try Forge couple! I've used it before on other TF sequence LORAs, its usually the only way i can even get them to work properly the way i want but i was intrigued by your mention of your Lora being able to do it even without it.
Usually in forge couple i'll have a "global" area that contains the lora and the general thematic tags like transformation, feral TF and so on. and then a "start" and "Finish" area, with the start describing the human and the end describing the finished result of the TF with tags. Do you suggest a different approach? Perhaps by having a region for each step instead of/alongside the "global" region?
I did try that specific approach with another lora in the past but i found that it can limit the models ability to vary a lot of stuff between generations.
I will see about posting the best examples i've been able to generate so far once i've done some more tweaking!
@vendilconius My method is pretty similar with a start and finish, but this set does contain a "mid transformation" tag that my be helpful to apply to the middle sections, though I have no idea how it will work with feral tf as it was all anthro, rubber, or bimbo tfs it was used for. But combining it with things like "all fours" may help you get what you're looking for a bit more consistently. I'm curious to see your final results when you have some you want to post to the model.
@nemesis2k7 I just posted some! These ones are WITHOUT forge couple/regional prompter but that will certainly be the next thing i try, as i imagine thats going to improve the results significantly. I will post those when im done tinkering as well!
@vendilconius Wow, those look really good, thanks for posting them. I love seeing the kind of stuff people can come up with using the stuff I train. I gifted buzz on my two favorite.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.