This is just the same dataset but trained for LTX 2.3 but I increased rank from 32 to 64 (thus the 1.2GB file size). The result is really great. We get 6 character likeness and voices which sound and look great. For one lora that's a lot of information! And the characters don't bleed into eachother.
Edit: All issues have been fixed in v1.1. Give it a try!
Add "subtitle text" to negatives if you get them, you should not get them otherwise.
Robert, Invisigal, Blonde Blazer (and without powers too), Prism, Malevola, Chase
Punchup is trained but not enough data, haven't tested. (char_punchup)
Royd is in the data as char_roy but also not much data.
Other characters are in the dataset but not tagged. Shroud is not in the data at all.v1.0
Style Trigger:
DISPSTYLE
Invisigal:
char_invisi, a woman with short dark hair accented by a purple streak, wearing a purple jacket over a dark top and distressed black jeans
Blonde Blazer:
char_bb is a woman with long blonde hair and a blue mask, wearing a blue and yellow superhero suit with a yellow cape and a red diamond-shaped gem on her chest.
(no costume, powerless)
char_bb has long, wavy dark brown hair and wears a strapless evening dress with long dark blue opera gloves
Robert Robertson:
char_rr has short brown hair and wears a light blue button-down shirt with the sleeves rolled up to his elbows and dark trousers.
Prism:
char_prism, She is a black woman with straight hair split down the middle, teal on the viewer's left and magenta on the right. She wears large, rectangular, reflective teal visors, blue lipstick, and a gold ring necklace. Her attire consists of a black sleeveless top, thick gold bands on her upper arms, a long magenta glove on her left arm, and a teal glove on her right hand
Malevola:
char_malev a woman with red skin, long black horns, and glowing yellow eyes, a long red tail,, is dressed in a white tank top and denim shorts. she has a sword attached to her back.
Chase:
char_chase, an older Black man with white hair in locs and a mustache, wears a yellow sweater over a light blue collared shirt and black trousers with a gold buckle belt. A pair of glasses hangs from the collar of his sweater.
Description
FAQ
Comments (13)
This has multiple characters under one lora, showing what LTX is capable of! Let me know how it works. Sometimes roberts beard is baked in, might fix it in next version. I wanna move onto a new lora now.
I've been working with a fine-tune level train of LTX and I gotta say that lora rank is a big thing when it comes to having diversified data. Only relevant to this because I notice an attempt to get all the characters into it-but I've used this a lot and voice-wise its still just Mechaman and BB, and my training runs end up being linear like that too. The big thing with getting that consistent diverse stuff might be increased rank, like 128,256+ etc, however big and fully trained it needs to be. rank 16/32/64 mostly seem only capable of single motion groups, style, or 2 very different voices. The one I use at 512 rank (9gig lol) is a train that essentially fine-tuned the entire model with all different sounds, positions, voices, styles, etc. Yeah file size is ridiculous, but you'd be surprised with what you can do with that increased rank. Can't guarantee it but with that rank and a good balance of weighting you'd get a lot more out of it especially other characters showing up, the linearity mostly seems to come from lower ranks. That should be helpful for anyone else trying to train on very diverse data as well.
@tenstrip Yeah I dont think a 9gig lora will work so well, it may be too much. I also have been thinking about increasing rank and playing with LR something I haven't done before. If I do another version, I probably more characters and try increasing the rank. And BB and Robert's voice has the most data, it probably comes out more. But invisigirl and prism's voice is definetely there, but probably doesn't come out when BB is prompted much. BB and Robert defintely become the default male/female voices too yeah. Thank you for the advice, I will give it a try!
Really great lora! May i ask how much time total it took to train with a 5090?
I think it was around 2-3 s/it , and maybe around 1.5k steps per hour. So maybe 12-15 hours max to train it.
@tazmannner379 Thanks!
@tazmannner379 Wow, that's a long ass time. Can you share your methods and settings for Mitsubi? I'd like to try this on my RTX 6000
Hey, pardon my stupid question but, is there any chance this would work with some characters of a different style ? or would it draw their appearance back to something that is similar to the style of the dataset ?
Do you mean use this with another style lora or prompting? I know someone who did another show lora and it made the characters in the show style and kept the appearance and voices. The other way around worked for an IRL character lora I had, it put it in the show style, sometimes have to play with the strengths. And just a few tries only, so I dont know for sure.
@tazmannner379 Hey thanks for your answer! I mean, I am more interested in the way the characters move than in their appearance in this lora. So lets say if I input my own character references as a start image (different style and design), do you think it will keep their appearance ?
I'm just enjoying watching the preview clips, if this were made into an animated series I'd honestly watch it!
Hollywood is fucked. Anime is Fucked. In 5 years, all of them, right up the ass.
omg nice