Creates videos in a modern HD anime style.
The primary trigger is "An1meStyl3," but it is recommended to also include the tag "AnimeStyle" alongside it and add "realistic, photograph, 3d animation" to your negative prompt. Additionally, it helps to refer to your subjects as anime characters somewhere in the prompt; for example, instead of "an old man," you should refer to him as "an old anime man" at least once.
This is intended to be a general-use anime LoRA, so it was trained from stills and clips of a diverse set of shows that have a common general art style, rather than fixating on a particular show or creator, as to avoid directly producing results that resemble any particular illustrator's signature style or creative choices.
California AB 2013 Training Data Disclosure
This LoRA was fine-tuned using visual data consisting primarily of still images sourced from animated television series, along with a limited amount of publicly available fan-created renderings and AI-generated images. The training data includes copyrighted material owned by third parties, including animation studios, production committees, distributors, and individual artists. No training data was licensed or purchased. This LoRA is provided for non-commercial use only under the terms of its distribution.
The dataset for Version 1 consists of over 1,500 images collected from publicly accessible sources across more than a dozen animated series between approximately 2000 and 2025. Image data was processed through standard resizing, cropping, normalization, and labeling steps. Synthetic images were included as part of the training dataset. It also incorporates approximately 200–250 video clips sourced from several different animated series. Clips were used as video sequences to support motion and temporal consistency.
The dataset for Version 2 consists of over 1,500 images and over 750 clips collected from publicly accessible sources across more than a dozen animated series between approximately 2000 and 2026. Data was processed through standard resizing, cropping, normalization, and labeling steps. Synthetic data was included as part of the training dataset. Clips were used as video sequences to support motion, lip sync, voice training, and temporal consistency.
This model is intended for non-commercial, experimental, and educational use. Generated outputs may reflect copyrighted visual styles or themes associated with the underlying training data. Users are responsible for ensuring compliance with applicable copyright law, other intellectual property laws, and all other applicable laws.
Description
V2 Release
Trained visual style focused on appearance and general movement.
Significantly more clip-focused training
Trained on clips with audio, helping lip sync and adherence to traditional anime voice acting cadence
FAQ
Comments (9)
just in time for 2.3
Yeah, that was really unfortunate timing lol. I think LTX-2 is going to have some staying power, because it is barely trainable on 24 GB of VRAM, which is most people's maximum, so I anticipate we're not going to see nearly as many community-sourced models for 2.3 for a while.
Why are developers still insisting on training in LTX2?... LTX 2.3 is FAR SUPERIOR in output quality, realism, movement, body physics, consistency, adherence to the prompt; it doesn't make sense that they haven't left the previous version behind yet.
I don't really see "insistence" so far. 2.3's first lora came out 4 days ago, model released 5. People might have been training on 2 for days or longer and to me it seems they are just dumping the last of their work.
And since model's been out less than a week, it would make perfect sense if people haven't left it. Did everyone transition from wan 2.1 to 2.2 instantly the moment it came out? Absolutely not. Even if the base is better, the platform (comfy), settings, workflows, loras all take time to adjust (recalibrate / retrain / extend code). Especially loras - some may not even get a substitute at all. It's not even the same architecture as was the case with 2.1 / 2.2 low, though there's a good deal of similarity and no binary model split so they might be partly functional, I dunno.
You're welcome to complain if it makes you feel better, but I think what is more effective at getting people to transition to competition is showing results and better support for the things they care for in particular. Show, don't tell.
@firemanbrakeneck This is exactly correct. I haven't tried 2.3 yet, but I know that LTX-2 is significantly more cumbersome on GPUs than things like WAN 2.2 or SDXL. This model trained at a rate of, at-most, 100 steps an hour, often even slower, so this was weeks in the making and finally finished this week.
In regards to OP 's original question, I assume LTX-2 will hang on a bit, because it is a 14B model and 2.3 is a 22B model, so 2.3 is likely going to be even more difficult to train on consumer hardware (both longer training times and higher base hardware requirements), so the pool of people capable of training on it will be even smaller.
It's also noteworthy that Ostris AI Toolkit is one of the biggest training platforms, and 2.3 is so new that it's not fully supported on that platform in the same way that LTX-2 is.
Chill down, the new version has been out like for just a couple of days?
You should chill
bruh
@ReltivlyObjectv LTX 2.0 is 19B while 2.3 is 22B. But you're right, there shall be a delay in loras for 2.3, afaik I can only see a lot goon lora updates for 2.3, the normal sfw ones shall take its time lol.