Full Video & Voice Training!

The full LTX2.3 LoRA for Anna (Frozen) is finally here! This LoRA was trained using ai-toolkit against a large set of short sample video clips for both video & audio! So that means that you can simply refer to your Elsa character by the trigger word Fr0z3nAnna naturally for entire videos, including speaking!

Start with close up on Fr0z3nAnna, she is wearing a dark green dress dancing by herself in a large elegant royal ballroom. She stops dancing, she looks at the camera and blushes with an emerrased expression. She looks down at her feet and says softly "Oh hi, I didn't know anybody else was here". She looks back up at the camera slowly and smiles, then says "Do you want to dance with me?"

The training was intentionally done for the most flexibility with the characters, so sometimes they might come out looking more like ... real humans haha. That might be what you want! And that's the point! The character LoRA should allow you to specify and override just about anything.

If you want to better ensure the original animation style, a simple prefix is all you need, and another LoRA works pretty well along side these characters (LTX 2.3 - Pixar CGI Toon Style)

Just prefix or specify something like this:

P1x4r 3D animated style.

I just ended up having this as a static prefix for all of my prompts, they all worked out pretty well for fidelity. The full example combining these would be as simple as:

P1x4r 3D animated style. Start with close up on Fr0z3nAnna, she is wearing a dark green dress dancing by herself in a large elegant royal ballroom. She stops dancing, she looks at the camera and blushes with an emerrased expression. She looks down at her feet and says softly "Oh hi, I didn't know anybody else was here". She looks back up at the camera slowly and smiles, then says "Do you want to dance with me?"

Finally, both Fr0z3nElsa & Fr0z3nAnna LoRA models were trained using AI Toolkit and specifically the new experimental feature: Differential Output Preservation (shoutout to Ostris!) - Which uniquely allows for these characters to be combined in scenes easily without warping or overwriting each other (too much). Some amount of that is unavoidable, but the better you separate them in the prompt and define explicit details for each, the better the results will be!

Simple example...

P1x4r 3D animated style. 
On the left side: Fr0z3nElsa has braided blonde hair wearing a blue swimsuit.
On the right side: Fr0z3nAnna has braided auburn red hair put up in a bun wearing a red swimsuit.

Scene 1: Fr0z3nElsa and Fr0z3nAnna are sitting on wooden folding beach chairs at a warm public beach with a tropical island background.

Scene 2: Fr0z3nAnna says to Fr0z3nElsa, "Isn't this better than freezing our butts off in Arendelle?" while both girls let out a loud relaxing sigh as they close their eyes and lean back in their chairs with a smile.

Scene 3: Fr0z3nElsa replies to Fr0z3nAnna, "hehe, yeah ... ummm ... my bad about all that frozen stuff".

Most of my testing was done at standard 1.0 strength, for best fidelity would expect 0.8 - 1.0, but play around and see what works best for you!

These have been tested with LTX2.3-10Eros Checkpoint and actually work pretty well, but as usual, the best character fidelity will be from using the full LTX2.3-DEV checkpoint.

This is my first serious attempt at a real LoRA, please share your generations here (Add Post Below) and provide any feedback you may have <3

** PLEASE CREDIT ME IF YOU DO COOL STUFF WITH THESE!

This LoRA is created as part of a fan project for research purposes only and is not intended for commercial use. It is based on the Movies "Frozen" which is protected by copyright. Users utilize the model at their own risk. Users are obligated to comply with copyright laws and applicable regulations. The model has been developed for non-commercial purposes, and it is not my intention to infringe on any copyright. I assume no responsibility for any damages or legal consequences arising from the use of the model