Id lora with reference image and audio
Hello, I'm trying to use the LTX-2.3_-_I2V_T2V_Basic_ID-Lora_reference_audio workflow. I want to maintain the exact same character from my reference image throughout the generated video. Is there a way to do that?
The ID-Lora is for the voice (only).
Unfortunately all video models do lose a bit of facial consistency over time.
There is a new lora that might help, I didnt try it yet, but seems to aim at helping character consistency
https://huggingface.co/elix3r/LTX-2.3-22b-AV-LoRA-talking-head
I'll try that lora soon ;-)
Other remedies can be to start with a close up face portrait as first frame, and also use the latent reference image node to guide (but i think that node is already in use in the workflow, i'll check)
a quick test run with the talking head lora. Didnt do a 1-1 comparison, but perhaps it helps a bit
oh great i'll try it out thanks alot!