@AIWarper i also played around with a workflow that runs a dirty audio latent when it can’t get the prosody right. that could work well for what you’re trying to do.
@AIWarper training an ltx lora actually works quite well but requires a lot of generations/editing. and perhaps overkill.
i def like it better for audio than rvc and tts. this is brand new lyrics but could easily be different person on original lyrics.
@Jecky666@ostrisai hmmm, might be a setting issue? i accidentally just trained the same, extremely difficult likeness/voice on h3, ltx 2.3 and ltx 2.5 (midway) over 72 hours and ltx 2.5 is blowing away the other two.
Packie runs are a Public Service 😤 JFK celebrates July 4th with his best impression of Independence Day's President Whitmore. Built primarily in @ltx_io , JFK trained with @ostrisai AI-Toolkit.
@OdinLovis thats sweet. did u train it w seedance gens or something from gaming industry?
i built my set in blender but so far have just been using it to generate consistent first frames for ltx 2.3. works great for that.