Local I2V finally feels less like image wiggle and more like shot direction with LTX Director
I’ve been experimenting with LTX Director for LTX 2.3, and I think this workflow has a lot of potential.
Local I2V often feels like “make this one image wiggle”: same angle, small motion, maybe blinking or hair movement. But with LTX Director, using multiple images of the same character as key poses/camera angles inside one timeline feels much closer to shot direction or a tiny MV editor.
For this test, I used three source images of the same character with the same outfit/background, but different poses and camera angles. I included the original three images as well, so you can see what LTX Director was working from.
I also added a custom K-pop-style audio track with Custom Audio ON.
After a lot of tuning, it was able to handle:
- multi-image I2V
- smooth pose changes
- camera and face movement between poses
- cute performance gestures
- custom audio timing
- usable lip-sync
It’s still experimental. Hands can break, identity can drift, and transitions need careful prompting. But when the input images are consistent — same character, outfit, background, and style — it becomes much more dynamic than normal single-image I2V.
The most useful prompt idea for me was to treat the images as key poses of the same character, not separate people:
“Treat all images as the same character in different poses and camera angles. Preserve the same face, hairstyle, outfit, and background throughout. Move smoothly between the poses as one continuous close-up performance. Natural lip-sync to the custom audio vocals, clear visible mouth movement, soft blinking, small head tilts, cute gestures, subtle shoulder sway, light hair motion.”
This still needs more testing, but I think LTX Director could be really useful for AI idol clips, character PVs, surreal mascot videos, short music videos, and anything where local video generation needs more than one static angle