(LTX-2.3 Lora) - Camera Zoom Out (I2V)
A lora to create videos where the camera zooms out relative to the subject. I2V use recommended.
Can be combined with my "Pose Helper" lora to improve animation:
https://civarchive.com/models/2402694/ltx-23-lora-pose-helper-i2v
Use "z00m047" as a prompt word
Use Lora Strength around 0.7-1.0. (if combined, put ' d3m0p0s3' lora between 0.3-0.5)
Preferably use with I2V on LTX-2.3 DEV (FP8).
May require 2-5 generations to get the best result (effect and overall speed is random)
Trained on 960x960 (24fps) and 704x704 (48fps) videos
Feel free to post your results, though keep it decent.
Read more below for prompt examples (recommended)
Known issues: When not used with the 'Pose helper' or other animation loras, may result in static subjects. Two to five generations may be required to get the desired effect. Consider specifying the exact number of subjects.
For prompting, use keyword "z00m047", followed by a sentence describing:
What is visible in the start frame + "zooms out revealing" + what should be visible in the end frame +description of the initial subject ' barely visible in the distance'
I2V Prompt example:
z00m047. a clean video of a closeup of a SINGLE blonde woman is posing and moving in a neon street. The camera zooms out, revealing a large cyberpunk like neon city street with shops, the SINGLE woman is barely visible in the distanceWas trained on captions like:
z00m047, the subject and surroundings are static.
Start Frame: closeup of a woman with gray eyes.
Action: The camera zooms out revealing the woman is standing up outside wearing a necklace and a light pink top and skirt. A large stone pillar and a church is revealed. z00m047, the subject and surroundings are static.
Start Frame: Closeup of a grassy hills and some trees
Action: The camera zooms out, revealing the landscape from a distance showing more hills, trees, mountains and a riverDescription
FAQ
Comments (3)
i have a question, and nothing against this lora, but i noticed that lightricks themself released the same kind of lora's when they released ltx2.0..
why lora's for camera movement when its so easy to prompt?
Hey!
In my trials I didn't get the type of zoom out I wanted from the base model, especially 2.0 . Of course it's possible you have gotten better results prompting.
I also used some fancy way of creating the dataset which should on average maintain high quality when zooming out in certain settings. I personally prefer using this lora together with the pose animate lora to make videos of subjects posing with some zoom out happening.
No need to use this if you got competent results of course.
@JonXL ok, no worries, i was just curious, my assumption was using these camera lora's to fight against other loras being stuck in certain angles/zooms. because that's the only time i struggle with prompting camera movement.