Move the camera, keep everything else (mostly).
Give it an image and a camera position. It renders the same subject, scene and style from the new angle. Orbit all the way around, raise or lower the camera, or pull in for a close-up. This is an edit LoRA for Qwen Image 2.1, so it always needs a source image.
Every prompt follows the same simple pattern:
<qma> {azimuth} {elevation} {distance}
Azimuth is where the camera orbits to:
front view
front-right quarter view
right side view
back-right quarter view
back view
back-left quarter view
left side view
front-left quarter view
Elevation is how high the camera sits:
low-angle shot
eye-level shot
elevated shot
high-angle shot
Distance is how far away it is:
close-up
medium shot
wide shot
Pick one of each, in that order. For example:
<qma> right side view eye-level shot medium shot
<qma> back view elevated shot wide shot
<qma> front-left quarter view low-angle shot close-up
Directions are relative to the original camera, not the subject. "Right side view" means the camera swings 90 degrees to its own right around the subject. "Back view" takes it all the way around behind.
Rather than typing prompts, you can drag a camera around to build them with the QMA Multi-Angle LoRA Node by CoreyCorza.
Use LoRA strength 1.0, 30 steps, CFG 3, the euler sampler and the simple scheduler. Connect your source image as the reference image. Output around 1 MP gives the best detail. 768 is great for quick previews.
Trained on a mix of photographs and Blender renders with AI-Toolkit.
Each pair shows the same subject and scene from two camera positions, covering all 96 combinations of angle, height and distance. Rank 64 LoRA trained for 3,000 steps at 512 and 1024 resolution.