Luma launches Modify Video in Dream Machine

Luma AI announced Modify Video, a video-to-video editing feature now available in Dream Machine on its Ray 2 model. Clips can be up to 10 seconds long.

The tool reads pose, facial expression, lip sync, and scene structure from an uploaded clip, then uses those signals to decide what to keep and what to regenerate. Luma groups its uses into three categories:

• Motion and performance capture: pulls full-body, facial, or lip-sync motion from a clip to drive new characters, props, or camera paths. One example is moving an actor's take onto a CG creature.
• Restyling and world swapping: changes the look of a scene without altering its motion or framing, such as turning day into night or a garage into a spaceship.
• Isolated element edits: changes one element, such as wardrobe, a face, a prop, or the sky, without green screen or manual tracking.

Users choose one of three presets that set how far the output departs from the source. Adhere mostly retextures, Flex balances fidelity against model creativity, and Reimagine allows a full reinterpretation of the scene, including turning human performers into non-humanoid creatures. A text prompt is optional. Users can guide results with visual references or a stylized first frame, and can generate several variants from the same base motion. Output supports native resolutions including 16:9 at 720p.

Luma says blind evaluations with external raters found Modify Video beat Runway's V2V at preserving motion and facial animation and at temporal consistency. The company did not publish the number of raters, the test set, or any scores. It also did not disclose pricing.