Luma Labs brings prompt-based video editing to Dream Machine

Fixing a mistake in post-production usually means tedious frame-by-frame rotoscoping or re-rendering an entire scene. When it came to generative AI video, tweaking a single detail was even worse: you had to re-roll the entire generation prompt and hope the composition didn't fall apart.

Luma Labs introduced Modify with Instructions inside Dream Machine. The feature is now available to all paying subscribers, letting creators change video elements, swap outfits, or overhaul virtual backgrounds using everyday text prompts.

Directing video edits in plain English

The workflow avoids complicated tracking masks. You simply freeze a frame where the target element is clearly visible and type a straightforward edit prompt, such as "change the red dress to a blue suit" or "replace the coffee cup with a wine glass."

Dream Machine parses the intent and extends that modification coherently across the full clip. Luma highlighted several practical workflows, including VFX object removal, virtual weather, digital product placement in ads, set extensions, and rapid wardrobe restyling for filmmaking.

Three strength levels and initial guardrails

To balance fidelity and creative freedom, Luma built an intensity slider with three presets: Adhere, Flex, and Reimagine. Keeping the dial low preserves subtle lighting and textures, while dialing it up lets the model dream bigger and reinvent the scene.

That said, there are real constraints to keep in mind. For the instruction to land properly, the target element must appear in the very first frame of the video clip. If the subject enters midway through, the prompt is likely to miss.

Aspect ratios are also strictly monitored. The tool natively supports 9:16, 3:4, 1:1, 4:3, 16:9, and 21:9. Uploading footage outside these presets will trigger automatic cropping and reformatting, which can degrade output quality.

Why it matters

For filmmakers, commercial animators, and solo creators, prompt-guided editing drastically cuts down the pain of iterative revisions. Instead of treating generative video as an unpredictable slot machine, users now have a native lever to tweak specific visual assets without throwing out the entire shot.

Across the AI landscape, the battleground is shifting from basic video generation to granular post-production control. Models can already dream up high-definition scenes, but enterprise clients need deterministic editing. By integrating prompt-based object swapping straight into Dream Machine, Luma puts direct pressure on competitors like Runway and Pika to make video manipulation just as accessible.

We think Luma's approach is smart because it brings intuitive VFX directly to non-technical storytellers. However, the requirement that targets must sit in the first frame underscores the persistent challenge of temporal consistency across AI video models. It is a fantastic asset for rapid prototyping today, though industrial-grade precision still has a bit of catching up to do.