Luma has launched Luma Scenes, a multi-shot workflow for AI-generated video that is now generally available.
The system creates a sequenced storyboard of editable keyframes before rendering the final video, shifting the process away from generating a complete sequence from a single prompt at the outset.
The approach aims to address a common problem in AI video production: users can spend time and money rendering an entire sequence before discovering that one section does not work.
With Scenes, users can regenerate individual shots, adjust timing across a sequence and approve the storyboard before rendering begins. That allows creators to change parts of a project without restarting the entire production.
The tool can generate a full storyboard in a single pass while maintaining visual consistency across characters, environments, products and style. It also adds timeline-level timing controls before the final render stage.
Workflow shift
The launch reflects a broader move in generative video software towards workflow and editing tools, rather than a sole focus on the underlying model. In practical terms, companies are trying to make AI video systems fit more closely with established production processes.
Luma's setup uses its Uni-1 model for planning, while Ray 3.2 and Seedance 2 handle rendering. That split suggests the company is separating planning from final image generation to give users more control earlier in the process.
Storyboard-first systems are familiar in conventional film and advertising production, where sequences are reviewed and adjusted before expensive final work begins. By applying a similar structure to AI video generation, Luma is positioning the product around predictability and revision control.
Editing control
One of the main challenges for AI video users has been continuity across multiple shots. Even when a single generated clip looks convincing, keeping the same character, setting or product appearance across a wider sequence can be difficult, particularly when separate clips must be produced and stitched together.
Scenes is designed to preserve that consistency across the storyboard while still allowing individual shots to be changed. That could matter for commercial users producing advertisements, social video, product demonstrations or branded content, where inconsistencies between shots can undermine the final output.
The ability to adjust timing at the timeline level before rendering also points to a more editing-led workflow. Instead of accepting the pacing that emerges from a generated output, users can shape the sequence in advance.
Market context
The AI video market has become increasingly crowded as developers compete not only on image quality and motion realism, but also on usability. Early tools often centred on text prompts and model performance, but commercial users have pushed for systems that reduce waste, support revisions and fit budget constraints.
That pressure has led vendors to focus on practical controls that make generated video more manageable in production environments. Storyboarding, selective regeneration and pre-render approval are examples of features aimed at narrowing the gap between experimental outputs and repeatable production work.
For Luma, the release adds a layer of process management to its existing video generation technology rather than simply introducing another rendering model. The emphasis is on helping users make decisions earlier in the creative cycle, before committing computing resources to a full output.
Scenes enters the market at a time when generative video companies are being judged less on novelty alone and more on whether their tools can support repeatable work for creators and businesses. The proposition is straightforward: review the sequence first, change what does not work, and render only when the project is approved.