Luma Layers adds editable scene elements to Luma Agents

• Luma shipped Layers inside Luma Agents: creators can ask the agent to extract layers from generated or uploaded images, then swap a product, restyle a background, or rewrite a headline without rebuilding the whole frame, according to LumaLabsAI's launch post.
• The mechanic is a transparent layer stack: Luma's learning guide says uploaded posters and ads can split into elements like background, headline, character, logo, CTA, product, shadow, text block, and backdrop, while LumaLabsAI frames the job as changing one element while preserving everything else.
• The strongest demos are element inserts: DreamLabLA's octopus tutorial adds a creature into an office scene, and DreamLabLA's weapon demo adds one prop while keeping the room and lighting intact.
• Layers is already feeding video workflows: the system makes it easier to build Modify-video keyframes with Ray 3.2 in mrjonfinger's Layers + Ray 3.2 test.
• Developers get the same model family through Luma Agents API, where Luma's model docs list uni-1 and uni-1-max for image editing, while LumaLabsAI's link post points users into the Creative Agents product surface.

Luma's Introducing Layers post says Uni-1 can generate new images as independent layers and decompose uploaded assets into layers. The intro guide gets more production-specific: a product ad can become product, shadow, CTA, text block, and backdrop. The editing guide has the Photoshop-ish detail creative teams will care about: layer edits create new variations while the original layer is hidden, grouped, and still available.

Layers starts with a command: extract the parts of an image into editable elements. Luma says this works on images generated in Luma and on uploaded assets.

Luma's intro guide breaks the output into production-friendly pieces:

• Poster: background, headline, character, logo, graphic treatment.
• Product ad: product, shadow, CTA, text block, backdrop.
• Each extracted element: transparent background, editable layer stack.
• Layer actions: move, reorder, hide, modify, replace, remix, remake, remove.

The official pitch is brutally practical: stop losing approved composition, lighting, and layout because one element needs a change.

The interaction model is chat first. Luma's launch copy says creators can extract layers, then keep prompting the agent to swap the product, restyle the background, or rewrite the headline.

The Creative Agents page frames Luma Agents as a shared-context workspace where agents plan, generate, iterate, and refine across creative stages. Layers gives that agent loop a precise object handle instead of asking the model to reinterpret the whole image on every revision.

DreamLabLA's tutorial uses Layers to add an octopus into an office scene without touching the rest of the frame. The post describes it as Luma's precision-editing tool powered by Uni-1.

The workflow detail is scoped insertion: one new element enters a live-action-looking shot while the desk, lighting, monitor, and actor remain the continuity anchor.

Another DreamLabLA clip credits Blake Johnson and describes the edit as one added weapon layer. The post says Luma did not regenerate the room, the lighting, the desk, or the actor's reaction.

This is the creative sell in one gag: absurd change, stable scene.

mrjonfinger tied Layers to video preproduction, saying the system makes it easier to quickly build keyframes for Modify video. A follow-up from mrjonfinger's tutorial post says he was testing a tutorial format for the workflow.

Luma's model docs list Ray 3.2 as the Agents video model for video, video_edit, and video_reframe. The same docs put uni-1 and uni-1-max on image generation and image_edit, with the same endpoint envelope and submit, poll, download flow across media types.

Luma's Agents quickstart exposes the developer workflow as three calls: submit a job with POST /v1/generations, poll GET /v1/generations/{generation_id}, then download images or videos from presigned URLs once complete.

The pricing docs list two API buying modes:

• Pay-as-you-go: no minimum spend, no contract.
• Provisioned Throughput: guaranteed capacity and latency SLA, billed monthly.

Per-image API pricing starts at $0.0404 for uni-1 text-to-image and $0.0434 for uni-1 image edit. uni-1-max is listed at $0.1000 for text-to-image and $0.1030 for image edit.