Runway adds editable image layers with Seedream 5.0

Runway has added Layers, a tool that decomposes a flat image into movable, transparent elements. After uploading a poster, product shot, or social graphic, users can separate subjects, backgrounds, text, and other detected regions without rebuilding the composition by hand.

The underlying model is ByteDance’s Seedream 5.0 Layers, which Runway hosts inside Tool mode. The integration gives Runway users a direct path from a finished image to an editable layer stack that can feed into design, image, and video workflows.

From flat image to layer stack

1. Open Tool mode from the left navigation bar in the Runway app.
2. Select the Image tab, open the model menu, and choose Seedream 5.0 Layers.
3. Upload one image and optionally describe which elements should become separate layers.
4. Click Generate to create the layer stack.
5. Select elements on the canvas or in the layer list to move, resize, rotate, hide, delete, or reorder them.
6. Download all generated layers as transparent PNG files from the Recents folder.

The model can infer a layer structure without a prompt. A specific instruction helps when the composition contains small objects, overlapping elements, or several items that should remain separate. Runway’s model guide recommends naming the elements that need their own layers.

One generation costs 18 credits

Runway charges credits when the user clicks Generate. Reordering, transforming, hiding, and downloading the resulting layers does not consume additional credits, so repeated layout changes can use the same generated stack.

Recovering structure from finished assets

Teams often receive exported PNG or JPEG files after the source PSD, Figma document, or design project has disappeared. Updating those assets usually requires manual masking, object removal, and background repair. Layer decomposition automates much of that recovery work by identifying meaningful regions and placing each one on a separate transparent canvas.

Removing a foreground object also exposes pixels that were absent from the original image. The model reconstructs those hidden areas so the background can function as an independent layer. That reconstruction is generated content and may require cleanup around complex edges, shadows, reflections, hair, or overlapping objects.

Useful jobs for recovered layers

• Remove rasterized headline artwork and add replacement typography.
• Extract a product from a photo and place it on another background.
• Adapt an existing ad layout for multiple products or aspect ratios.
• Separate character parts for animation and game-production pipelines.
• Recover usable components from client assets whose source files are missing.
• Move isolated subjects into Runway video-generation workflows.

Where the seams can show

Seedream produces a best-effort interpretation of the image rather than its original document structure. Closely overlapping objects may be merged, small details may disappear, and reconstructed backgrounds may contain artifacts. Another generation with a more precise prompt can produce a different split, although it also incurs another 18-credit charge.

Exported elements remain raster images. Text may appear on its own transparent layer, but its font, wording, spacing, and vector properties do not become editable. Users who need to change the copy must remove the extracted text and recreate it in a design tool.

Runway exports transparent PNG layers rather than a native PSD with typography, vectors, masks, and adjustment layers intact. Teams can import the PNG files into Photoshop, Figma, or another editor, then rebuild any document-specific controls they need.

Image decomposition moves into Runway

Standalone services including IntoLayers, Playyy, and PixMind already offer image-to-layer conversion, with some supporting direct PSD export. Runway’s implementation keeps decomposition beside its image editing, Agent, and video-generation tools, reducing the number of exports required before an isolated element can enter another workflow.

The feature expands practical control over flattened assets while leaving copyright and license terms unchanged. A generated layer stack can simplify editing, repurposing, and production handoffs, but it does not grant permission to modify or redistribute material the user does not own.