Raster used to be where images lived. Now it's where they get made, changed, and shipped, by your team and by the agents working beside it. Same libraries, same URLs, same keys. Everything around them is new.
A new design
The home screen shows each library as a card of its newest photos, with its counts on the card. The image sidebar is built around variants: one row each, with metadata, description, and tags a drill-in away. Edit tools stack in one scrollable view with Save and Discard pinned.
Photos paint a placeholder in their own colors before the pixels land. Uploads appear the moment they're ready. Trash and restore are instant. Switching libraries keeps the current photos on screen until the next set arrives. Transparency renders on a checkerboard, SVGs stay vectors, and videos and PDFs get the same redesigned panel. Video tiles carry a poster and duration, and play on hover.
Variants, and promoting one
An asset holds an original plus its variants. Upload one, in the sidebar or over the API with parentId, and it sits beside the original to compare, download, and copy. Photos take image variants; videos take video variants.
Set as Default replaces the pixels an asset serves at its existing URL. Every site, CMS, and plugin already embedding that URL shows the new image with nothing to update. The old default becomes a variant, so you can promote it back. Same ID, same URL, same tags. Only the image changes.
Saved variants remember their adjustments. Reopen one and every slider is where you left it; pull one back and the original pixels return.
AI editing
AI editing is rolling out to Pro plans now. Open the Automatic tab and describe the edit, or select Enhance. The generation lands in a history above the composer. Re-prompting refines the entry you have selected, so "warmer" and then "a bit less" build on each other. Select the starting image to revert. Hover an entry to preview it on the canvas.
Generations outlive the tab that asked for them. Close it, come back, and the result is waiting in the sidebar. Anything saved from an AI result carries an AI-generated mark that survives re-saves and duplicates. Each member's monthly AI spend is capped, and the cap shows under the composer.
The agent layer
One set of keys and one model of access, across every way an agent reaches Raster.
- MCP. Raster is a remote Model Context Protocol server at
mcp.raster.app. Agents list, search, upload, tag, transfer, and delete on your behalf, and browse libraries through stableraster://URIs. - OAuth. Point Claude, Cursor, or any MCP client at Raster and sign in with one click. The connection runs as you, sees the libraries you can, and follows your access live. Revoke it under Connected apps in Settings.
- REST and GraphQL. Every endpoint is available over plain HTTP with an
Api-Versionheader on each request, alongside the GraphQL API. REST is in alpha. - CLI.
npm install -g @raster-app/cliputs every endpoint in your shell and CI, with--jsonfor scripts. The CLI is in alpha. Read the CLI docs → - Agent skill.
npx skills add raster-app/agent-skillsteaches Claude Code, Cursor, Copilot, Codex, and other coding agents how to use Raster. - Scoped keys. One key per integration, each with None, Read, or Write access per library.
- No account required. An agent can create an organization with a starter library and upload before anyone signs up, then hand over a claim link. Read the agent guide →
Raster also publishes llms.txt, an OpenAPI spec, and a plain-markdown connection guide, so a tool that goes looking for Raster can set itself up on its own. Every upload generates several sizes and Raster serves the one that fits the screen; over the API, each asset carries a sizes map and a blurhash.
What's next
Legacy DAMs were built for a slower era: a portal you upload finals into, far from where the work happens. AI made image creation instant, and the asset layer has to keep up, for the people making images and for the agents that increasingly make and fetch them. That's what Raster is now.
Sign in and see the new Raster.