LoginSign Up
← Back to all sparks
A

Air

DESIGN
Velocity8.8

Air's asset library grew an agent, and the agent is learning to keep what it does.

damcreative-agentscanvaslocalizationvideo-editingbrand-kit
◆Current state
Canvas has become the centre of the product. In the last three weeks it gained an agent that edits existing video rather than only generating new clips, two additional video models, the ability to save any completed workflow as a reusable skill, @ mentions that pull in board assets alongside Brand Kit colours, fonts and logos, and now in-image text translation across 33 languages. The library side kept moving too, with manual person tagging where facial recognition falls short, current-version share links, frame-level timeline zoom, and a batch of video features covering burned-in captions, transcript editing and sub-clipping.
◆Where it's heading
Air is converting a digital asset manager into a place where assets are produced and adapted, not just stored and approved. The skills feature is the pivot point: it turns one-off agent instructions into named, repeatable operations, which is how a creative team's conventions become part of the tool. Brand Kit entering the @ mention namespace points the same way — the agent is being given the brand's own vocabulary as addressable material, so generated work starts from house colours and logos rather than a generic prompt.
◆Prediction
Expect skills to gain sharing or workspace-level scope, since a saved workflow is far more valuable to a team than to the person who recorded it. The translation and captioning work together suggests localisation is being assembled as a workflow rather than left as isolated features.

◆Recent moves

  1. 5d ago

    NEW: Translation in Canvas

    ⚡ SPARK

    Translating text inside a flattened image removes the source-file dependency that made localisation a designer's job. Combined with automated captions and transcript editing, it turns Air from where localised assets are stored into where they are made.

  2. 5d ago

    New: One-click duplicate

    Right-click duplication for any canvas, asset or file, with copies named after the original and a confirmation message that names what was created. Standard interface housekeeping.

  3. 7d ago

    @ mentions now live in Canvas

    Typing @ on a canvas references any asset on the board or a linked board, plus Brand Kit colours, fonts and logos, with a kit switcher for multi-kit workspaces and colour chips from hex codes. It gives the agent the brand's own vocabulary as addressable input rather than leaving it to prompt description.

  4. 8d ago

    New: Manual Tagging

    Manual person tagging fills the gaps facial recognition leaves, allowing existing workspace people, newly created ones, and the same person tagged more than once in a photo. A correction path for automated metadata, which matters most in the large shoot libraries this product is bought for.

  5. 12d ago

    New: Video to Video in Canvas

    Canvas's agent can now take an existing video and change it rather than only generating new footage, handled by Kling's model alongside the other Canvas tools. It extends the agent to owned footage, which is where an asset manager's library actually is.

  6. 13d ago

    Two new AI video models available in Canvas

    MiniMax H3 and Alibaba's Wan 2.7 join Canvas, the former with 2K output and audio generated alongside the video, the latter spanning text, image, reference and video inputs. Model additions broaden what Canvas can produce without changing how it works, and Canvas's direction is still set by the skills and Brand Kit work rather than by model count.