Design tools moved generation off the canvas and into agents, while 3D became a first-class format.
The week in design
Two patterns ran through the design sector this week, and they reinforce each other. The first is that AI generation is leaving the canvas. VEED shipped both a ChatGPT app and an MCP server, making its video engine addressable from outside its own editor; Jitter launched an AI that builds custom effects from a prompt rather than offering a fixed library; Webflow extended hosting and AI-answer optimization well past the visual builder. The common thread is that the tool's interface is no longer the only place the work happens — the engine is being exposed to assistants, agents, and repositories.
The second pattern is 3D arriving as a first-class format. ComfyUI added native 3D Gaussian Splats, turning a single image into renderable 3D inside a node graph, and Frame.io promoted 3D to a reviewable asset type with automatic turntables and degree-stamped comments. These are different ends of the pipeline — generation and review — converging on the same conclusion: 3D is no longer a specialist detour. Underneath the leaders, a steady cadence of model-support and motion-design work filled out the week, with a long tail of products shipping only incremental changes or nothing observable.
Leaders
Webflow posted the week's highest velocity by pushing past website building on several fronts at once. Webflow Cloud now deploys apps straight from a repository with no site required, decoupling hosting from the visual builder. Alongside it, Webflow AEO launched for Enterprise to measure how a brand appears in AI answers and ship fixes with agent help — an entirely new product surface rather than a refinement of the editor.
VEED made the clearest agent-era move in the sector. In roughly a month it has exposed its generation engine through an MCP server and a ChatGPT app, so a prompt can become a finished video without ever opening VEED's canvas. The strategy is unbundling: the editor becomes one of several front doors, and the engine becomes something agents drive on the user's behalf.
Jitter reframed its motion-design product with Jitter AI, which generates a custom tool or effect from a description — a glitch transition, a 3D rotation, a ripple — inside the editor, then lets users reuse and share it. It shifts the product from a fixed set of capabilities to an open-ended one, on top of a deepening manual toolset of components, shaders, and pen-tool work.
Frame.io added 3D asset review in beta, treating 3D as a first-class format with automatic turntable previews, USD-based ingestion, and degree-stamped comments running through the same versioning and approval workflow as video and imagery. It landed alongside Frame.io's promotion into Adobe's Top App Bar — the clearest sign yet of its absorption into Creative Cloud.
ComfyUI continued its day-0 breadth strategy, this week stretching from images and audio into native 3D. TripoSplat adds native support for 3D Gaussian Splats, converting one image into renderable 3D Gaussians directly in a node graph — extending the 3D push from mesh generation into splats and making 3D a permanent part of the workflow rather than an experiment.
Wildcards
Mentimeter is doing something off-pattern: pushing past the live-presentation moment entirely. Menti Pulse, in closed beta, delivers recurring anonymous pulse checks inside Microsoft Teams chats with trends tracked over time — engagement that happens between meetings rather than during them. In parallel it made its AI features a default included part of the product for all users.
Octopus.do is repositioning a niche tool — the visual sitemap — into the front of a production pipeline. It rebuilt its content editor from scratch (adding columns, tables, and alignment) and updated its Figma plugin to turn a planned project into a full high-fidelity prototype with every component wired to variables, so the planning stage now feeds directly into design.
Themes that compounded
- Generation moved off the canvas: VEED, Jitter, and Webflow all exposed capability outside their own editor this week.
- MCP and assistant surfaces appeared in design too — VEED's MCP server and ChatGPT app make its engine agent-addressable.
- 3D became first-class at both ends of the pipeline — ComfyUI for generation, Frame.io for review.
- Prompt-built tooling replaced fixed libraries, with Jitter generating effects on demand rather than offering a menu.
- Model-support velocity stayed a moat for open-source tooling, with Krita's plugin first to support Anima, ERNIE Image, and Flux 2.
Watch this week
The pattern to watch is whether more design tools follow VEED in exposing their engines over MCP and into assistant apps. VEED has now shipped both an MCP server and a ChatGPT app; Webflow is adding agent-assisted workflows inside its platform. If a second editor ships an MCP surface next week, the off-canvas shift becomes a sector trend rather than one company's bet. Watch also whether the 3D momentum continues — ComfyUI and Frame.io both treated 3D as first-class this week, and Krita's relentless model cadence suggests the open-source side will keep adding 3D and image models as fast as they appear.