QuiverAI

QuiverAI's pitch is narrow on purpose: it generates vectors, not pictures. Text-to-SVG turns a prompt into editable paths, image-to-SVG vectorizes raster PNGs into structured geometry, and the canvas lets you select and duplicate child elements, step through undo history, zoom, and fit artwork to the view. Every creation is real, editable SVG — not a raster export pretending to be vector — which is the entire reason a designer can hand the output to a developer without a cleanup pass. The September 2026 release cycle is what makes the tool worth re-evaluating today. Arrow 2 and Arrow 2 Telos (September 7, 2026) brought faster generation and improvements across animation, vectorization, and asset creation. The App now starts from a conversation and streams the drawing as it is made, and animation runs on the canvas so you can replay it without leaving the page. On the API side, POST /v1/responses runs on Arrow's Open Responses contract with function and custom tools, is stateless (it rejects store: true and previous_response_id rather than ignoring them), and bills by measured token usage. A sandbox key with the sk_test_ prefix hits the same routes but never reaches a model — it returns deterministic output and bills nothing, which is how you test a pipeline without burning balance. Distribution is the real differentiator. QuiverAI MCP (June 9, 2026) puts SVG generation into agent workflows; plugins exist for Cursor and Codex; npx quiverai add <creation> installs an SVG as a React component; @quiverai/sdk covers Node.js and @ai-sdk/open-responses covers the Vercel AI SDK. Timbal AI (May 19, 2026) already produces clean SVGs and brand assets through QuiverAI, which is a live proof point rather than a promise. Where it does not fit: photorealistic raster generation is out of scope by design, and manual path editing is shallower than Illustrator or Inkscape, so most teams pair the two. Typography and animations are still flagged coming soon on the marketing site even though canvas animation shipped in the App — treat custom font work as not ready. The billing split is the sharpest edge: subscriptions start at $8/mo for Go and include weekly app usage with no rollover, but API usage is billed separately from a prepaid API balance. Anyone automating at volume should budget for the subscription and the API balance as two line items. Arrow 1.x is the deadline everyone needs to know. The IDs arrow-1, arrow-preview, arrow-1.1, and arrow-1.1-max retire at 08:00 UTC on October 16, 2026 and return 404 model_not_found — requests are not silently routed to Arrow 2. A key that allows all models only needs its request model changed to arrow-2; a key scoped to an exact Arrow 1.x model is immutable and needs a new key created, the new secret deployed, then the old key revoked after traffic moves. Rotation preserves authority and is not a model migration. Plan the cutover before the cutoff, not after.