Midscene is no longer just a browser automation project. The official site now sells it as AI-powered, vision-driven UI automation for every platform, and the repo is already on v1.7.6 while v1.7.5 added Studio polish, multi-platform playground work, and a hard 180 second timeout on AI HTTP calls. That points to a team shipping for breadth and reliability at the same time.
What the data says
The homepage groups the product into Web, PC, Mobile, and any interface automation. It also pushes a Skills and MCP layer, plus a multi-model strategy with Doubao Seed, Qwen3-VL, and Gemini 3 Pro. That is a deliberate bet. Midscene wants to be the control plane for UI automation, not a single-platform helper.
v1.7.5 backs that up. The release notes call out dark mode in Studio, a playground shell, a collapsible execution flow, a multi-platform playground for Android, iOS, HarmonyOS, and computer, plus fixes for scroll behavior, replay crashes, and duplicate retries. The 180 second AI timeout is the clearest signal. The team is hardening the core path, not just adding new demos.
What this does not tell you
ToolVitals sees stars, commits, tags, and releases. It does not see whether the agent clicks the right thing, whether the UI feels sane under load, or whether users trust it after week three. The 12.8k stars and 95 health score show attention. They do not prove user success.
How it compares
Relative to nearby tools, Midscene is not the fastest shipper in the set. OpenClaw logged 43 release events in the last 30 days, and React Email logged 26. Midscene logged 19. That is still a serious cadence, just not the loudest one in the room.
Bottom line
If your team needs vision-based automation across web, desktop, and mobile, Midscene is worth a close look now. If you only need DOM-level browser tests, it is probably too much machine for the job. The bet here is clear, a multi-platform UI control layer that ships fast and keeps widening its surface area.