BUI is an ambitious act of surgery: take the interface that millions of 3D artists have trained their hands on — the drag-to-adjust number fields, the fuzzy menu search, the damage-only redraws — and extract it from Blender's codebase into a freestanding C core with bindings for six languages. The pitch is that decades of optimization shouldn't be locked inside one application. The technical surface area is enormous. The kit at the center holds the widget tree, handles layout, input, and drawing using code ported directly from Blender's interfacewidgets.cc. On top sit thin binding layers for Python (including full tkinter and ttk API compatibility — all 7,846 public names), Rust with HTML/CSS/Tailwind rendering, JavaScript via React and Svelte over napi-rs, Dear ImGui, and raw C. You pick the row that matches your existing code and, in theory, your app inherits Blender's feel. The performance claims are specific and bold. BUI reports 2–4× less CPU per frame than Zed's gpui, with less RAM and a roughly 5ms first frame. Damage rendering means an idle UI costs zero CPU; a hover redraws only the touched widget. The GPU renderer (Direct3D 11 on Windows, OpenGL 3.3 on Linux) models Blender's own instanced widget drawing. Image decoding happens off the UI thread with JPEG DCT scaling, a 256 MiB memory-budgeted cache, and settled frames with zero allocations. Video playback via OS decoders (Media Foundation, GStreamer, AVFoundation) runs 1080p at 0.99ms render frames. What's actually here today: tkinter compatibility is marked 'Complete,' as are the 45 native Blender component bindings from Python. Rust, JavaScript/React/Svelte, and Dear ImGui integrations are 'Working.' The README is transparent about this — it's a status table, not marketing. Go bindings are listed as 'Not yet.' Nine real tkinter programs reportedly behave identically on BUI as on the standard library, which is the kind of concrete claim that invites verification. The project's core bet is that Blender's UI idioms — hidden-cursor dragging, math expression evaluation in number fields, press-drag-release pie menus, fuzzy search with typo forgiveness — constitute a genuinely superior interaction vocabulary that deserves to escape its host application. Whether that vocabulary translates outside 3D workflows is the open question. Artists who live in Blender will feel at home instantly. Developers building creative tools have an obvious use case. For general-purpose desktop apps, the Blender-specific idioms might feel alien to users who've never opened Blender. The README functions as both documentation and manifesto, structured as a seven-part book covering concepts, per-language guides, performance benchmarks, and build instructions. The rendering architecture section explaining CPU vs GPU paths, damage rendering, and font handling (Inter embedded, FreeType rendering, Slug for GPU glyph outlines) reads like genuine systems engineering rather than marketing. The image and video pipeline descriptions include specific latency numbers — 2.9ms p99 frames scrolling 1,000 thumbnails, 0.76ms p50 for 200 simultaneous animations. BUI sits at an interesting intersection: it's neither a web-tech wrapper (no browser engine, no webview) nor a traditional retained-mode toolkit. It's a port of a production UI that's been battle-tested by millions of users in one of the most demanding real-time creative applications. The 'twice as good as GPUI' headline claim is provocative but backed by specific benchmark methodology and an invitation to run them yourself. The real test will be adoption outside the Blender-adjacent community.
