Render Studio

Render Studio

Turns the 3D view you are looking at into a photoreal image, without leaving Revit and without a render farm. Open the view you want, capture it, say what the finished thing should look like, and render. Every attempt is kept, so the one you liked two tries ago is one click away.

The loop

1. Capture the view. Deliberately a separate, visible step: you see exactly what is about to be rendered before spending anything on it. The plugin exports the active view at 1600px into an empty temp folder, so it never picks up a neighbouring file.
2. Say what it should look like. Start from a preset — exterior daylight, exterior dusk, warm interior, white model, watercolour — then edit the text. The presets fill the box; they do not replace it.
3. Choose how close to the model to stay. Faithful keeps the geometry, proportions and camera exactly and changes only materials, light, weather and context. Balanced keeps the massing and viewpoint but refines detail and surroundings. Reinterpret treats the model as a starting point.
4. Render, or Refine this render to iterate on the image on screen instead of the original capture — the way you chase a look you nearly have.
5. Or mark one part and render just that. Paint it, outline it corner by corner, or drag a box; say what should be there; the marked area is rendered and blended back with a soft edge. This works on a render, and on the model view itself before anything has been rendered — the cheapest way to try one idea.

Save writes a PNG next to your other renders and can open the folder.

How it renders

Through BimCode's own AI, on the agent-loop lane (aiTurn) with the GenerateImage server tool. There is no API key, no model name and no provider account in this plugin: the model is whatever the signed-in user picked under Model for AI plugins, and every render is metered to them like any other AI usage. If their plan does not include AI plugins, or the allowance is spent, the refusal is shown in the plugin rather than swallowed, and everything else keeps working.

A note the AI lane makes necessary: there is no view-capture tool on the server side. The plugin exports the view itself in C# and attaches the image to the request — which is also why a plan or elevation warns before you render it, since those export fine and render badly.

Shape

• Bridge/RenderStudioBridge.cs — the JS-callable surface: read the active view, capture it, save a render, reveal it in Explorer, and the AI self-test behind Check AI.
• Bridge/BimCodeAi.cs, Bridge/AiSelfTest.cs — the AI kit, copied in.
• web/src/render/useRenderStudio.js — the whole workflow as one hook: capture, fidelity clauses, render, history, save.
• web/src/App.jsx — canvas-first: one image stage, an inspector, and a filmstrip of attempts underneath.
• web/src/components/Stage.jsx — the picture and everything that must sit exactly on it: the compare wipe, the mask, the outline, the box. It measures the stage and fits the picture into it, so the overlays share the picture's box rather than the stage's.
• web/src/components/Rail.jsx — the inspector's four pages (Render, Style, References, Setup) and the action bar pinned under them.
• web/src/components/AreaToolbar.jsx — the tools that appear while marking an area.
• web/src/bridge/devScenario.js — what the fake bridge answers in npm run dev: a drawn model view and every bridge call, so the screen can be worked on in a browser with no Revit and no spend.

Build with npm run build in web/, then dotnet build — web first, or the window opens to a bare WebView2 error.