
Render Studio turns a Floor Plan Studio project into photoreal visuals through a chat transcript instead of a slider panel: describe what you want changed — materials, lighting, mood — in plain language, and each message becomes a new render saved to the project's history. Every render starts from either an uploaded image or a capture of the current plan (capture is a deliberate action, not automatic, so a render is never silently generated from whatever happens to be on screen), with a hard-coded system prompt instructing the model to preserve wall, opening and room geometry rather than redesigning the layout.

Sketch to Floor Plan, Plan to 3D (axonometric), Elevation and Perspective — each with its own geometry-fidelity prompt clause.
A capture dialog toggles walls, room shading, openings, plot/setback lines, furniture, dimensions and more, with a live preview before anything is sent.
Each render is scored and badged excellent / good / review / retry, with a short strengths/findings note — and can auto-retry with a corrective prompt if the score is weak.
The server-rewritten "enhanced prompt" actually sent to the image model is shown in the transcript, not hidden behind the scenes.
Pick a region (whole/center/top/bottom/left/right) and only that part regenerates, composited back into the original.
Generate 1-6 variants per request, lock a seed to keep composition consistent, or set one manually to reproduce a result.
Five visual styles (Neutral, Realistic, Clay, Warm, Luxury), four lighting conditions (Daylight, Studio, Golden hour, Night) and three detail levels (Clean, Balanced, Rich) combine into a prompt without you having to write one from scratch. A separate quality mode sets the actual target resolution generated — Fast (1024px) for a quick check, up to Ultra (4096px) for a presentation-ready export — and a style reference image can steer material mood specifically, kept separate from the geometry-preserving base prompt.
Renders are organized as chats grouped by project, not one long undifferentiated feed — a chat's first prompt becomes its title, and older pre-chat history is kept under "Earlier renders" per project so nothing gets lost when the organization changed. History is saved locally and synced to your account when signed in, and a saved render's link is automatically re-resolved if it expires, rather than just breaking.
Beyond region edits, dedicated workflow modes cover the most common follow-ups directly: Furnish stages an empty room with furniture, Enhance upscales a chosen result for a larger export, and every generation can include or exclude labels/annotations, furniture/fixtures and site context independently — so a follow-up message can ask for exactly one change without re-specifying everything else.
Every generation includes a system prompt instructing the model to preserve wall, opening and room geometry rather than redesigning the layout — you choose exactly what's included (walls, furniture, dimensions, site context, and more) via a capture dialog before it's sent.
Yes — a region-edit mode lets you pick whole/center/top/bottom/left/right, and only that part regenerates, composited back into the original so the rest of the image is untouched.
Every output gets a quality score and badge (excellent/good/review/retry) with a short findings note, and you can turn on auto-retry so a weak result automatically regenerates with a corrective prompt.
They're AI-generated and illustrative — useful for communicating design intent to a client, not a substitute for a construction rendering with verified materials and lighting.
A dedicated Furnish mode stages an empty room with furniture as a starting point, which you can then refine with the usual chat-driven edits.
Yes — the server-enhanced prompt actually used for generation is shown in the chat transcript, not hidden, so the enhancement step is never opaque.
Yes — 1 to 6 variants per request, each with its own seed unless you lock the seed to keep the composition consistent across variants.