Rhino AI Render

No, and it is worth being blunt about it. There is no 3DM parser, no Grasshopper component and no Rhino add-in. The only thing Archome reads is an image — a PNG or JPG of the view you captured yourself, usually with ViewCaptureToFile or a plain screenshot.

That is a deliberate trade, and one side of it is genuinely good news. Because the input is a picture, your NURBS surfaces, layer structure, block definitions and Grasshopper logic never leave your machine; you decide exactly what is shared, one frame at a time. It also means the workflow is identical on Rhino 7 and Rhino 8, on Windows and on Mac, and nothing breaks when you upgrade.

The other side is equally real. The AI only knows what the camera saw. Geometry behind a wall, a material you assigned but did not display, and anything outside the frame do not exist as far as the render is concerned. Frame the view you actually want before you capture it, because there is no second chance later in the pipeline.

If you need geometry back out rather than an image in, that is a different tool: image to 3D generates a mesh from a picture — but it is a newly generated mesh, not your Rhino model, and it is for presentation rather than fabrication.

Not from Archome, and pretending otherwise would just waste your afternoon. Archome ships real, downloadable plugins for exactly two applications: SketchUp as an .rbz extension and 3ds Max as a MAXScript. Rhino, Revit, Blender and ArchiCAD are browser workflows with no add-in at all.

In practice the gap is smaller than it sounds. Run ViewCaptureToFile into a folder you keep open, drag the PNG into the browser tab, type the prompt. That is two extra actions compared with a one-click add-in, and they are the same two actions every time.

If your studio genuinely wants Rhino-side automation, there is a supported route: renders can be triggered over the API with a key generated in your account settings, so a RhinoScript or Python macro can capture a view and post it without a human in the loop. That is a build-it-yourself path rather than a product handed to you, and it is described honestly as such.

What no browser workflow gives you is real-time navigation. Tools that live inside Rhino redraw as you orbit; here you choose a frame and then render it. Before comparing against any specific add-in, check that product’s own site for its current requirements — those change often enough that repeating them here would only mislead you.

This is the question people actually type, and the accurate answer has two halves. Nothing is imported here — no file crosses over — so what stays fixed is not the file but the view. Within that view the answer is a clear yes: the render is conditioned on the edges, depth and silhouette of your capture, not generated from the prompt alone.

Concretely, the massing, curvature, window positions, floor lines, roof pitch and camera angle in your Rhino capture survive into the render. What is generated on top is everything you did not model: surface material, weathering, glazing reflections, sky, light direction, vegetation, wet asphalt, the people on the pavement.

That is a different promise from a text-to-image tool, where every generation invents a new building. If you want that behaviour instead — a building from a description with no source view — Imagine is the tool built for it, and the distinction between the two is the whole point of this page.

Where the rule bends: a capture with almost no tonal information, such as a thin wireframe on white, gives the model very little to hold and the result drifts. Arctic, Shaded and Rendered modes carry depth, so they hold much better. For deliberate line work, sketch to render is tuned for exactly that input.

Arctic is the strongest default. It gives clean white surfaces with real ambient occlusion, so the AI reads form and depth without being distracted by placeholder colours or half-finished materials.

Shaded and Ghosted work well too, and Ghosted is useful when you want interior volumes to read through a facade. Rendered mode is fine, and you do not need materials assigned first — an untextured capture usually produces a cleaner result than a partial material set, because there is nothing for the prompt to argue with.

Technical and Pen modes, and Make2D output, are line drawings. They render, but through a different door: send them to sketch to render, which expects line work rather than shaded surfaces. Interior views go to interior rendering; building exteriors and site views to exterior rendering.

Capture at the largest resolution your screen allows. A small, soft screenshot limits how much detail there is to preserve in the first place. If you need a large file at the end, run the finished render through 4K upscaling rather than starting bigger.

There is no Grasshopper component, so this is a capture-and-batch workflow rather than a live one. Bake or preview each option, run ViewCaptureToFile for each, then render the set one image at a time.

What makes a study set look deliberate is prompt discipline: the same prompt, word for word, across every iteration. Change the geometry, hold the material and light language constant, and the comparison reads as a study rather than a grab bag. Keep the sentence in a text file and paste it each time; the prompt library has starting points worth stealing.

The arithmetic matters when you are rendering twenty options. A render costs 10 credits on Fast and 30 on Ultra, so twenty Fast iterations is 200 credits — well past the 100 free credits an account gets each month, and a fifth of the entry subscription’s monthly allowance. The numbers are laid out on the pricing page.

A practical sequence: render the whole set on Fast to find the direction, then re-render only the two or three survivors on Ultra for the presentation. If you are still deciding where the production render belongs, the comparison of AI rendering against V-Ray covers that trade-off directly.

A good prompt names four things: material, light, context and camera. "Board-formed concrete and blackened steel, overcast late afternoon, mature birch behind, shot on a 35mm lens" is far more useful than "make it photorealistic", because every clause gives the model something specific to change.

Name materials by their construction rather than their mood. "Vertical larch cladding at 100mm centres" lands; "warm and natural" does not. The same applies to glazing — "low-iron glass, minimal reflection" and "dark tinted glass, strong sky reflection" produce genuinely different facades from an identical capture.

Say what should not change if something keeps drifting. Explicit instructions such as "keep the roof flat" or "no additional balconies" hold better than hoping. And if one region is wrong while the rest is right, inpainting edits that area instead of re-rolling the whole image and losing what already worked.

There is a library of prompts written for architecture rather than general image generation in the prompt collection, and a longer explanation of the structure in the guide to architectural render prompts.

It cannot open your model. No .3dm, no .gh, no STEP, no IGES — the input is an image, every single time, and no roadmap promise is being made here.

It is not physically accurate. Light is not simulated, so reflections, caustics and shadow angles are appearance rather than analysis. Nothing produced here belongs in a daylight study or a compliance document, and no dimension in the output is measurable.

It cannot render what is off-camera, cannot reliably reproduce text, logos or signage, and will not repeat a variation exactly — two generations from the same capture and prompt will differ. It is also not a walkthrough engine: video produces a short cinematic clip from a finished still, not navigable geometry.

Free renders are for evaluation and personal use and free output has no commercial licence. Both change with any purchase, including the smallest credit pack.

Every account gets 100 free credits to start, renewed, with no card. A render costs 10 credits on Fast and 30 on Ultra, so the free allowance is 10 Fast renders or 3 Ultra renders a month. Free output is for evaluation and personal use — that is the honest shape of the free tier, not twenty finished images.

One-time credit packs are $15, $39 and $99 and stay valid for twelve months, which suits project-driven practices that render in bursts. Subscriptions are $29, $49 and $119 a month for a renewing credit allowance.

Any purchase, including the $15 pack, adds a commercial licence. The full breakdown, including annual billing, is on the pricing page.

Heavier tools cost more from the same balance: 4K upscaling is 30 credits and a video clip is 100. Plan the last mile of a presentation rather than discovering it at the end.