## The short answer
AI rendering turns a simple input, a sketch, a view of your model, a photo or a sentence, into a finished-looking image in minutes. The useful versions for architects keep your design fixed: the same volumes, proportions and camera, with only the surfaces, the light and the setting changing. Used that way it is a fast way to explore options, present early work and prepare competition images. It is not a replacement for a calculated render when you need one, and every image needs a check before a client sees it.
## What AI rendering actually is
A render engine such as V-Ray calculates light from the materials and lamps in a scene. An AI render tool works from a picture: it looks at your input and draws a new image, using what the model has learned about how buildings, materials and light look together. That is why it is fast, and also why it needs checking. Light in an AI image looks right but is not measured.
## Choose the right input
- **A sketch.** Fast, loose, good for finding a direction. The more clearly the sketch shows massing and openings, the better the result. See [AI render from a sketch](/news/ai-render-from-sketch-for-architects).
- **A view of your model.** The strongest input for presenting a design, because the building and the camera come from your model. Send it straight from [SketchUp](/plugins/sketchup), [Rhino](/plugins/rhino) or [Revit](/plugins/revit) with the OctopusWave plug-ins.
- **A photo.** Useful for existing buildings and sites: render a proposal over the real context, or turn the photo into an editable model to design against.
- **A text prompt.** Good for ideas and mood boards, least reliable for a specific building.
## Keep the geometry and the camera
If an AI render changes your building, it is no longer a picture of your design. Three habits prevent that:
1. **Start from a view of your model, not from a description.**
2. **Keep the frame.** Send and render at the same proportions as your window, because a different frame makes the model recompose.
3. **Use a tool that locks geometry.** OctopusWave tells the model to reproduce the volumes, heights, openings and camera exactly and to change only surfaces, light and setting. Compare the result with the model on every image anyway.
The step-by-step versions are in the [SketchUp](/news/sketchup-to-photoreal-render-workflow), [Rhino](/news/rhino-to-photoreal-render-workflow) and [Revit](/news/revit-to-photoreal-render-workflow) workflows.
## Get one consistent look across a project
A plan, a section and three street views should look like they belong to one project. Three tools help:
- **Render the set together,** in one style, instead of one image at a time.
- **Use a reference image** that carries the materials, the colours and the photography you want. The render takes the look and keeps your building.
- **Train your own style.** With [My Signature](/news/train-your-own-ai-render-style-my-signature), you teach the system a look from 15 to 80 of your own images and apply it to every render.
## Beyond the still image
Once you have a good render, the same canvas can take it further:
- A short slow camera film from the render, or a longer [architectural walkthrough](/news/ai-video-model-for-architectural-walkthrough).
- A [360 panorama](/news/ai-360-panorama-from-render) you can step into.
- An editable 3D model of an existing building from a photo, see [image to 3D for architects](/news/image-to-3d-for-architects).
- A PDF deck and first drafts of client documents written from your renders.
## What AI rendering will not do
- **It will not calculate daylight, glare or energy.** Use proper simulation for that.
- **It will not give you drawings.** Plans, sections and elevations come from your model.
- **It will not show what you did not send.** Each side of the building needs a view.
- **It will not remove the need to check.** Count floors, check openings and read any text in every image.
## When to use something else
For live design review inside the model, a real-time plug-in is better. For the one image that has to be exact, an offline renderer is better. For pure ideas, a general image tool is fine. See [AI render or Enscape, V-Ray, Lumion and D5](/news/ai-render-or-enscape-vray-lumion-d5) and [Midjourney or a model-aware tool](/news/midjourney-or-model-aware-ai-render-for-architects).
## A checklist before you show a client
1. Does the image match the model: floors, volumes, openings, camera?
2. Is every repeated element right: mullions, balusters, columns?
3. Is any text correct, or removed?
4. Do all the views of the project look like the same building in the same light?
5. Is it clear to the client that this is an artist's impression, not a measured result?
## How to start
Install the plug-in for your program from the [plugins page](/plugins) and send one real view. New accounts get five free renders, so you can judge the result on your own project before choosing a plan. Model Link is part of the [Studio plan](/pricing), and you can also pay per render with credits and no subscription.