AI for architects

Sketch to render with AI: turn a sketch or floor plan into a photoreal image

By Hadhi Muhammed P. · · 9 min read

Short answer: To turn a sketch into a render with AI, upload a clean photo of the sketch to Gemini or ChatGPT and write a prompt with five parts: keep the geometry, name the materials, set the light, set the camera and describe the surroundings. Then change one thing at a time. You get a photoreal concept image in a few minutes. It is a picture, not a scaled drawing, so use it for early client approval and not for construction.
Key takeaways
  • You need only a browser: Gemini or ChatGPT, on a free plan to start.
  • The prompt formula is Keep + Materials + Light + Camera + Context.
  • Always tell the AI to keep the form, openings and proportions of your image. This one line prevents most bad results.
  • Change one thing per message. Asking for five changes at once makes the building drift.
  • AI renders are concept visuals. They are not to scale and can invent details, so label them clearly for clients.

What do you need to turn a sketch into a render?

Three things, and none of them is rendering software or a powerful PC:

  1. An input image. A hand sketch, a printed floor plan, or a screenshot of a SketchUp, Revit or Blender model.
  2. An AI chat tool that edits images. Gemini and ChatGPT both do this. Gemini's image model is nicknamed Nano Banana.
  3. A clear prompt. That is the rest of this guide.
InputWorks well forWhat to tell the AI
Hand sketchEarly concept, elevations, perspectivesWhich lines are walls, glass and roof
Floor plan3D cutaway views, furniture layout ideasRoom names and which way is north or the entry
Model screenshot (clay or white)The most accurate resultsMaterials for each surface
Site photoShowing the design on the real plotWhat to keep and what to replace

A model screenshot gives the most faithful render, because the perspective and proportions are already correct. A loose sketch gives the AI more freedom, which is good for ideas and bad for accuracy.

How do you prepare the sketch?

The AI can only keep what it can see. Two minutes of preparation saves ten minutes of fixing.

  • Shoot it flat and straight. Photograph or scan the sketch from directly above, in even light, with no shadow of your phone.
  • Darken the lines. Go over the main outline in pen. Faint pencil gets ignored.
  • Crop tight. Remove the desk, your hand and other drawings from the frame.
  • Label what is unclear. Write "glass", "void" or "sloped roof" in small letters where the drawing could be read two ways.
  • For a model screenshot, switch off axes, guides and section planes, and pick the camera angle you want in the final image.

What is the prompt formula for an AI render?

Use five parts, in this order. Google's own guidance for its image model says the same thing in different words: be specific about the subject, the lighting and the camera, and say what must stay the same.

PartWhat it controlsExample words
KeepStops the building from changing"Keep the exact form, openings and proportions"
MaterialsWhat every surface is made of"Laterite walls, exposed concrete slab, teak louvres"
LightTime of day and mood"Soft morning sun from the left, light monsoon clouds"
CameraViewpoint and lens"Eye-level view from the road, 24mm wide lens"
ContextWhat surrounds the building"Coconut palms, compound wall, wet road after rain"

1. Hand sketch to photoreal exterior

Works in Gemini · ChatGPT
Turn this hand sketch into a photorealistic architectural photograph.

Keep: the exact building form, roof shape, window and door positions and proportions from the sketch. Do not add or remove floors, windows or balconies.
Materials: [FOR EXAMPLE laterite stone walls, white plaster, exposed concrete slab edges, teak wood louvres, clay roof tiles].
Light: [soft morning sunlight from the left, a few clouds].
Camera: [eye-level view from the road, 24mm lens, vertical lines straight].
Context: [tropical garden with coconut palms, stone paving, low compound wall, quiet residential street in Kerala].

Style: real photograph for an architecture magazine. No people, no cars, no text.

2. Clay model screenshot to render

Works in Gemini · ChatGPT
This is a screenshot of my 3D model in plain white. Turn it into a photorealistic render.

Keep the camera angle, the geometry and every opening exactly as in the screenshot. Only add materials, light and surroundings.

Materials by surface:
- Walls: [MATERIAL AND COLOUR]
- Roof: [MATERIAL]
- Windows: [FRAME COLOUR, CLEAR OR TINTED GLASS]
- Floor or paving: [MATERIAL]
- Feature element: [FOR EXAMPLE vertical timber fins on the first floor]

Light: [TIME OF DAY AND WEATHER].
Surroundings: [LANDSCAPE, NEIGHBOURS, ROAD].
Photographic, natural colours, no people, no text.

How do you turn a floor plan into a 3D view?

A plan has no heights, so tell the AI what kind of 3D view you want. A cutaway view from above is the most reliable. Treat the result as a way to explain the layout to a client, not as a measured model.

3. Floor plan to 3D cutaway view

Works in Gemini · ChatGPT
This is a 2D floor plan. Create a 3D cutaway view of it, seen from above at a 45-degree angle, with the roof removed.

Keep the room layout, wall positions, doors and windows exactly as drawn. Do not add or remove rooms.

Rooms: [LIST THE ROOMS AS LABELLED, FOR EXAMPLE living, dining, kitchen, 3 bedrooms, 2 toilets, sit-out].
Walls: 3 metres high, white plaster.
Floors: [FOR EXAMPLE vitrified tile in living areas, wood-look tile in bedrooms].
Furniture: simple modern furniture that fits each room. Keep walking space clear.
Light: bright even daylight, soft shadows.

Clean architectural model style, plain light background, no text or labels.
Architects cohort

We are building an AI course together with architects, interior designers and civil engineers. Want to be part of it?

25+ course students, 500+ live webinar attendees and ₹1,00,000+ earned by students. See what they made Join the architects cohort

How do you fix a render that looks wrong?

Do not start again. Reply in the same chat and change one thing. The AI remembers the image it just made, so small follow-up edits stay consistent.

4. One-change follow-up edits

Works in Gemini · ChatGPT
Use these one at a time, as replies to the image you just got:

- Keep everything exactly the same. Only change the wall material to [MATERIAL].
- Keep everything exactly the same. Only change the time to dusk, with warm lights on inside.
- Keep the building exactly the same. Only replace the sky with heavy monsoon clouds and make the ground wet.
- Keep everything exactly the same. Remove the [OBJECT] on the [LEFT / RIGHT].
- The window on the first floor, left side, has changed. Restore it to match my original image.
ProblemWhy it happensFix
The building shape changedNo "keep" line, or too many changes at onceAdd the keep line and make one change per message
Extra windows or floors appearedThe sketch lines were unclearLabel the sketch, or say "exactly 2 floors, 3 windows on the front"
Materials look plasticVague words like "modern" or "nice"Name the real material and finish: "rough laterite", "matt white plaster"
Verticals are leaningWide-angle distortionAdd "vertical lines straight, two-point perspective"
It looks like a paintingNo photo words in the promptAdd "real photograph, natural colours, 24mm lens"
Text and signs are garbledImage models are weak at small textAdd "no text" and place text later in your own software

5. Same design, three moods

Works in Gemini · ChatGPT
Using the render you just made, create 3 versions. Keep the building, camera and materials exactly the same in all three. Change only the light and weather:

1. Bright morning, clear sky
2. Golden hour, long warm shadows
3. Dusk after rain, interior lights on, wet reflections on the paving

Show them as 3 separate images.

6. Place the design on the real site photo

Works in Gemini · ChatGPT
Image 1 is a photo of the empty plot. Image 2 is my building design.

Place the building from image 2 onto the plot in image 1. Keep the real road, neighbouring buildings, trees and sky from image 1. Match the sunlight direction and shadows to the site photo. Keep the building's form and materials exactly as in image 2.

Photorealistic. No people, no text.

What are the limits of AI renders?

  • They are not to scale. Proportions are approximate. Never take a dimension from an AI image.
  • They can invent things. Check every opening, level and railing against your drawing before you show a client.
  • Fine detail and text are weak. Google notes that small text and fine details may not come out right.
  • Images may carry an AI watermark. Read each tool's terms if you plan to use the image in paid advertising.
  • Tell the client it is a concept. A line such as "AI concept visual, not final design" avoids arguments later.

Once a still is approved, the next step is motion. See how to make an AI walkthrough video from a render. For the writing side of a project, use our ChatGPT prompts for architects.

Questions people ask

Can AI turn a sketch into a realistic render for free?

Yes. Gemini and ChatGPT both accept an uploaded sketch and return a photoreal image on their free plans. Free plans limit how many images and uploads you get each day, so plan your prompts before you start.

Which AI is best for sketch to render: Gemini or ChatGPT?

Both work with the prompts in this guide. Gemini's image model is designed to edit an uploaded image while keeping the rest unchanged, which helps a building stay true to the sketch. Run the same sketch and prompt in both and compare.

Can AI convert a 2D floor plan into a 3D model?

It can create a 3D-looking image of the plan, such as a cutaway view, which is useful for explaining a layout to a client. It does not create a real, measurable 3D model. For that you still need SketchUp, Revit, Blender or similar software.

Will AI replace V-Ray, Lumion or Enscape?

Not for final, accurate visuals. Rendering software works from your real model, so every dimension is right. AI is faster for early concepts, mood options and quick client approvals. The two work well together: AI first, rendering software for the final set.

Do I need a powerful computer for AI rendering?

No. The image is generated on the tool's servers, so a normal laptop or even a phone is enough. This is the main reason small studios pick it up quickly.

Sources

Facts in this guide were last checked on . AI tools change their plans and features often, so confirm limits and prices on the official pages above.

Hadhi Muhammed P., founder of Raise AI
Hadhi Muhammed P.

Founder of Raise AI in Kozhikode, Kerala. He makes AI video ads and teaches working professionals, including architects, interior designers and civil engineers, to use AI in their own job. About us

Read next

Prompts are step one. A real workflow is step two.

Join the architects cohort