Gains:
- Ability to choose text-to-image, image-to-image and regional editing methods according to the right job
- Ability to enrich rendering by writing a layered visual prompt and preserving its own geometry
- Ability to sort out physical impossibilities and hallucinations and present the visual honestly to the customer as a 'representation'
Visual and Render Production: Making the Atmosphere Visible with AI
Customers can't read the plans, but they feel the visuals. Render (a realistic-looking computer image of a space) is the most powerful tool to visualize an abstract design in the customer's eyes. Traditional rendering requires setting up a 3D model, assigning materials and calculating light, and takes hours. Visual generator AI (artificial intelligence models that generate images from text or transform an image) reduces this process to seconds: producing an atmospheric image from a text description or a rough sketch. In this unit, we will cover AI visual production from end to end — from prompt writing to customer presentation and ethical boundaries.
But let's be clear from the beginning: an AI visual is a representation, not a technical project. The furniture inside may not be commercially available, the dimensions may not be realistic, the light may be physically impossible. Use visuals for inspiration, communication and discovery; It is your measured drawing and actual material selection that will be applied.
Types of AI tools that produce images
There are three basic ways to use it:
- Text-to-image. You write a recipe, AI produces a space visual from scratch. Ideal for concept and atmosphere exploration; It is weak in measurement and accuracy.
- Image-to-image. You give a sketch, rough 3D screenshot or photo to AI, and it turns it into a "realistic rendering" according to the style you specify. It's the most valuable method for designers because it preserves scale and order — because it's based on the geometry of your measured sketch.
- Regional editing (inpainting). You change only one area of an existing image (for example, only the chair or wall color), and the rest remains constant. Powerful for variation and customer revision.
Tip: If you want to maintain scale and order, choose the image-to-image method instead of the text-to-image method: "reference" a rough perspective from SketchUp/CAD to the AI and just enrich the materials, lighting and atmosphere. This way the geometry remains under your control.
Anatomy of a good visual prompt
An effective visual prompt is layered. It includes the following components:
- Space type and function: "living room", "boutique hotel lobby".
- Style/concept: "Scandinavian minimalism", "warm industrial".
- Material and colour: "light oak floor, matt white wall, anthracite detail".
- Light: "soft morning light, natural light from the left".
- Camera/angle: "wide angle, eye level, two-point perspective".
- Atmosphere: "calm, spacious, inviting".
- Technical attribute: "photorealistic, high resolution".
three mini cases
Case 1 — Sketch 90 minutes to 5 minutes. To demonstrate a bedroom concept to a client, a designer built a rough mass model in SketchUp in 10 minutes, fed the screenshot from image to visual AI, and produced three atmospheric renderings in 5 minutes with the description "warm Scandinavian, morning light, photorealistic." Traditional rendering would take 90 minutes. Because the geometry came from its own model, the measurements were consistent; only the atmosphere has been enriched by AI.
Case 2 — Hallucination removed from presentation. In one kitchen render, AI produced a very stylish but physically impossible island: tap water was suspended in the air, there were no cabinet handles, and the stove was not connected to the hood. The designer kept this image as an "atmosphere reference" but included the measured drawing in the technical presentation and told the client that "the details will become clear in the technical project." Details were not trusted, feelings were trusted.
Case 3 — Revision accelerated by inpainting. In a living room image he liked, the customer said, "I just want the color of the sofa to be mustard yellow instead of grey, and the carpet to be darker." Instead of reproducing the entire image, the designer changed only the sofa and carpet with regional arrangement; composition and light were preserved. The 20-minute revision was reduced to 3 minutes.
Four copyable templates
1) Image from text (atmosphere exploration):
Interior image in [space type], [style/concept].Materials: [floor], [wall], [main furniture material].Color palette: [3 colors]. Light: [direction and character, e.g. softmorning light from left]. Camera: eye level, wide angle, two-point perspective. Atmosphere: [feeling]. Photorealistic, high resolution. NOTE: dimensions are not realistic; This is an atmospheric reference.
2) Image to image (turning your own sketch into a render):
Preserve the GEOMETRY and LAYOUT of the reference image I provided (my own 3D/sketch); Just enrich the material, texture, light and atmosphere. Style: [concept]. Material: [list].Light: [character]. CHANGE furniture positions and window/door locations.
3) Regional regulation (revision):
In this image ONLY [item to change, e.g. sofa] replace: [new feature, e.g. mustard yellow velvet]. Keep the rest of the image (lighting, composition, other furnishings) the same.
4) Variation series (option to customer):
Generate 3 variations for the same [space] by moving a single variable:Variation A: [wall light wood], Variation B: [wall dark green],Variation C: [wall white + texture]. Everything else (furniture, light, angle) should remain constant so that the customer can only compare this difference.
Weak prompt / Strong prompt
Weak: "Render a modern living room."
Strong: "Photorealistic living room interior image. Warm Scandinavian style; light oak floor, matte lime white walls, boucle cream sofa, black thin metal details, a single olive green accent. Soft afternoon light enters through the large window on the right. Camera eye level, wide angle, two-point perspective. Calm and spacious atmosphere. Note: atmosphere reference, dimensions are not binding."
The strong prompt style defines material, lighting, camera and atmosphere layer by layer; The result is consistent and manageable.
Comparison of image types
Method
Where it's strong
Weakness/risky place
Image from text
Quick atmosphere/concept exploration
No control of size and order, hallucination
visual from visual
Preserves your own geometry, dimension is consistent
Dependent on reference quality
Regional regulation
Fast, targeted revision
Inconsistency with major changes
traditional rendering
Size and material are exactly correct
Slow, labor intensive
Honesty when presenting to customers
The AI visual is so convincing that the customer may mistake it for the “sure result”. This subsequently creates disappointment and loss of confidence: “But that's what it looked like in the picture!” To prevent this:
- Clearly label the image as “atmosphere/concept representation.”
- Tell that the exact product, size and colors will be determined in the technical project and physical sample.
- If an item in the image (a special piece of furniture) cannot be supplied, please specify in advance.
Attention: Putting non-existent branded products, special parts that you cannot actually supply, and presenting them as "the product you will offer" in the AI image puts both customer trust and the contract at risk. Sell the soul of the image, not its impossible detail.
Common mistakes
- Entrusting the measurement to the AI visual. Mistaking the perspective in the image for real measurement.
- Not noticing the hallucination. Overlooking physical impossibilities such as suspended faucets, handleless cabinets, disconnected hoods.
- Mislead the customer. Presenting the visual as a "definitive result" and creating expectations that cannot be realized.
- Not fixing the variable in variations. Changing multiple things in each variation and the customer not being able to compare.
- Ignoring copyright/product rights. Copying exactly the registered design of a certain brand and presenting it as one's own work (we will cover the copyright issue in unit 10).
In summary
Visual generating AI makes the atmosphere visible in seconds and strengthens communication with the customer. It is used for visual exploration from text, for rendering while preserving your own geometry, and for regional editing revision. But it is a visual representation: you guarantee the size, actual product and physical accuracy. Write the prompt layer by layer, extract the hallucinations, honestly explain the nature of the image to the client.
Application task
Choose a venue. First, write an atmosphere image description with the "Image from text" template. Then, prepare a rough sketch or a simple mass model and describe the same space using the "Visual to Visual" logic, preserving your own geometry. Compare the output (or prompts) of the two approaches: which one preserves the measure better? Look for at least two physical impossibilities/hallucinations in an image you produce.
checklist
- [ ] I chose the correct method (discovery/preservation/revision).
- [ ] I wrote the prompt with layers of style, material, light, camera, atmosphere.
- [ ] If the size is critical, I preserved my own geometry with the image from the image.
- [ ] I looked for physical impossibility/hallucination in the output and weeded it out.
- [ ] I tagged the image to the client as "atmosphere representation".
- [ ] I have stated upfront the specific items that cannot be supplied.
- [ ] I moved a single variable in the variations.