What is Hallucination (in rendering)?
When an AI invents details that were not in the input — an extra window, a shifted wall. Architecture-focused tools minimise this with geometry-preserving pipelines; general image tools do not.
Updated 4 November 2026
In AI rendering, a hallucination is a detail the model adds that was not in your input or prompt: an extra window, a shifted door, a railing that appears from nowhere, or distorted text on a sign. The model fills gaps with what looks statistically plausible, not with what your design says.
Why it matters
For architecture, accuracy is the point. A client who approves an image with a feature that is not in the design will expect that feature. Hallucinations that slip into a presentation can cause disputes, redesign work and lost trust, so catching them is part of the process.
Example
A render of a facade shows five windows in the upper row where the drawing has four. Comparing the render and the elevation side by side reveals it immediately.
Common mistake
Skipping the check. Always compare against the original, focusing on openings, rooflines, counts and proportions. If something is wrong, rerun with a clearer prompt, reduce the number of requests in one pass, or fix the area with inpainting. Do not rely on the render to prove compliance with the design.
Practical tips
Build a short checklist for review: counts of windows and floors, door positions, rooflines, balcony and railing details, signage, and anything added to the background. Review at full size, not as a thumbnail. Keep the original drawing next to the render when you check, and record any approved deviations so that nothing unexpected is carried into the next stage.
In Archome AI
Archome AI uses geometry-preserving pipelines to reduce this, and Inpainting lets you correct one area without redoing the image. Free accounts get 3 free renders, no card.