A visualization brief is often assembled from a camera sketch, a handful of references and an adjective such as “welcoming.” That may be enough to start a conversation, but it is not enough to govern an AI-assisted workflow. A generator can satisfy the adjective while changing the entrance, concealing the servicing route or replacing a difficult facade junction. The result may feel right while answering the wrong question.
Start instead with a decision sentence: “This view will help the project team compare the perceived depth of two facade options from the pedestrian approach.” That sentence identifies an audience, a comparison and a viewpoint. It also limits the claim. The view is not automatically evidence of daylight performance, accessibility or planning acceptability.
Name the stage, not just the mood
The RIBA Plan of Work offers a useful vocabulary for discussing project maturity. An early concept image and a coordinated technical presentation should not look equally certain simply because the software can finish both convincingly. Record the actual project stage, the information available and the decisions still open. Do not imply that referencing a framework makes a particular image compliant with it.
Divide the brief into three categories: fixed information, controlled alternatives and atmospheric interpretation. Fixed information might include the site boundary, approved massing and entrance locations. Controlled alternatives might be two cladding modules. Atmosphere might include a proposed lighting condition or planting character. Every category needs an owner who can approve a change.
A reference should explain a quality to examine, not become a building to imitate.
Use references as evidence
Architects often discuss light through precedents such as Louis Kahn’s Kimbell Art Museum in Fort Worth, or material continuity through Peter Zumthor’s Therme Vals. Those are useful conversations only when the reference is precise. “Study the rhythm of indirect daylight” is different from “make it look like the Kimbell.” Neither statement proves that your roof geometry can produce the same effect.
Annotate each reference with the quality being borrowed, the qualities being excluded and the source’s permitted use. Separate photographs you may circulate from images you may merely discuss internally. If a reference contains identifiable people, private interiors or another client’s work, check permissions before uploading it to an external service. A convenient mood board is not a licence.
Choose a camera that admits the question
A low viewpoint may emphasize civic presence while hiding a service enclosure. A wide lens may make a small room seem generous. Neither choice is inherently dishonest, but both should be intentional. Record the camera position, target, focal length or field of view, and the relationship to the model coordinates. Include a simple location diagram when the audience will not know the site.
For a public-facing view, consider whether a second camera would reveal what the first excludes. A pedestrian approach and a service-side view can tell a more responsible story than several variations of the same flattering corner. Where a view is intended to support a formal visual assessment, obtain the relevant specialist methodology rather than adapting a marketing composition after the event.
Make acceptance testable
Information management principles in the ISO 19650 series can inform how a studio labels revisions, responsibilities and exchanges. They do not prescribe a universal prompt or guarantee a correct image. Use a proportionate record: source model identifier, camera identifier, reference list, permitted transformations, reviewer and intended output.
- Decision: What should the reviewer be able to compare or understand?
- Source: Which model revision and site records establish the baseline?
- Limits: Which elements must not be invented, removed or relocated?
- Approval: Who checks architecture, communication and release separately?
Finally, agree what failure looks like. A shifted window grid, an inaccessible threshold or an apparently occupied area that has no safe route should trigger revision even if the image is attractive. Set a review budget and a stopping rule. Without them, “one more version” can become a substitute for resolving the design.
The brief should remain short enough to use at the review table. Attach supporting records rather than burying the essential distinctions in a long document. Its job is to protect the connection between the picture and the decision, while giving the visualization team a clear, legitimate space in which to experiment.