OpenAI Image Model 1.5 – Strong in Quality, but with Clear Limits for Architecture

With OpenAI Image Model 1.5, OpenAI has made another significant leap in the field of AI-based image generation. Of course, we have thoroughly tested the model and compared it with other models used at Vizera.

Our conclusion upfront:
The image quality is convincing, especially for creative and illustrative applications.
However, for professional architecture and interior workflows, there are currently still relevant limitations.

In this post, we explain transparently where the strengths lie – and why we continue to prefer other models for architecture.

Image Quality: Visibly Improved

Visually, Image Model 1.5 delivers very good results:

• realistic lighting mood

• clean color reproduction

• overall high image coherence

• significantly better than previous OpenAI models

For mood images, concept visualizations, or general creative applications, the model is absolutely competitive.

Aspect Ratios: Limited Flexibility

A central point for architectural visualizations is the aspect ratio. Plans, facades, interiors, or presentation images require different formats.

Supported Aspect Ratios:

• 1:1 (Square): supported

• 3:2 (Landscape): supported

• 2:3 (Portrait): supported

Currently not supported are, among others:

• 4:3 / 3:4

• 16:9

• 9:16

• 21:9

Especially 16:9 and 4:3 are essential for architecture, presentations, and competitions. Here, the limitation is clearly noticeable.

Resolution: Limited for Professional Output Formats

There is also a clear limitation regarding resolution:

• 1K: available (square)

• 1.5K: available (Landscape / Portrait)

• 2K: not available

• 4K: not available

For architectural projects, 2K and 4K outputs are often necessary – whether for print, competitions, or large-format presentations.
This requirement is currently not met by Image Model 1.5.

Generation Speed: Too Slow for Iterative Workflows

A crucial factor in the daily life of architects is speed.

Our tests show:

• 1K output: approx. 30–45 seconds

• 1.5K output: approx. 50–60 seconds

For iterative processes – multiple variants, material studies, facade options – this is clearly too slow.
Especially in comparison to other models, this point is significant.

Reference Images & Control

Positively noteworthy is:

• Reference images are supported

• a configurable fidelity level allows for closer adherence to the original

However, there are limitations regarding:

• maximum 5 reference images at high fidelity

For complex architectural projects, where multiple references (materials, typologies, contextual images) are necessary, this is limiting.

Why We Currently Prefer Other Models for Architecture

Vizera is intentionally model-agnostic. We always integrate the models that objectively deliver the best results for the respective use case.

Currently, the following points speak against Image Model 1.5 as the primary model for architecture and interior design:

• restricted aspect ratios

• missing 2K and 4K outputs

• slow generation times

• limited scalability for professional workflows

Other models we use offer:

• significantly higher flexibility in formats

• better resolutions

• faster iterations

• higher reliability in proportions and spatial logic

Our Conclusion

OpenAI Image Model 1.5 is a powerful model with clear advancements.
For creative applications, it is already very well suited.

However, for professional architectural visualization, it currently does not meet all the requirements that architects need in their daily work.

We are closely monitoring its development – and as soon as an OpenAI model meets these points, it will of course be integrated at Vizera.

Our commitment remains unchanged:

Always the best model for architecture – regardless of the provider.