Perspective · February 2026
What Makes a Generated Image Actually Good
Technical correctness is not the same thing as quality. Here is the difference — and why it matters more than the model you use.
By Look & Seen
The image generation debate mostly happens in the wrong register. People argue about which model is better, which prompt technique unlocks more realism, which tool is winning. These are interesting questions if you are a developer. They are the wrong questions if you are a brand.
The question that matters is the same one it has always been: is the image good?
“Good” Is Not a Feeling
Good is specific. In photography, in generated imagery, in any visual medium — quality is not vibes, it is a set of properties that either are or are not present. Light behaves according to physics. Color has relationships that either hold or break. Composition creates tension and resolution. These things do not change because the image was made by a camera or a model.
What changes is who is responsible for them. In traditional photography, the photographer owns the light. In post-production, the retoucher owns the surface. In generated imagery, both responsibilities collapse into the prompt — and then into the judgment of whoever is reviewing the output.
The Four Properties That Separate Good From Correct
Light Coherence
Does the light source make sense? Is the shadow where physics says it should be? Correct images often have light from everywhere. Good images have light from somewhere.
Material Truth
Does skin feel like skin? Does glass feel like glass? Models trained on stock imagery have learned to fake materials. Trained eyes catch it immediately.
Compositional Intent
Is the image arranged to communicate something, or is it arranged because that is what the model defaults to? Default compositions are safe. Intentional ones are memorable.
Color Discipline
Are the colors in relationship with each other? The difference between a frame that looks professional and one that looks generated often comes down to whether someone made a color decision or accepted the model's default.
Why This Is a Skill Problem, Not a Tool Problem
Every model available today can produce a technically correct image. Very few people can look at that output and identify the three things that are slightly wrong and know how to fix them. That is a trained eye. It takes years to build.
The brands producing the best AI imagery right now are not the ones with access to better models. Every major agency has access to the same models. The differentiator is the people in the loop — specifically, whether those people have the visual vocabulary to evaluate output against a standard that goes beyond “this looks fine.”
“This looks fine” is not a standard. It is a default.
What to Look For in a Partner
If you are working with someone to produce AI-generated imagery for your brand, ask them one question before anything else: show me something you rejected.
The answer tells you everything. A generator who has never rejected output does not have a standard. A retoucher who cannot articulate why one frame is better than another cannot be trusted with your brand. Good work requires the ability to say no to mediocre work — clearly, specifically, with reasons.
We built Look & Seen around the idea that craft travels. The skills that make a great retoucher — the eye for light, the understanding of material, the discipline around color — apply directly to generating and evaluating AI imagery. The tools are new. The standard is not.