Field Note · August 3, 2026
Kate kept passing at full resolution and failing where readers actually saw her.
The fix was not more realism. It was a better production boundary. This is the useful part of the failure: what looked like a cosmetic problem exposed a systems problem, and the correction changed how we will build Kate from this point forward.
The convincing master
At 100 percent, the image looked finished.
Kate looked fantastic when the file was large enough to inspect every eyelash, highlight, and fold. The pose was right. The environment was right. The editorial intention was right. That made the image easy to trust.
But a website does not display a master file at inspection size. It makes thumbnails, feed cards, responsive crops, social previews, and compressed versions. A visual identity has to survive all of them.
Newbie translation: the big image was the showroom model. We had not yet taken it onto the road.
The scale failure
The pixels did not shrink politely.
When the image was reduced, neighboring tones were forced to merge. Tiny transitions across Kate’s skin stopped reading as smooth form and began reading as bumps and dark bands. Hair detail, jacket texture, specular highlights, compression, and local contrast all competed for too few pixels.
Technically, this is a scale-dependent failure. Information that was harmless—or even attractive—at full size became unstable after resampling. The environment held together because desks, windows, and envelopes can lose small details without changing identity. A face cannot.
Newbie translation: the room only had to remain a room. Kate still had to remain Kate.
Why Kate is the test
We made the hard thing hard on purpose.
Kate has always been our ultimate consistency test. She is intentionally more demanding than a logo, an icon, a product photograph, or a simple brand illustration. Her face, hair, body proportions, skin, expression, jewelry, wardrobe, lighting, and environment all have to agree—then keep agreeing after the image is cropped and reduced.
If a production system can keep Kate recognizable and coherent, less complex brand imagery becomes comparatively simple. That makes her valuable beyond the image itself. She reveals where a workflow is pretending to be reliable.
The useful overcorrection
Less realism worked. Then it went too far.
Moving toward illustration reduced the fragile micro-texture that kept breaking at smaller sizes. Kate became cleaner and more stable. The lesson was correct.
But the entire room moved with her. That solved one problem by changing the visual language of the whole scene. It was useful evidence, not the final production boundary.
What we actually wanted was more precise: Kate can be drawn into the world without forcing the world to become a drawing. Her identity surfaces can be deliberately simplified while real clothing, furniture, materials, light, and atmosphere keep their dimensional weight.
The boundary
Kate is becoming less realistic so she can become more herself.
The goal was never photorealism for its own sake. The goal was a character who remains herself wherever the work appears.
The new direction keeps the office, table, paper, clothing, shadows, and environmental light grounded. Kate’s face, hands, and identity-defining surfaces receive a more controlled illustrated treatment. Think less “cartoon universe” and more “a drawn character photographed inside a real production.”
That gives the eye enough realism to believe the scene while removing the fragile skin detail that failed in thumbnails.
The production change
One clever prompt is no longer the workflow.
Future Kate production moves to a compositing standard. That means we stop asking one generation to solve identity, anatomy, wardrobe, pose, environment, lighting, materials, and final scale behavior in a single pass.
- Build the worldCreate the realistic environment and its lighting as a stable plate.
- Build KateCreate the controlled character plate from her approved identity and body standards.
- Join and testComposite the plates, match contact light and shadow, then inspect the result at full size, 33 percent, and feed-card size.
For the technical reader: the gain is separation of concerns, deterministic assembly, narrower failure domains, and testable output gates. For Suzie: if Kate’s face fails, we can fix Kate without rebuilding the whole office.
What remains open
The perfect renderer remains an open question. Production does not.
We will keep evaluating more deterministic rendering paths—including whether a tool such as Blender eventually belongs in the system. That is later work. We are not stopping useful publishing today while waiting for a theoretical perfect engine.
The practical standard is now stronger than the tool: separate the problem, preserve the approved identity, composite deliberately, and reject anything that fails where the audience will actually see it.
Imperfection is not the embarrassing part. Hiding the evidence and repeating the mistake would be.
No Lies. No Stealing. Also no pretending the first version was fine.
Behind the work
Jacqueline & Kate
Jacqueline did the work and formed the point of view. Kate helps make it visible.
OutloudMarketer documents the decisions, experiments, failures, and useful conclusions behind what Black Cat Media is building.
No Lies. No Stealing.