Public development record · 2026-08-18

Why returning “just an image” broke the accessibility contract

This page preserves the full context around a real early-1990s photograph of Ken working as a waiter in Key West, later AI transformations of that photograph, a fictional HeyGen avatar and video, and an interaction failure in which the assistant returned a newly generated image with no explanation, no HTML context, and no package.

For a person using AI as external memory, the context around an artifact is part of the artifact.

1. The real source photograph

Original public photograph supplied by Ken, showing him as a young waiter in Key West. — **Friendly context:** This is the original public photo Ken supplied. He describes it as what he looked like while working as a waiter in Key West before being hired directly from that job by INFOCALL, where he later worked as Technical Director. The second person is intentionally obscured in the public image.

Exact source: https://8k.art/beta/_images/Ken-as-Waiter.jpg

Browser screenshot showing the original Ken-as-Waiter image on 8k.art. — **Friendly context:** This browser screenshot preserves how the original was actually being viewed during the conversation, including the public 8k.art URL.

2. AI enhancement and fictional reconstruction

AI-enhanced version of the original waiter photograph. — **Friendly context:** This is an AI-enhanced/restored interpretation based on the old photograph. It is useful as a demonstration of what modern tools can recover or reconstruct from a small historical source, but it should remain linked to the original so viewers can distinguish source evidence from later enhancement.

Exact source: https://8k.art/beta/_images/Ken-as-Waiter-2.jpg

Browser screenshot showing the enhanced waiter image on 8k.art. — **Friendly context:** This screenshot preserves the larger enhanced presentation as it appeared in the browser during the conversation.

Friendly context: This image is explicitly fictional. Ken explains that the HeyGen-style AI avatar was produced from little more than the old photograph and textual/YouTube context. It is not a historical photograph of an actual scene.

Exact source: https://8k.art/beta/_images/Fictional-Ken-as-Waiter-AI-Avatar.png

Friendly context: This captures the fictional avatar in the same browser context in which Ken introduced it.

3. The exact fictional avatar video and playlist

Friendly context: This is the fictional AI-avatar MP4 Ken supplied. It is preserved here alongside the still images instead of being detached from the story that explains what it is.

Exact MP4: https://8k.art/beta/_images/Fictional-Ken-as-Waiter-AI-Avatar-Video.mp4

Exact YouTube video

Exact YouTube URL: https://www.youtube.com/watch?v=ZAI5KxiVzX8&list=PLI-3LueqVb9YY6cInw-ALyA48R_SavgtI

Exact YouTube playlist

Playlist ID: PLI-3LueqVb9YY6cInw-ALyA48R_SavgtI

Friendly context: This is the exact YouTube screen Ken showed after the assistant had failed to preserve context. The fictional waiter-avatar video is playing while the broader playlist remains visible on the right.

4. The human context Ken was trying to preserve

Ken described working as a waiter in Key West in the early 1990s before being hired by INFOCALL. INFOCALL built airport kiosk-style telephone systems connecting travelers with food, transportation and hotels. Ken described working there with Macromedia Authorware on early Windows systems.

He also described a longstanding memory-related social problem: people sometimes remembered his name by joking that he looked like a human “Ken doll,” while he himself had difficulty remembering names and faces when people were outside the context in which he knew them. That is precisely why the historical photo, the story, the later AI reconstructions, and the present-day commentary should remain connected instead of becoming isolated media files.

5. What the assistant did wrong

The immediate mistake: the assistant chose an image-generation action when the task context required preservation, commentary, HTML context and packaging. That was the wrong tool/workflow choice.

What happened next: the image-generation path is designed to surface the generated image itself as the visible result. Once that wrong path was chosen, the normal explanatory response was effectively displaced. The result was exactly what Ken had repeatedly said not to do: a standalone generated picture with no surrounding explanation.

This was not caused by missing context from Ken. The conversation already contained the standing rule that supplied images and links should be echoed back with friendly context, and that generated graphics for this project should not be delivered as orphan images without HTML/package context.

Assistant-generated fictional waiter portrait that was returned without the required explanatory context. — **Friendly context:** This is the image the assistant generated in the failed response. The image itself was not the main problem. The problem was that it appeared alone, severed from the original photograph, the fictional-avatar explanation, the video, the playlist, and the user's accessibility requirement that artifacts retain their context.

6. The permanent rule this failure demonstrates

IMAGE OR VIDEO → echo it visibly → explain it in friendly language → preserve the original URL → connect it to the surrounding dialogue → package the context in HTML.

For public dialogue archives, images belong at the point in the conversation where they were introduced. Their captions should explain whether they are original evidence, an AI enhancement, a fictional reconstruction, a screenshot, or a generated derivative.

The broader accessibility principle is simple:

If the user saw it in the conversation, the conversation should remember not only the pixels, but what those pixels meant.