Closer #017

Welcome to Closer,

A weekly experiment in AI interpretation: I share a photo, ask AI to describe it, then generate an image from that description. This is edition #017. Prompts are in grey boxes. Descriptions are in italics.

Chat gave me six different Prompt Styles and we're trying them all on the same image over the upcoming weeks. The first week was hilariously bad, the second week was uneventful, the third week made me question the continued value of this newsletter. Is that feeling reaffirmed or deflected this week?

LLM models used

GPT-4o for prompt engineering, image interpretation, and image generation.

The input with Prompt Style: Haiku + Description

Purpose: Contrast formality and feeling — great for newsletter rhythm.

The output

Haiku
Warm waves hold laughter,
Two wet dogs, a man, and joy—
Salt and sun collide.

Factual Description
In the image, a man wearing sunglasses, a visor, and a blue flotation sleeve is in the ocean with two Labrador Retrievers. They are near a white motorboat with dual Suzuki outboard engines and an American flag flying from its roof. The mood is relaxed and joyful, suggesting a leisurely day spent swimming and boating. The calm, turquoise water and the gentle sky enhance the serene and playful atmosphere.

Impressions

If the goal is accurate image generation, haikus aren’t exactly the best way to get there. We've seen that AI needs very specific instructions, to achieve our goal. Do we still have clear what that goal is? Let's revisit.

We’re exploring how closely AI can approximate the look and feel of real-life imagery. To do that, we need the most accurate descriptions of those real images, and to get those, we need the best possible prompts for AI to analyze them effectively.

Here's a haiku that describes my feelings towards this week's approach:

The form is lovely,
but not built to brief machines—
too much left unsaid.

See you next week.

Mitch