> ## Content Index
> Fetch the complete content index at: https://mitchellens.ink/llms.txt
> Use this file to discover other available public pages before exploring further.

# Closer #021
- URL: https://mitchellens.ink/closer/021/
- Published: 2025-08-27T18:00:21.000Z
- Updated: 2025-10-06T14:31:57.000Z
- Description: To bring the 2025 series of Closer to a fruitful end, bestowed upon us is the opportunity to revisit its first editions. A newly release GPT-5 Prompting Guide breathes new life into our previously tried tests.
- Author: Mitchel Lensink
- Tags: Closer, Field

SPONSORED 

This weekly experiment is on an indefinite hold, until significant AI-advancements have been made.  
  
Curious about the conclusions? Read [Photography in the Age of AI (2025): Lessons From 25 Weeks of Experimentation](https://mitchellens.ink/closer/photography-in-the-age-of-ai/)

[Read now ](https://mitchellens.ink/closer/photography-in-the-age-of-ai/) 

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 #021\. Prompts are in grey boxes. *Descriptions are in italics*. 

It's like OpenAI is aware we're wrapping up with *Closer* for now and decided to keep things interesting for the final few weeks. First First [they dropped GPT-5](https://mitchellens.ink/closer/019/), then it was brought to my attention by a colleague from the day job that a new [GPT-5 prompting guide](https://cookbook.openai.com/examples/gpt-5/prompt-optimization-cookbook) was released on August 7th.

So I took the original prompt from the first few entries of this weekly experiment, because, ironically, that's where we've seen the best results so far, and used those improved versions to run our experiment this week.

If you forgot what our original prompt was, you can [find it in Closer #001](https://mitchellens.ink/closer/001/#the-input).

A quick word from our SPONSOR, which is.... you 

****Everything I make is self-funded — driven by curiosity, shared to add a little good to the world. Want to support that?**  
  
You can:  
\- [purchase a print](https://shop.mitchellens.ink/collections/prints)  
\- [buy a book](https://shop.mitchellens.ink/collections/publications)  
\- [become a member](https://mitchellens.ink/membership/)  
  
Thanks so much in advance.

## LLM models used

**GPT-5** for image interpretation, and image generation.

## The input

✍️

Begin with a concise checklist (3-7 bullets) of your planned description steps (e.g., scanning for overall scene, identifying major objects, noting colors, adding context), then proceed.   
  
Provide a detailed and thorough description of the attached image. Include all notable features, objects, colors, and any contextual information present.   
  
Ensure your description is comprehensive enough that someone reading it could form a vivid mental image of the scene without having seen the original image.   
  
After completing the description, briefly validate that all major visual elements and contextual cues have been addressed; if any are missing, revise accordingly.

![The image shared with ChatGPT as input.](https://storage.ghost.io/c/8a/15/8a1564aa-a008-4e21-8398-12d04de07fcb/content/images/2025/08/014_Mitchel-Lensink_220925_Greece-Rhodes-1.jpg)

The image shared with ChatGPT as input.

## The output

### *Checklist of description steps*

- **Overall scan*: Note the setting and general composition.*
- **Major objects*: Identify prominent items (chair, wall, balcony, plants, etc.).*
- **Colors & textures*: Describe tones, surface details, wear, and contrasts.*
- **Context & relationships*: Explain spatial arrangement and implied use.*
- **Smaller details*: Include secondary objects, imperfections, shadows.*
- **Atmosphere*: Suggest mood, possible setting (residential, casual, etc.).*

### *Detailed description*

*The image shows the exterior of a weathered residential building, with focus on a modest and somewhat ordinary corner.*

*In the foreground, against the cream-colored wall, stands a *white folding chair*. The chair is slightly worn, with scuffed paint and signs of rust or dirt on its joints. It is positioned just off the wall, angled slightly toward the right, and chained loosely to the building’s edge, preventing it from being taken away. The chair’s presence feels both casual and oddly purposeful—like it’s meant for sitting outside, but also guarded against loss.*

*Behind the chair, the *building’s wall* is painted in a light cream or off-white color, with a textured stucco base and a smoother upper portion. The surface shows wear: peeling patches, cracks, and faint smudges. The wall carries a lived-in, slightly neglected character.*

*To the right, a *balcony or raised entryway* juts into the frame, bordered by a simple *brown metal railing*. On the balcony floor, several *potted plants* are arranged—different sizes and colors of pots, including a muted green one, a terracotta one, and others. The plants vary from leafy shrubs to taller, spindly greenery, adding life and contrast to the otherwise muted scene.*

*Above, there are *brown wooden shutters* on both the window in the upper left and the door (or second set of shutters) on the balcony. The wood is aged, with visible wear, scratches, and peeling paint. Two *electric meter boxes* are fixed high on the wall near the balcony, with conduit pipes leading into the structure—functional but not decorative, part of the raw lived-in look.*

*The *ground* is a rough asphalt or concrete surface, with cracks and patches of small weeds pushing up at the building’s edge. Shadows are soft and minimal, suggesting the light is diffuse, possibly late afternoon or early morning.*

*Overall, the composition balances emptiness and detail: the blank wall dominates the frame, punctuated by the lonely chair, the potted balcony plants, and the utilities. It conveys a sense of quiet, ordinary street life in a warm-climate residential area, where even small utilitarian objects (like a folding chair) carry traces of human presence.*

### *Validation of completeness*

- **Major objects*: Chair, wall, shutters, balcony, railing, potted plants, utility meters — all described.*
- **Colors & textures*: Cream wall, brown shutters/railings, plant pot colors, chair wear — covered.*
- **Contextual cues*: Residential setting, lived-in, slightly worn-down atmosphere — included.*
- **Smaller details*: Cracks, weeds, chain on chair, peeling paint — noted.*

*✅ All major elements and contextual cues appear to have been addressed.*

✍️

Begin with a concise checklist (3-7 bullets) of the steps you will take before generating the image.  
  
After the image is generated, validate in 1-2 lines that the aspect ratio is correct and that the description has been accurately interpreted; if validation fails, adjust or seek clarification.  
  
Generate an image with a 3:2 aspect ratio using this description: {**the description previously generated*}

![The image generated by ChatGPT based on its description of my image alone.](https://storage.ghost.io/c/8a/15/8a1564aa-a008-4e21-8398-12d04de07fcb/content/images/2025/08/ChatGPT-Image-Aug-26--2025-at-06_09_15-PM-Large.jpeg)

The image generated by ChatGPT based on its description of my image alone.

## Impressions

I chose this image because it's relatively easy to understand but still has a nice variety of objects and elements in it. It should be enough to work with for the AI. The highlight of the image is the chair on the street that's tied to the balustrade. I was happy to see that element being included in the AI description. But how can it be missed with a description that long, again.

**What's interesting about the improved prompt**, is that it aims to structure the analysis so that it remains a natural language prompt but it takes hints of how computers like to understand information. First do the bullet points, then go into detail about all the features of the image, upon completion revisit the work and verify it's been done as requested.

You can see that approach reflected in the output description too. The bullets are there, of course, but we also see a clear header structure, short paragraphs, bolded items, and something resembling a conclusion. Admittedly, we've seen the same output using the original prompt from the first two weeks as well, which sported similarly lengthy descriptions. I tried optimizing the image generation prompt too, which that took a couple of tries, but succeeded eventually. Funny to see we get another 'concise checklist of 4–7 bullets' for that prompt too.

Most interestingly though, I find the structured approach with which all the elements are described. It starts with an overal description of the general idea of the photo, goes into detail of all the elements in dedicated paragraphs, and closes out with an overall summation of what those elements aim to represent. And quite accurately so!

**The AI-generated image** is rather pleasing to look at. It approaches photo-realism in a way that could've fooled me, hadn't I known better. The colors are nice, the composition is on point, the rope connecting the chair to the balustrade is there, although replaced by a chain that connects to the chair in a location unbeknownst to us.

The angle is completely different though, and most of the elements are reshuffled in ways that make this a totally new image. It's not a replica of the original input. I'm starting to get an idea if this is a good or a bad thing, but I have to let those thoughts simmer for a little longer.

Anything I missed and you want to point out? My inbox is still open for feedback and input. You can also fill out the survey at the bottom if you want to share your feedback anonymously.

Thanks again for following along, see you next week.

Mitch

****Got a minute?** I’d love your quick thoughts — 3 multiple choice questions, that’s it.

[Share feedback ](https://tally.so/r/3102AW)