ChatGPT (DALL·E) · GPT Image 2

Underground Cavern Map

Tested: 2026-08-09

3:2 · 768 × 516

AI-generated image for the prompt "Underground Cavern Map", produced with GPT Image 2
Generated with ChatGPT (DALL·E) · GPT Image 2 at 3:2. Full prompt below.

The prompt

Prompt

An underground cavern system map drawn in top-down cross-section view, a network of tunnels branching from a central cavern, the main cavern large and irregular with a subterranean river flowing through it drawn in pale blue with a ripple pattern, a waterfall dropping from a higher passage on the east side drawn as parallel blue lines, a cluster of stalactites marked as small triangles hanging from the cavern ceiling in the north chamber, a narrow passage to the south leading to a smaller dry chamber with drawn crystal formations as small geometric gems, a pit in the centre drawn as a dark circle with a depth-marking cross-hatch, three tiny drawn mine-cart rails as parallel lines leading from a tunnel entrance at the top, a small blank rectangular area in the main cavern for a label, no text, no letters, no room names, hand-drawn cartography style with ink linework on aged grey parchment

How to use this prompt

  1. 1

    Open the generator that matches the model and version above, and start a new image rather than editing an existing one.

  2. 2

    Set the output to 3:2. That is the only output setting recorded for this entry, and it is what produced the 768 × 516 image on this page.

  3. 3

    Paste the prompt exactly as written. Changing the wording changes the image, so copy it before you edit.

  4. 4

    Generate, then change one thing at a time. The result will not be pixel-identical to the image on this page, because no seed is recorded for it.

Only an aspect ratio is recorded for this entry. GPT Image and DALL·E do not expose a sampler, step count, or CFG value in the interface this was tested in, so there are no numbers to match — the prompt text and the aspect ratio are the whole instruction. Resize by describing the framing in words rather than reaching for a parameter.

More prompts from the same model, and prompts that share a use case with this one.