Gadgetify shares a GPT Image 2 prompt for figure-ground optical illusions in a 2×2 grid

In the early hours of June 19, Gadgetify (@Gdgtify) posted on X a new round of experiments with optical illusions generated by GPT. «I didn't even pick the subjects. AI figured it out”—did not choose the topics; The AI ​​deduced them. The prompt calls for a 2×2 grid with four pairs of objects in which the black solid shape and the white empty space offer two equally valid readings, in the style of Rubin's glass or FedEx's hidden arrow.

AI Prompt · How to do it

Example posted by Gadgetify on June 19, 2026. @Gdgtify

Steps

1. Open chatgpt.com with GPT Image 2 active.
2. Copy the prompt below as is — the AI infers the four pairs of objects.
3. For a single panel, remove "2x2 grid" and specify [Object A] and [Object B] (e.g. "a violin" + "a woman's face").
4. If grays or shadows appear, reinforce the Negative block at the end.

Prompt to copy

2x2 grid, do this for 4 AI inferred topics: Anchor: A black-and-white figure-ground illusion where the solid shape reads as [Object A] and the empty space around it reads as [Object B] ::5 Duality Rule: Both readings must be equally valid — ink is [Object A], void is [Object B]. One contour, two pictures ::5 Composition: Centered, near-symmetrical, generous breathing room ::2 Material Physics: Pure solid black ink, no gradients, no texture, razor-clean edges ::3 Illumination: None — flat 2D graphic ::1 Render Stack: Smart logo design, Rubin's-vase illusion, FedEx-arrow cleverness ::2 Negative: [gray tones, shadows, 3D, gradients, detail noise, more than two readings] :: -1

Syntax with ::n weights (Midjourney style). In English according to the original post. Results vary by ChatGPT plan and region.

What makes this prompt special

Unlike a conventional logo, here the rule of duality (::5) requires that ink and void have the same perceptual weight: it is not enough for an object to be seen; The negative contour must form another equally recognizable one. The Render Stack block explicitly points to the Rubin's glass and the FedEx arrow—classic references of double-reading design.

The Material Physics (::3) and Negative (::-1) weights combat common AI failures in this type of graphics: intermediate grays, 3D shading, detail noise or more than two interpretations. The result is a flat, binary, almost typographic graph.

Relationship with other Gadgetify prompts

It's the same family of 2x2 templates with :: weights that Gadgetify already tested with knolling and themed silhouettes: four AI-inferred themes in a single image. Here the challenge is not the overhead composition of objects, but the perceptual ambiguity of a single shared contour.

A/B Pair Ideas

• Violin ↔ woman's face (classic Rubin vase)
• Guitar ↔ mono
• Tree ↔ two people facing each other
• Arrow ↔ letter E (FedEx style)

Before publishing

The illusions generated can accidentally evoke recognizable faces or brands. Review the result before using it in commercial campaigns and respect OpenAI's policies for image generation.