GPT Image 2 Prompt Examples: 12 Templates That Actually Work

What's actually new with GPT Image 2

OpenAI launched GPT Image 2 (also branded as ChatGPT Images 2.0) on April 21, 2026. It's the third generation of OpenAI's image models, following GPT Image 1 (April 2025) and GPT Image 1.5 (December 2025).

The two genuinely new capabilities are:

1. Built-in reasoning before generation. This is the first OpenAI image model to integrate O-series reasoning into the image pipeline. Before generating, the model analyzes the prompt, plans the composition, and reasons about constraints — which is why it handles complex layered prompts (UI mockups, multi-element layouts, scenes with text) much better than its predecessors.

1. Multilingual text rendering at production quality. Text inside images now renders accurately across Latin, Chinese, Japanese, Korean, Hindi, Bengali, and Arabic scripts. Independent reviews put accuracy above 95% on first-generation attempts. Mixed-script layouts (Japanese poster with English brand name, etc.) actually work now — they didn't in any previous model.

The model is available in ChatGPT, Codex, and the API. Pricing on the API is $5 per million text input tokens, $8 per million image input tokens, and $30 per million image output tokens — which works out to roughly $0.006 to $0.21 per generated image depending on quality and resolution. (Source: OpenAI pricing page, April 2026.)

This article gives you 12 prompts that work well with GPT Image 2's reasoning model — battle-tested across the categories most creators care about.

The structural pattern these prompts follow

Every prompt below shares the same underlying structure:

[Subject + specifics] + [Action] + [Environment + cultural anchor] + [Composition] + [Lighting] + [Style/medium]

For photoreal: add camera body, lens, aperture. For text in images, wrap the copy in quotes and specify font/weight/color/placement. For edits, use the CHANGE / PRESERVE / MATCH structure.

This is a synthesis of patterns we've seen work across hundreds of prompts — not an industry standard, but a useful framework for thinking about what's missing in any given prompt.

1. Event poster

Rough idea: "poster for a tech conference"

Polished prompt:

Why it works: All text is in quotes with placement, weight, and color locked. The hex code prevents palette drift. "Swiss design aesthetic" handles composition without needing more rules.

2. Process flow infographic

Rough idea: "infographic showing a 4-stage workflow"

Polished prompt:

Why it works: Layout topology is stated upfront. Each stage has explicit fill, label, and icon. Connectors specified. The model handles the rest.

3. Web dashboard mockup

Rough idea: "dashboard mockup for a SaaS analytics product"

Polished prompt:

Why it works: Specific device frame, specific component layout, specific data. The shadcn reference handles styling without you having to describe every spacing decision.

4. LinkedIn carousel slide

Rough idea: "slide with a stat about productivity"

Polished prompt:

Why it works: The stat-as-hook follows what actually performs on LinkedIn. Two-tier typography (bold + italic serif) creates rhythm. Negative space gives the eye room.

5. Cinematic still

Rough idea: "cinematic shot of a wedding"

Polished prompt:

Why it works: Pose, gaze direction, and clothing are specific. 50mm + f/1.8 locks the cinematic feel. Two named light sources prevent flat lighting.

6. Interior photograph

Rough idea: "calm reading nook"

Polished prompt:

Why it works: Specific materials throughout. Light source named with quality. The worn book detail makes it feel lived-in.

7. Food photography

Rough idea: "breakfast spread"

Polished prompt:

Why it works: Universal breakfast items recognizable across cultures. Specific positioning. Light direction named.

8. Background swap edit

Rough idea: "change my photo's background to a rainy city at night"

Polished prompt:

Why it works: The PRESERVE list is exhaustive — every locked element named. The MATCH block prevents the "pasted in" look by addressing color spill from the new environment.

4 more shorter examples

Concept art: Concept art illustration, 16:9 cinematic, matte painting style. Sprawling low-profile research facility built into the side of a rocky crater on an alien planet, glowing blue lights through narrow window slits, solar arrays across rust-red plain. Two figures in white pressure suits walking toward an airlock. Sky: dusty pink with two small moons.

Magazine cover: Editorial magazine cover, 8.5x11 portrait. Masthead "WANDER" in bold sans-serif, white, top center. Headline "Mountains at Dawn" in elegant serif italic, lower left third. Sub-headline "12 hidden viewpoints" in thin sans-serif beneath. Cover photo: silhouetted mountain range against deep blue-to-orange gradient sky.

Comparison infographic: Two-column comparison infographic, 4:5 portrait. Left header "OPTION A" in blue (#2B6CB0); right header "OPTION B" in amber (#D69E2E). Vertical divider. Five rows: Speed / Cost / Flexibility / Best For / Trade-off. Statements max 6 words each.

Object removal edit: CHANGE: Remove the green trash can in lower-left and the power line crossing the upper third. Reconstruct the wall behind using existing brick pattern; reconstruct sky as clean continuous gradient. PRESERVE: Every person, the storefront awning, parked car, street signs, sidewalk, all shadows. MATCH: Golden hour from camera-right, warm temperature, film grain.

How to use these

Don't just copy them verbatim. Use them as templates: find the example closest to your use case, swap out the subject specifics for your own, keep the structural skeleton (camera specs, lighting language, aspect ratio), and iterate by changing one variable at a time.

Or paste your rough idea into Depikt and let it generate the structured prompt for you.