ChatGPT Image 2 (GPT Image 2) processes prompts with high instruction-following fidelity — it handles multi-part, complex descriptions reliably. The prompts below are organized by category to match the model's core strengths.
Why Prompt Structure Matters for GPT Image 2
Unlike diffusion models that respond to weighted keywords, GPT Image 2 processes prompts more like a language model. This means:
- Full sentences outperform keyword lists — describe the scene like you're explaining it to a photographer
- Negations work — "no people", "without shadows", "not centered" are interpreted correctly
- Text content should be quoted exactly — the model will attempt to render the exact text you specify
- Lighting and camera directives translate — "studio lighting", "natural side light", "overhead view", "shallow depth of field" produce predictable results
25 ChatGPT Image 2 Prompts
Product Photography (6 prompts)
A minimalist perfume bottle — clear hexagonal glass with a silver cap — on a white marble slab. One eucalyptus sprig beside it. Soft diffused lighting. Commercial beauty photography.
A leather wallet, dark brown, open to show card slots. Flat lay on a light grey linen surface. Natural overhead lighting. Clean shadows.
A protein powder tin with black and gold branding, standing upright on a gym floor. Blurred weight rack in the background. Dramatic side lighting.
A ceramic coffee mug, matte white, filled with black coffee. Steam rising. Placed on a wooden saucer with a small spoon. Natural morning light.
An open hardback book resting on a reading chair arm. Warm lamp light. Reading glasses resting on the open page. Cozy interior background.
A running shoe, white with orange accents, suspended in mid-air against a clean white background. Shadow below. Dynamic angle — shot from 45 degrees.
Interior and Architecture (4 prompts)
A Scandinavian bedroom interior. Platform bed, white linen, single pendant lamp above. Floor-to-ceiling window, morning light. Minimal — nothing on surfaces.
A Japanese tea house interior. Tatami mats, a low wooden table, a ceramic teapot and two cups. Natural light through shoji screens. No people.
A rooftop bar in a Mediterranean city at sunset. White-painted furniture, terracotta tiles, potted bougainvillea, city rooftop view. Golden light.
A modern home office. L-shaped standing desk, dual monitors, bookshelves behind. Cable-managed, clean. Afternoon natural light from the left.
Text-in-Image Content (5 prompts)
A vintage store front sign reading "SECOND HAND" in hand-painted serif lettering. Weathered wood. Slightly faded. Natural outdoor lighting.
A minimalist poster with large bold sans-serif text reading "DO THE WORK" centered on a cream background. No decoration. Print-ready.
A coffee shop blackboard menu listing: "ESPRESSO — $3 | FLAT WHITE — $5 | COLD BREW — $4". White chalk lettering, clean and legible. Wooden frame.
A book cover design: a dark forest scene at dusk, with the title "The Last Observer" in large white type at the top, author "A.R. Sinclair" below. Moody and literary.
A warning label on a yellow background reading "CAUTION: HIGH VOLTAGE" in standard black industrial type. Sticker format. Clean edges.
Photorealistic Scenes (5 prompts)
A fog-covered mountain road at dawn. Pines on both sides. No cars. Wet pavement reflects the pale sky. Cinematic wide shot.
A crowded fish market in Tokyo. Vendors, ice-covered stalls, bright fluorescent lights. Morning rush. Slightly elevated camera angle.
A corner of a botanist's greenhouse. Humidity visible in the air. Rows of orchids in terracotta pots. Green light through glass panels.
A late-night diner from outside. Warm fluorescent interior light. One customer at the counter. Wet street, reflections. Noir aesthetic.
A watchmaker's workbench. Dozens of watch parts, magnifying lenses, small tools. One half-assembled watch in the center. Bright task lighting.
Creative and Conceptual (5 prompts)
A chess board mid-game, one king standing, all other pieces knocked over around it. Dramatic top-down lighting. Black marble board.
A single candle burning in a completely dark room. The flame is the only light source. Wax has melted and dripped.
A sand hourglass, half-empty, on a stone surface. The falling sand is lit from behind. Time metaphor. No additional props.
A map of a fictional island, hand-drawn, aged parchment. Detailed coastlines, mountains indicated in pencil, a red "X" marking one location. No text.
A folded origami crane on a wooden surface. Side lighting creates a long shadow. Only the crane and shadow — nothing else in frame.
Prompt Tips for GPT Image 2
For product photography: Always specify the surface, background, and lighting style. GPT Image 2 handles material textures (marble, linen, wood, metal) reliably when named.
For text-in-image: Put the exact text in quotation marks within the prompt. Specify the font style (serif, sans-serif, hand-lettered, chalk) and the background it sits on.
For scenes: Work from foreground to background. Specify what's in the foreground, what's at mid-ground, and what's in the background. GPT Image 2 populates all three.
For realism: Add one camera/lighting directive per prompt — "natural side light", "overhead studio lighting", "golden hour", "fluorescent indoor lighting" all translate predictably.
Using These Prompts on aicut
aicut gives you GPT Image 2 alongside all other image models from the same interface.
For any of the above prompts, try comparing GPT Image 2 against nano-banana-pro on the same input — the difference in approach is immediately visible. GPT Image 2 produces higher instruction fidelity; nano-banana-pro often produces richer aesthetic quality on the same scene description.
Key Takeaways
- GPT Image 2 handles complex, multi-part prompts better than most image models — write full sentences, not keyword lists.
- Text-in-image content is GPT Image 2's clearest advantage over diffusion-based models.
- Product photography and photorealistic scenes are consistent strengths.
- All 25 prompts are available to try at aicut — no OpenAI subscription required.
Generate these prompts with ChatGPT Image 2 at aicut — compare outputs against nano-banana-pro on the same prompts.