ChatGPT Photo Prompts to Try With GPT Image 2

Below is a library of chatgpt photo prompts written the way OpenAI's image model likes them: full-sentence photo briefs, one clear subject, one clear light source, and explicit "keep this, avoid that" rules. Every prompt is copy-paste ready, tuned for GPT Image 2 on Omnely, and paired with a settings line so you know what aspect ratio and resolution to pick. You also get a prompt formula, edit templates for photos you upload, and quick fixes for the yellow-tint, waxy-skin, and identity-drift issues people run into most often. The example images in this post were generated on Omnely with GPT Image 2.

How ChatGPT photo prompts work with GPT Image 2

ChatGPT's image generation runs on OpenAI's GPT Image model family. Omnely gives you GPT Image 2 from that family, so the plain-language photo-brief style that works well in ChatGPT carries over cleanly here. That means the same phrasing patterns — describe the shot, the subject, the light, and the finish — produce reliable results in both places, though the outputs are not guaranteed to be identical and ChatGPT's own image model has changed over time.

A few things to internalize before you paste anything:

• Write like a photographer briefing an assistant, not like a stock-tag search. Skip "8k masterpiece, ultra realistic, trending on artstation." Say "85mm lens look at f/2.8, soft overcast daylight from the left, visible skin pores, no retouching."
• One subject, one light, one job per prompt. If you try to describe three products, two people, and a mood board in one paragraph, GPT Image 2 will average it into mush.
• Edits on Omnely are separate runs. There is no in-chat "change the shirt" brush. To edit a photo, you upload it as a reference in Image to Image and describe the change plus everything that must stay the same. Full workflow is in how to use GPT Image 2.
• Resolution is the quality lever. GPT Image 2 on Omnely has no low/medium/high setting — you pick 1K, 2K, or 4K, and aspect ratio doesn't change the credit rate.

The photo prompt formula

Every prompt in this library follows the same nine-part shape. Use it as a checklist when you write your own.

Part	What it does	Example phrase
Shot type	Frames the image	"Environmental portrait", "overhead flat lay", "3/4 hero shot"
Subject	Who or what, with age/material	"A woman in her seventies with silver hair"
Action / pose	Adds life	"Looking into the camera with a calm half-smile"
Setting	Where	"Beside a potter's wheel, shelves of unglazed bowls behind her"
Light	The single most important line	"Soft overcast daylight from a large window on the left"
Lens / camera feel	Depth and compression	"85mm lens look at f/2.8, shallow depth of field"
Color & texture	Realism cues	"Neutral white balance, visible pores, matte finish"
Keep / avoid	Guardrails	"No retouching, no beauty filter, no text"
Use & aspect ratio	Ties it to output	"4:5 for a magazine feature"

Two templates you can reuse. The first is for a new photo; the second is for editing an image you upload as a reference.

[Shot type] of [subject with 2-3 concrete details], [pose or action], in [setting]. Light: [direction + quality + color temperature]. Camera: [focal length look + aperture feel + angle]. Color and texture: [white balance + skin/material detail]. Do not include [things to avoid]. Intended use: [use], aspect ratio [ratio].

Using the uploaded reference photo, change only [one specific thing]. Keep [subject identity, pose, framing, lighting direction, background, colors] exactly as they are. Do not add or remove anything else. Match the original light and white balance.

Pick aspect ratio and resolution in Omnely's settings panel — the prompt should only mention intended use so the model biases the composition correctly.

Portrait and headshot prompts

Skin is where GPT Image 2 either wins or looks like plastic. Always call out pores, fine lines, and neutral white balance, and always say no retouching.

Photorealistic environmental portrait of a woman in her seventies who runs a small pottery studio, silver hair loosely tied back, clay-dusted canvas apron over a faded denim shirt, standing beside a potter's wheel and looking into the camera with a calm half-smile. Soft overcast daylight from a large window on the left, neutral white balance, 85mm lens look at f/2.8, shelves of unglazed bowls softly out of focus behind her. Real skin texture with visible pores, fine wrinkles and age spots, no retouching, no beauty filter, no text.

Generated with GPT Image 2 on Omnely from the prompt above (4:5, 1K, 10 credits).

Settings: 4:5 at 1K (10 credits), or run 2:3 at 2K (15 credits) if you want a magazine crop.

LinkedIn-style corporate headshot generated from the uploaded selfie. Same person, same face, same age — keep identity exactly. Reframe to chest-up, straight to camera, subtle smile with closed lips, wearing a charcoal wool blazer over a white cotton shirt. Soft two-light studio setup: large softbox front-left at 5600K, gentle fill on the right, plain warm-grey seamless backdrop. 85mm lens look, f/4, neutral white balance, visible pores and natural skin texture, no beauty filter, no logos, no text.

Settings: 4:5 at 1K (10 credits). Upload the selfie as one reference image; add a second reference if you want a specific outfit.

Candid street-style portrait of a 28-year-old man walking past a rain-soaked shop window in Copenhagen, dark green overshirt, cropped trousers, mid-stride, turning his head slightly toward the camera. Overcast diffused daylight, neutral cool white balance, 35mm lens look, deep depth of field, faint reflections on the wet pavement. Documentary feel, real skin texture with stubble and pore detail, no retouching, no text.

Settings: 3:2 at 2K (15 credits) for editorial use.

Outdoor family portrait of two parents in their late thirties and their eight-year-old daughter sitting on a moss-covered log in a quiet birch forest, all three laughing at something off-frame, hands touching. Warm late-afternoon backlight filtering through leaves, soft fill on faces, 50mm lens look at f/2.5, natural depth. True skin tones for all three, visible pores and freckles, matte finish, no beauty filter, no lens flare, no text.

Settings: 3:2 at 4K (20 credits) if it is going on a printed card.

Product photography prompts

For products, protect the details GPT Image 2 tends to invent: shape, label wording, color, and logo placement. If accuracy of a real product matters, upload a reference photo (you can add up to 16 reference images).

Catalog product photograph of a stainless-steel refillable fountain pen with a brushed silver barrel and a polished chrome nib, cap posted, laid flat and centered on a pure white seamless backdrop. Soft even top light, no harsh shadows, faint contact shadow directly under the pen, neutral white balance, macro lens look at f/8 so the entire pen is sharp. E-commerce style, no props, no text, no reflections beyond a subtle highlight along the barrel.

Settings: 1:1 at 2K (15 credits) — 1:1 is capped at 2K on GPT Image 2.

Lifestyle-in-use shot of a compact matte-navy foldable travel umbrella held open by a woman's hand in the frame, held up against a light drizzle over a cobblestone street. Cool overcast daylight, tiny water droplets visible on the fabric, shallow depth of field with the background bokeh of blurred red bicycles and a bakery sign. 35mm lens look, natural color, no logo unless it is on the reference, no text overlay.

Settings: 3:2 at 2K (15 credits). Upload the real umbrella as a reference if you need brand color accuracy.

Dark moody hero shot of a small ceramic incense holder in matte terracotta, a single thin stick of incense with a curl of pale grey smoke rising, placed on a dark walnut side table. One soft rim light from the back-right, deep shadows falling to near-black on the left, no fill light, 100mm macro lens look at f/4. Rich color, visible ceramic texture, smoke rendered as thin natural wisps not clouds, no text.

Settings: 4:5 at 1K (10 credits) — 4:5 is 1K only.

Overhead flat lay of stationery arranged on a cream linen backdrop: a hardcover notebook with an olive-green fabric cover in the center, a brass fountain pen at a 20-degree angle to its right, three loose kraft envelopes fanned above, a small ceramic dish with two paperclips on the lower left. Soft even daylight from the top of the frame, faint natural shadows, neutral white balance, 50mm lens look. Photorealistic textures, no text on the notebook cover, no watermarks.

Settings: 1:1 at 2K (15 credits).

Food and drink photography prompts

Food goes wrong when GPT Image 2 makes everything glossy and orange. Ask for true colors, clean neutral whites, and diffuse daylight.

Overhead food photograph of Moroccan chicken tagine with preserved lemon and green olives in a glazed terracotta dish, a torn round of crusty bread and a small bowl of chopped cilantro beside it, on a weathered teal wooden table. Diffuse daylight from the top of the frame, soft natural shadows, true-to-life colors with clean neutral whites, a light wisp of steam, sauce glossy but not greasy. Square framing for a menu or social post, no hands, no text, no extra cutlery.

Generated with GPT Image 2 on Omnely from the prompt above (1:1, 1K, 10 credits).

Settings: 1:1 at 1K (10 credits) for a menu tile, or bump to 2K (15 credits) for print.

45-degree restaurant plating shot of pan-seared salmon on a bed of pea puree with charred asparagus and a lemon-butter drizzle, served on a wide off-white ceramic plate at the center of a dark oak table. Soft window light from the back-left, natural highlights on the fish skin, gentle shadows, neutral white balance, 50mm lens look at f/3.5. Real portion size, visible sear marks and pea texture, no oversaturation, no herbs falling off, no text.

Settings: 3:2 at 2K (15 credits).

Editorial cocktail photograph of a smoky mezcal negroni in a rocks glass with a single large clear ice cube and a wide orange peel twist, set on a dark slate bar top, a second glass softly out of focus in the background. One warm tungsten spotlight from the upper-right, deep shadows, tiny water droplets on the glass, 85mm lens look at f/2.8. True color of the amber drink, no fake bubbles, no text, no branded bottles.

Settings: 4:5 at 1K (10 credits) for a menu portrait.

Real estate and interior prompts

Listing photos live or die on straight verticals, wide-angle look, and neutral daylight. State them every time.

Real estate listing photo of an empty, freshly painted living room in a mid-century house: white walls, light oak herringbone floor, a large black-framed window on the right looking onto a green garden, a white brick fireplace on the back wall. Shot from the doorway at chest height with a 16mm wide-angle look, vertical lines perfectly straight, bright even daylight, neutral white balance. No furniture, no people, no text.

Generated with GPT Image 2 on Omnely from the prompt above (3:2, 1K, 10 credits).

Settings: 3:2 at 1K (10 credits) for the web listing; regenerate at 4K (20 credits) for a printed brochure.

Twilight exterior real-estate photograph of a two-story modern farmhouse with black cedar cladding, white trim, a metal standing-seam roof and a wide gravel driveway. Deep-blue sky just after sunset with a warm ambient glow, every interior window softly lit warm from inside, exterior path lights on, wet driveway reflecting the porch light. 24mm wide-angle look, vertical lines straight, 3:2 framing centered on the front facade. Neutral white balance outside, warm interior light, no cars, no people, no text.

Settings: 3:2 at 4K (20 credits) — this is the money shot of a listing.

Interior detail photograph of a modern kitchen: matte deep-green shaker cabinets, brushed brass hardware, a honed white marble countertop, an under-mount stainless sink with a matte black gooseneck faucet, a small ceramic bowl of lemons on the counter. Bright even daylight from a window off-frame left, faint reflections on the brass, 35mm lens look at f/5.6, vertical lines straight. Neutral white balance, no smudges, no dishes, no text.

Settings: 4:3 at 2K (15 credits).

Photo edit and restyle prompts for your own photos

On Omnely each edit is a fresh Image to Image run: upload the photo as a reference, describe the one thing you want changed, and restate what must stay. Edits cost the same as new images — 1K 10, 2K 15, 4K 20 credits.

Background swap using the uploaded portrait as the reference. Keep the person's face, hairstyle, expression, skin tone, clothing, pose, framing and body proportions exactly as they are — identity must be locked. Replace only the background with a softly out-of-focus modern office interior, warm ambient lighting, muted olive and oak tones. Match the original light direction on the subject's face and keep the original edge lighting. Do not restyle the person, do not change age, do not add text.

Settings: 4:5 at 1K (10 credits).

Relight edit using the uploaded outdoor photo as the reference. Keep the subject, pose, clothes, background objects and framing exactly the same. Change only the lighting: convert harsh noon sun with hard black shadows into soft open shade under a bright overcast sky, evenly diffused daylight from above, no directional shadows on the face, gentle cool 6000K white balance. Preserve natural skin tone and true clothing colors, do not add haze, do not change composition, do not restyle the subject.

Settings: 3:2 at 2K (15 credits)

Outfit change edit using the uploaded full-body photo as the reference. Keep the person's face, hair, skin tone, body proportions, pose, background and lighting exactly as they are — identity locked. Replace only the clothing with a tailored beige linen suit over a plain white t-shirt and clean white leather sneakers. Match the original light direction and shadows on the new fabric. Do not change stance, do not change background, do not add accessories or text.

Settings: 2:3 at 2K (15 credits).

Old photo restoration using the uploaded damaged photograph as the reference. Repair only physical damage: remove scratches, creases, dust spots, mold and fading, and gently rebalance color if it is faded. Keep every face, expression, hairstyle, clothing detail, background object and composition exactly as it appears in the original — do not invent details that are not visible. Do not sharpen skin into plastic, do not remove wrinkles or freckles, do not add color that was not implied by the original, do not add text.

Settings: 3:2 at 2K (15 credits) — good for prints up to A4.

Film restyle edit using the uploaded digital photo as the reference. Keep subject, pose, framing, background and every object exactly the same. Apply only a Kodak Portra 400 analog color grade: warm midtones, slightly lifted shadows, natural pastel skin tones, gentle organic grain across the whole image, subtle halation around bright highlights. Neutral white balance overall, do not add light leaks or dust, do not add borders or text.

Settings: 3:2 at 2K (15 credits).

Virtually stage this room for a listing photo. Add a low cream bouclé sofa facing the fireplace, a round walnut coffee table, a jute rug, one tall fiddle-leaf fig by the window and two framed abstract prints above the mantel. Keep the walls, floor, window, fireplace, ceiling height, daylight and camera angle exactly as they are. Do not change the room's shape and do not add windows or doors.

GPT Image 2 Image to Image edit on Omnely (3:2, 1K, 10 credits), using the empty-room image above as the reference. Walls, floor, window and fireplace stayed in place; the model also added two accent chairs that were not in the prompt.

Settings: 3:2 at 1K (10 credits) — upload the empty-room image above as your reference.

Travel, lifestyle and pet photo prompts

Travel photograph of a lone hiker in a red shell jacket standing on a rocky ridge in the Scottish Highlands, back to camera, looking out at layered green and grey mountains under a broken sky. Cool overcast daylight with a shaft of sunlight breaking through onto the distant valley, 24mm wide-angle look, sharp foreground rocks with texture, subtle atmospheric haze in the distance. Neutral cool white balance, no lens flare, no text, no watermark.

Settings: 21:9 at 4K (20 credits) for a hero banner (21:9 supports up to 4K).

Everyday lifestyle photograph of a man in his early forties sitting on the floor of a sunlit apartment reading a book with a large orange tabby cat curled in his lap, morning coffee on a low wooden table beside him, houseplants on the windowsill. Soft warm window light from frame-left, 35mm lens look at f/2.8, gentle depth, natural skin texture with visible stubble, cat fur clearly detailed. Neutral warm white balance, quiet mood, no text.

Settings: 4:3 at 2K (15 credits).

Studio pet portrait of a border collie with black and white markings and one blue eye, sitting alert, head slightly tilted, looking just past the camera. Plain deep teal seamless backdrop, single large softbox front-left, subtle rim light on the right, 85mm lens look at f/4. Sharp eyes with catchlight, individual fur strands visible, natural black-and-white tones, faint contact shadow beneath, no props, no collar text, no watermark.

Settings: 1:1 at 2K (15 credits).

Candid birthday party photograph of a small group of friends around a wooden dining table indoors, someone leaning forward to blow out candles on a two-tier chocolate cake, the others laughing and clapping. Warm mixed light from string lights above and one soft window off-frame, slight motion blur on hands, 35mm lens look at f/2.5, shallow depth. Natural skin tones across different ethnicities, real party clutter on the table, no text on cake, no logos.

Settings: 3:2 at 2K (15 credits).

Fixes for common ChatGPT photo problems

Problem	One-line fix to add to your prompt
Yellow / warm tint on everything	Neutral white balance at 5600K daylight, clean neutral whites, no warm color cast.
Plastic or waxy skin	Real skin texture with visible pores and fine lines, no retouching, no beauty filter, matte finish.
Face changes on an edit (identity drift)	Keep the person's face, hairstyle, age and skin tone exactly as in the reference — identity locked. Change only [one thing].
Wrong or warped text on signs / labels	Spell exactly "[YOUR TEXT]" in [font style], centered, high contrast. No other text, no misspellings, no extra words.
Edits change too much of the image	Change only [one thing]. Keep everything else — pose, framing, background, colors, light direction — exactly as in the reference.
Cluttered or over-styled scene	Minimal composition, one subject, one prop, plenty of negative space, no extra objects.
Warped hands or extra fingers	Hands relaxed and clearly visible, five fingers each, natural anatomy, no distortion.

Add these as their own sentences at the end of your prompt — they act as guardrails, not decoration.

GPT Image 2 settings on Omnely

Pricing on GPT Image 2 is set purely by output resolution, and it's the same whether you're generating from text or editing an uploaded photo:

Resolution	Credits	USD
1K	10	$0.10
2K	15	$0.15
4K	20	$0.20

1 credit = $0.01, and your credits are shared between the web workspace and the API — same balance, same rate.

Aspect ratio cheat sheet by use:

• Headshot / portrait: 4:5 (1K only)
• Feed post: 4:5 or 1:1 (1:1 goes up to 2K)
• Story / Reel: 9:16 (up to 4K)
• Real estate listing: 3:2 (up to 4K)
• Kitchen / interior detail: 4:3 (up to 4K)
• Web banner: 16:9 (up to 4K)
• Cinematic hero: 21:9 (up to 4K)

Clamps to remember: auto is 1K only. 1:1 caps at 2K. 4:5, 5:4, 3:1, 1:3 and 9:21 are 1K only. Everything else — 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9, 2:1, 1:2 — supports 4K. Aspect ratio does not change the credit rate; resolution does.

Workflow rule of thumb: draft at 1K (10 credits), regenerate the winner at 2K or 4K. Ten drafts + one 4K final = 120 credits = $1.20. Full step-by-step is in the GPT Image 2 workflow guide. You can attach up to 16 reference images per run when you need identity, product accuracy, or scene matching.

If you're comparing generators before you commit, our best AI image generators for 2026 roundup covers the field, and if you're working with Google's model instead, the Nano Banana Pro prompt library uses the same brief style tuned for that model.

FAQ

Do these prompts work in ChatGPT too?

Yes, the same brief-style phrasing carries over well because both tools respond to plain-language photo direction. Outputs won't be pixel-identical though, since the model version and settings differ between the two workspaces.

Is GPT Image 2 the model ChatGPT uses?

ChatGPT's image generation runs on OpenAI's GPT Image model family, and that family has changed over time — so it isn't accurate to say ChatGPT currently uses GPT Image 2. Omnely gives you GPT Image 2 from that family, which is why the prompting style feels familiar.

How many credits does a photo cost on Omnely?

GPT Image 2 costs 10 credits at 1K, 15 at 2K, and 20 at 4K ($0.10 / $0.15 / $0.20 at 1 credit = $0.01). Edits on uploaded references cost the same as new generations, and your credit balance is shared between the web workspace and the API.

Start writing your own photo prompts

Open GPT Image 2 on Omnely, paste any prompt from this library, and iterate at 1K before locking a final at 2K or 4K. When you're ready to build a full project — portraits, product shots, listings — the Image workspace keeps every run in one place.