← Back to all free guides

ChatGPT Images 2.5
The Prompt Playbook

The four looks everyone is copying, the real prompts behind them, and the craft words that decide whether your image looks expensive or looks like AI.

Where The Prompts Live YouMind's GPT Image 2.5 library holds 593 prompts pulled from X, each one credited to whoever wrote it, with the resulting image next to the words. Free to browse, no account needed. Sort by views and you can see exactly what is working.

Saving a prompt gets you one image. Learning why it works gets you every image after that. So this is the craft layer underneath the library, taken from the prompts with real numbers behind them and from OpenAI's own prompting documentation.

If you want the five business jobs to point it at instead, from sketch-to-mockup through merch and flyers, I covered those in ChatGPT Images 2.5 for your business. This guide is how to make any of them look good.

Two facts that change how you work. There is no seed, so the same prompt twice gives you two different images, and consistency has to come from reference photos instead. And 2.5's real leap is editing, not generating. It holds an image steady while you change one thing, which means you stop re-rolling and start directing.

The Twelve Lines

Every prompt worth copying has the same bones. OpenAI's guidance is to write it as labeled sections rather than one long paragraph, and the viral ones all do exactly that. Type, format, subject, identity lock, materials, pose, environment, lighting, camera, grade, text, constraints.

Four of those do most of the work:

The full fill-in-the-blank version is in the pack at the bottom.

The Words That Change The Picture

Camera language is the fastest lever you have. One caveat from OpenAI first: these are cues for appearance, not physics. "85mm f/1.8" is a style word, not a setting. It biases the look, it does not simulate a lens.

Delete these five words

8K, ultra-realistic, masterpiece, hyper-detailed, award-winning. They push the image toward waxy skin and the glossy AI sheen. Replace them with "realistic skin texture, visible pores, subtle natural imperfections." Golden hour has gone the same way, so reach for flat overcast light or window light instead.

How To Fake A Real Photo

The camera-roll look is the one people cannot believe, and it works by inversion. You stop describing beauty and start describing a camera failing. Every lever is a flaw.

  1. Bad framing. "Place the person slightly to the right." "Create mysterious whitespace above the head." Tilted horizons and dead space.
  2. A missed focus. "AF pulled to the background" is the most convincing tell there is. The subject goes slightly soft and the wall behind them is sharp.
  3. Exposure failure. "Incomplete HDR correction, partial highlights blown out, deep shadows crushed." Nothing tone-mapped into looking pretty.
  4. Motion. A blurred passing hand, a turning head, a train moving through frame.
  5. Dirty glass. "Veil-like flare." "Obvious fingerprints, smudges and scratches with a direct light source creating glare."
  6. Humble gear. "Shot on iPhone." "RAW quality, unprocessed, unedited." Naming a pro rig makes it look like stock photography, so name a phone instead. Dating it works too: "a photo from 2003, shot with a digital camera" pulls in a whole era of color science.
  7. Ordinary faces. In-between expressions, mid-sentence, looking away. Not aspirational.

The line I would not cross

These are brilliant for content and concepts. They are designed to look like documentary evidence, so use your own face, or someone who gave you written permission, or an invented person. Never use a real person who has not agreed to it, and never dress something up as though it actually happened. OpenAI's own policy bans generating a real person's likeness without their consent.

Edit, Do Not Regenerate

When an image is 90% right, almost everyone throws it away and rolls again. That is the expensive mistake, because there is no seed and you will never get back to the version you liked.

Instead, use the formula. Tell it to change only X, then hand it a long explicit list of everything to preserve, and finish with make no other changes. The longer and more specific that preserve list, the less the rest of the image drifts. One change per turn, never five.

Two honest limits. Repeated edits can still move details you asked to keep, so check the whole image every time rather than just the bit you changed. And if something has to be pixel-perfect, like your actual logo, paste it over the top in Canva afterwards rather than trusting the prompt. On text generally, 2.5 handles short bold copy well and still fumbles small dense type, prices and dates, so typeset the final numbers yourself.

The Prompt Pack

The skeleton, one full working prompt for each of the four looks, the identity-lock lines that keep a face or a product the same across images, and the edit formulas. Swap anything in brackets for your own details.

ChatGPT Images 2.5 Prompt Pack
THE SKELETON ================================================================ Fill these twelve lines in this order. Delete any you do not need. Labeled sections beat one long paragraph, and this is the order OpenAI's own prompting guide recommends. 1. TYPE: [editorial product photograph / studio fashion portrait / candid snapshot / art poster] 2. FORMAT: [4:5 vertical / 9:16 / 1:1 / 16:9] 3. SUBJECT: [who or what, with age, build, expression, texture] 4. IDENTITY LOCK: Maintain the exact facial features and identity of the person in the uploaded reference photo. 5. WARDROBE / MATERIALS: [name the fabrics and finishes, not the vibe: chocolate brown linen, frosted glass, brushed brass] 6. POSE / ACTION: [what the body is doing, loosely and specifically] 7. ENVIRONMENT: [where, and what must NOT be in shot] 8. LIGHTING: [direction + quality + modifier: soft window light from the left, casting irregular leaf-like shadows] 9. CAMERA: [85mm, f/1.8, shallow depth of field, slightly low angle] 10. GRADE: [name four to six actual colors, not a mood word] 11. TEXT: Render exactly once, word for word: "[YOUR COPY]", placed in the [upper third]. No other text. 12. CONSTRAINTS: No watermark, no extra logos, no people, no clutter. LOOK 1 / THE LOOKBOOK PORTRAIT ================================================================ Maintain the exact facial features and identity of the person in the uploaded reference photo. A three-quarter fashion lookbook portrait of the subject leaning casually against a light beige stucco wall. Styled in a buttoned rich chocolate brown linen shirt, pleated beige linen trousers, small gold hoop earrings, a thin gold necklace, and retro tortoise-shell sunglasses. Tousled textured hair blowing gently. High-contrast golden-hour sunbeams cutting across the background, crisp details, natural linen fabric weave, 85mm f/1.8 LOOK 2 / THE PRODUCT EDITORIAL ================================================================ Swap the bracketed parts for your product. This skeleton works for anything you sell. Create a premium editorial product photograph in [a sunlit neighborhood coffee bar]. [A clear amber glass bottle] labeled "[YOUR BRAND / YOUR PRODUCT]" stands on [a light walnut table]. [A woman in a cream linen shirt] places it with one hand. Morning window light comes from the left, soft realistic shadows, shallow depth of field, warm neutral color grade, [4:5] composition. Keep the label fully readable and centered. Leave copy space on the right. No extra logos, no watermark. Preserve the recognizable identity of the product. Do not invent unrelated packaging, alter its core design, add unnecessary text, or introduce extra branding. LOOK 3 / THE SURREAL CONCEPT ================================================================ The rule for this category: the surreal idea gets ONE sentence. Everything else stays literal and physical. Create a breathtaking, photorealistic [luxury food-art] image of [YOUR SUBJECT] transformed into [YOUR SURREAL OBJECT]. [The object] lies diagonally on an elegant dark stone surface, photographed from a slightly elevated cinematic angle. It is dramatically broken open in the centre, and from inside it an incredibly detailed miniature 3D world emerges. Include: [landmark] as the central focal point, [landscape] surrounding it, tiny recognizable architecture, miniature streets, [small objects] as decorative elements. Make the materials extremely realistic: glossy surface, crisp snapped edges, tiny crumbs, realistic highlights and shadows. Create a magical transition where the pieces gradually become mountains and buildings. Cinematic lighting, ultra-detailed miniature world, macro photography, shallow depth of field, atmospheric perspective, rich textures, surreal but believable. Composition: vertical 4:5, centered hero object, clean luxurious background, no people, no subtitles, no watermark, no text. LOOK 4 / THE CAMERA ROLL ================================================================ Every line here is a camera FAILURE. That is the whole technique. A daily failed snapshot casually taken by a friend. 9:16 vertical aspect ratio. The subject is [a woman in her 20s with dark hair]. [A sunny city street in autumn.] Place the person slightly to the right. Create mysterious whitespace above the head. Auto-exposure failure, incomplete HDR correction, unevenly exposed snapshot, slight focus shift, AF pulled to the background, partial highlights blown out, deep shadows crushed, extreme contrast, strong direct sunlight, veil-like flare. Shorter variant that also works: Vertical 3x3 failed photos x [subject]. Whole-day variant, needs a reference photo: Take a day's worth of camera roll from the person in the reference image and turn it into a single image captured directly from the iPhone Photos app screen. Make it look like a real camera roll of casually shot everyday life. KEEPING A FACE OR PRODUCT THE SAME ================================================================ There is no seed. The same prompt twice gives you two different images. Consistency comes from references and repetition, never from re-rolling. Strong identity lock, paste this in every time: STRICT IDENTITY LOCK: Preserve her recognizable facial identity exactly from the face reference. Keep the same facial structure, face shape, eyes, eyebrows, nose, lips, jawline, and skin tone. Multiple references, give each one a job: Image 1 is my product. Image 2 is the lighting and mood I want. Image 3 is the background. Combine them so [explain how]. Continuing a character: Continue using the same character. Same outfit, same facial features, proportions, and color palette. New scene: [describe it]. EDITING WITHOUT WRECKING IT ================================================================ The universal formula: Change only [X]. Preserve [long explicit list]. Make no other changes. Removing something: Remove [the object] from [location]. Do not change anything else. Changing clothes on a person: Edit the image to dress the person using the provided clothing image. Do not change her face, facial features, skin tone, body shape, pose, or identity in any way. Replace only the clothing, fitting the garment naturally to her existing pose and body geometry. Changing label copy with the comment pin: Only change the front label in the marked area. Replace "[OLD TEXT]" with "[NEW TEXT]" in the same style. Preserve the face, hand position, bottle shape, glass reflections, table, background, crop, camera angle, light direction, shadow softness, and all unmarked details. Make no other changes. Three rules: one change per edit, always include a preserve list, and if it starts drifting go back to your last approved image instead of trying to repair a repair.

How to set it up. Fastest way. Copy the pack, pick the look you want, fill in the brackets, paste it into ChatGPT with your reference photo attached. To keep it, download the file and save it as a Claude Skill or a ChatGPT Project so you can say "write me a Look 2 prompt for this product" and get the finished prompt back.