Repository Public
1# Image prompting23> For the complete documentation index, see [llms.txt](/llms.txt). Markdown versions of documentation pages are available by appending `.md` to the page URL.45<header className="not-prose mb-8">6 <h27 id="gpt-image-2.5-guide"8 className="m-0 text-3xl font-semibold text-default"9 >10 {"GPT Image 2.5 prompting guide"}11 </h2>12 1314 Choose a model, write effective prompts, and preserve details across15 edits.16 1718 </header>19 2021## Overview2223Start with the image you need, then describe the subject, composition, style, and constraints. For edits, identify what should change and what must stay the same. Refine one thing at a time and inspect the result.2425GPT Image 2.5 includes two model choices. GPT Image 2.5 Flare is the small model, optimized for speed, with image quality comparable to GPT Image 2. GPT Image 2.5 Sunburst is the base model, optimized for quality, with higher image quality than GPT Image 2. Both models offer improvements in precise editing and subject preservation.2627For API setup and request examples, see the [image generation guide](https://developers.openai.com/api/docs/guides/image-generation).2829## Choose a model3031For a new workflow, start with GPT Image 2.5 Flare when speed is the priority, or GPT Image 2.5 Sunburst when demanding quality requirements are the priority. Once the output meets your requirements, look for opportunities to reduce latency.3233For migrating from a current image model, use your current image quality as the starting point. Both models support image generation, editing, and transparent backgrounds.3435| Your current workflow | Start by testing |36| ----------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------- |37| An existing, validated GPT Image 2 workflow already meets your quality requirements | GPT Image 2.5 Flare. Check whether you can retain acceptable quality while reducing latency. |38| A complex use case where GPT Image 2 does not meet your quality requirements | GPT Image 2.5 Sunburst. First establish that it delivers the quality you need. |3940If GPT Image 2.5 Sunburst meets your quality requirements, then test GPT Image 2.5 Flare with the same prompts and inputs. Switch to GPT Image 2.5 Flare if it also meets those requirements and improves latency. Keep GPT Image 2.5 Sunburst when its quality advantage is necessary for your workflow.4142Measure response time and quality on your own workload. Results depend on your prompts, reference images, output dimensions, and quality settings; a speed improvement on one workload doesn't establish a fixed improvement on another.4344## Model parameters4546Set API parameters separately from the prompt.4748| Parameter | GPT Image 2.5 settings |49| ------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |50| `model` | `gpt-image-2.5-flare` (small model) or `gpt-image-2.5-sunburst` (base model) |51| `quality` | `auto` (default), `low`, `medium`, `high`, `xhigh`, or `max` |52| `size` | `auto` or a custom resolution. Common sizes: `1024x1024` (square), `1536x1024` (landscape), `1024x1536` (portrait), `2048x2048` (2K square), `2048x1152` (2K landscape), `3840x2160` (4K landscape), and `2160x3840` (4K portrait). |53| `background` | `auto`, `opaque`, or `transparent` |5455For a custom resolution, use `WIDTHxHEIGHT` and follow these constraints:5657- Each edge must be no more than 3,840 pixels.58- Both edges must be multiples of 16 pixels.59- The ratio of the longer edge to the shorter edge must not exceed 3:1.60- The total pixel count must be between 655,360 and 8,294,400.6162Outputs with more than 3,686,400 total pixels (`2560x1440`) are experimental.6364Choose the model using the workflow above before tuning `quality`. For the first comparison, keep an explicitly selected quality setting unchanged when both models support it, along with the prompt, reference images, and output dimensions. The same quality label does not imply the same image quality or response time across models.6566If the output falls short, test a higher quality setting. Once it meets your requirements, test lower settings to see whether they preserve acceptable quality while reducing latency. Use `xhigh` or `max` only when they improve an unmet quality requirement within your latency budget. A higher setting doesn't guarantee a better result for every prompt.6768For transparent assets, explicitly request `background="transparent"` and use PNG or WebP. Check the decoded image's alpha channel, including hair, glass, shadows, and object edges. Use `output_compression` only for JPEG or WebP output, not PNG.6970## Migrate an existing workflow71721. **Save a baseline.** Collect representative production prompts and reference images, including difficult edits, exact text, faces, product geometry, and transparent assets. Record the current model, request settings, and results.732. **Choose the first candidate.** If GPT Image 2 already meets your quality requirements, start with GPT Image 2.5 Flare and test for a latency improvement. If GPT Image 2 falls short on a complex use case, start with GPT Image 2.5 Sunburst and first establish that it meets your quality requirements. Keep the prompt, references, dimensions, and output format unchanged for the first comparison.743. **Check the complete result.** Compare instruction following, identity and product preservation, text accuracy, unwanted changes, and transparency. Repeat requests to measure consistency. For editing workflows, test the complete sequence of edits as well as individual steps.754. **Test for a latency gain after quality passes.** If you started with GPT Image 2.5 Sunburst and it meets your quality requirements, evaluate GPT Image 2.5 Flare against the same requirements. Switch only if the quality remains acceptable and latency improves; otherwise, keep GPT Image 2.5 Sunburst.765. **Tune one setting at a time.** Compare quality levels before rewriting the prompt. Measure typical and slow responses, failures, retries, and cost per accepted image. Confirm current pricing rather than assuming the faster model costs less.776. **Roll out by workflow.** Once the released model passes your acceptance criteria, move a small share of traffic, monitor the same measures, and expand gradually. Keep the previous model available for rollback while it remains supported.7879When migrating from GPT Image 1 or 1.5, use the reference tabs to check parameter differences and shutdown dates. Test the candidate model's supported request settings rather than copying older settings unchanged. For GPT Image 2, keep your existing resolution and transparency requirements in the comparison.8081Repeated edits can still change details you intended to preserve. Restate those constraints and inspect each result. If a region must remain pixel-identical, composite the approved edit into the original image instead of relying on prompting alone.8283## Prompting fundamentals84851. **Define the result.** Name the subject and intended use, such as a product photograph, advertisement, or diagram. Specify the composition, aspect ratio, and important placement constraints. For complex requests, organize the prompt as scene, subject, details, and constraints, using labeled sections.862. **Choose a maintainable format.** Short prompts, descriptive paragraphs, JSON-like structures, instructions, and tags can all express the same intent. Choose the format that makes the requirements easiest to read and update rather than relying on special syntax.873. **Describe visible details.** Name materials, lighting, colors, and the visual medium. Request “photorealistic” or “real photograph” explicitly when that is the goal, and describe framing and texture. Treat camera specifications as cues for appearance, not a guarantee of exact physical simulation. For wide, cinematic, low-light, rainy, or neon scenes, specify scale, atmosphere, and color instead of relying on mood words alone.884. **Specify people and actions.** Describe body framing, relative scale, gaze, and interaction with objects. Instructions such as “full body visible, feet included,” “looking down at the open book,” or “hands naturally gripping the handlebars” make the intended pose and action clearer.895. **Specify exact text.** Put required wording in quotes and describe its position and typography. Spell unusual words or brand names letter by letter when needed. Ask for no extra text, then check spelling and legibility in the output. Compare medium or high quality for small text, dense information, or multiple fonts.906. **Separate changes from constraints.** For edits, say “change only X” and list the details to preserve, such as identity, geometry, layout, lighting, or labels. State exclusions such as unwanted text, logos, or watermarks. For precise local edits, also identify saturation, contrast, arrows, camera angle, and surrounding objects that must remain unchanged.917. **Assign roles to references.** Identify each input by number and purpose: subject, style, clothing, or background. Explain how the inputs should combine and which elements should move where.928. **Iterate deliberately.** Pass the previous output as the next edit input, request one change, and repeat the details to preserve. References such as “same style as before” can carry context, but restate critical constraints if the result drifts. Compare results before adding more instructions.9394The examples below each demonstrate a different technique. Keep their prompts as starting points and adapt them to your own images and requirements.9596## Generate images9798### Control style and lighting99100Describe a photograph through its subject, framing, light, and texture. This example specifies a candid composition and explicitly excludes heavy retouching.101102Generation settings: `size="1024x1536"`, `quality="medium"`.103104```text105Create a photorealistic candid photograph of an elderly sailor standing on a small fishing boat.106He has weathered skin with visible wrinkles, pores, and sun texture, and a few faded traditional sailor tattoos on his arms.107He is calmly adjusting a net while his dog sits nearby on the deck. Shot like a 35mm film photograph, medium close-up at eye level, using a 50mm lens.108Soft coastal daylight, shallow depth of field, subtle film grain, natural color balance.109The image should feel honest and unposed, with real skin texture, worn materials, and everyday detail. No glamorization, no heavy retouching.110```111112Example outputs:113114115116 <figure className="m-0 min-w-0">117 <figcaption className="mb-2 min-h-10 text-sm font-semibold">118 GPT Image 2.5 Flare119 </figcaption>120 121122123124125 </figure>126 <figure className="m-0 min-w-0">127 <figcaption className="mb-2 min-h-10 text-sm font-semibold">128 GPT Image 2.5 Sunburst129 </figcaption>130 131132133134135 </figure>136137138139### Explain a process visually140141Name the process, audience, and information the image should communicate. For diagrams and information graphics, verify labels and factual relationships as well as appearance.142143Generation settings: `size="1024x1536"`, `quality="medium"`.144145```text146Create a detailed Infographic of the functioning and flow of an automatic coffee machine like a Jura.147From bean basket, to grinding, to scale, water tank, boiler, etc.148I'd like to understand technically and visually the flow.149```150151Example outputs:152153154155 <figure className="m-0 min-w-0">156 <figcaption className="mb-2 min-h-10 text-sm font-semibold">157 GPT Image 2.5 Flare158 </figcaption>159 160161162163164 </figure>165 <figure className="m-0 min-w-0">166 <figcaption className="mb-2 min-h-10 text-sm font-semibold">167 GPT Image 2.5 Sunburst168 </figcaption>169 170171172173174 </figure>175176177178### Render exact text179180Quote the required copy and tell the model how many times it should appear. Specify the audience and visual treatment without adding unrelated instructions.181182Generation settings: `size="1024x1536"`, `quality="medium"`.183184```text185Give me a cool in culture ad / fashion shot for a brand called Thread.186It's a hip young street brand. The ad shows a group of friends hanging out together with the tagline "Yours to Create."187Make it feel like a polished campaign image for a youth streetwear audience: stylish, contemporary, energetic, and tasteful.188Use clean composition, strong color direction, natural poses, and premium fashion photography cues.189Render the tagline exactly once, clearly and legibly, integrated into the ad layout.190No extra text, no watermarks, no unrelated logos.191```192193Example outputs:194195196197 <figure className="m-0 min-w-0">198 <figcaption className="mb-2 min-h-10 text-sm font-semibold">199 GPT Image 2.5 Flare200 </figcaption>201 202203204205206 </figure>207 <figure className="m-0 min-w-0">208 <figcaption className="mb-2 min-h-10 text-sm font-semibold">209 GPT Image 2.5 Sunburst210 </figcaption>211 212213214215216 </figure>217218219220### Design a reusable logo221222Describe the brand and the shapes that should define the mark. Specify a clear composition that remains legible at different sizes. Use `n` to request multiple variations.223224Generation settings: `size="1024x1536"`, `quality="medium"`, `background="transparent"`, `output_format="png"`, `n=1`.225226```text227Create an original, non-infringing logo for a company called Field & Flour, a local bakery.228The logo should feel warm, simple, and timeless. Use clean, vector-like shapes, a strong silhouette, and balanced negative space.229Favor simplicity over detail so it reads clearly at small and large sizes. Flat design, minimal strokes, no gradients unless essential.230Fully transparent background. Deliver a single centered logo with generous padding, clean alpha edges, and no solid backdrop, scenery, checkerboard, or watermark.231```232233Each row compares one variation from each model.234235Example outputs:236237238239 <figure className="m-0 min-w-0">240 <figcaption className="mb-2 min-h-10 text-sm font-semibold">241 GPT Image 2.5 Flare242 </figcaption>243 244245246247248 </figure>249 <figure className="m-0 min-w-0">250 <figcaption className="mb-2 min-h-10 text-sm font-semibold">251 GPT Image 2.5 Sunburst252 </figcaption>253 254255256257258 </figure>259260261262263264 <figure className="m-0 min-w-0">265 <figcaption className="mb-2 min-h-10 text-sm font-semibold">266 GPT Image 2.5 Flare267 </figcaption>268 269270271272273 </figure>274 <figure className="m-0 min-w-0">275 <figcaption className="mb-2 min-h-10 text-sm font-semibold">276 GPT Image 2.5 Sunburst277 </figcaption>278 279280281282283 </figure>284285286287288289 <figure className="m-0 min-w-0">290 <figcaption className="mb-2 min-h-10 text-sm font-semibold">291 GPT Image 2.5 Flare292 </figcaption>293 294295296297298 </figure>299 <figure className="m-0 min-w-0">300 <figcaption className="mb-2 min-h-10 text-sm font-semibold">301 GPT Image 2.5 Sunburst302 </figcaption>303 304305306307308 </figure>309310311312313314 <figure className="m-0 min-w-0">315 <figcaption className="mb-2 min-h-10 text-sm font-semibold">316 GPT Image 2.5 Flare317 </figcaption>318 319320321322323 </figure>324 <figure className="m-0 min-w-0">325 <figcaption className="mb-2 min-h-10 text-sm font-semibold">326 GPT Image 2.5 Sunburst327 </figcaption>328 329330331332333 </figure>334335336337### Use historical and real-world context338339Name the place and date to establish a historical setting. The model can infer contextual details, but inspect clothing, staging, and surroundings for historical accuracy.340341Generation settings: `size="1024x1536"`, `quality="medium"`.342343```text344Create a realistic outdoor crowd scene in Bethel, New York on August 16, 1969.345Photorealistic, period-accurate clothing, staging, and environment.346```347348Example outputs:349350351352 <figure className="m-0 min-w-0">353 <figcaption className="mb-2 min-h-10 text-sm font-semibold">354 GPT Image 2.5 Flare355 </figcaption>356 357358359360361 </figure>362 <figure className="m-0 min-w-0">363 <figcaption className="mb-2 min-h-10 text-sm font-semibold">364 GPT Image 2.5 Sunburst365 </figcaption>366 367368369370371 </figure>372373374375### Turn a story into a comic strip376377For story-to-comic generation, define the narrative as a sequence of clear visual beats, one per panel. Keep descriptions concrete and action-focused so the model can translate the story into readable, well-paced panels.378379Generation settings: `size="1024x1536"`, `quality="medium"`.380381```text382Create a short vertical comic-style reel with 4 panels.383Panel 1: The owner leaves through the front door. The pet is framed in the window behind them, small against the glass, eyes wide, paws pressed high, the house suddenly quiet.384Panel 2: The door clicks shut. Silence breaks. The pet slowly turns toward the empty house, posture shifting, eyes sharp with possibility.385Panel 3: The house transformed. The pet sprawls across the couch like it owns the place, crumbs nearby, sunlight cutting across the room like a spotlight.386Panel 4: The door opens. The pet is seated perfectly by the entrance, alert and composed, as if nothing happened.387```388389Example outputs:390391392393 <figure className="m-0 min-w-0">394 <figcaption className="mb-2 min-h-10 text-sm font-semibold">395 GPT Image 2.5 Flare396 </figcaption>397 398399400401402 </figure>403 <figure className="m-0 min-w-0">404 <figcaption className="mb-2 min-h-10 text-sm font-semibold">405 GPT Image 2.5 Sunburst406 </figcaption>407 408409410411412 </figure>413414415416### Create an interface preview417418Interface previews work best when you describe the product as if it already exists. Focus on layout, hierarchy, spacing, and real interface elements, and avoid concept art language so the result looks like a usable, shipped interface rather than a design sketch.419420Generation settings: `size="1024x1536"`, `quality="medium"`.421422```text423Create a realistic mobile app UI mockup for a local farmers market.424Show today’s market with a simple header, a short list of vendors with small photos and categories, a small “Today’s specials” section, and basic information for location and hours.425Design it to be practical, and easy to use. White background, subtle natural accent colors, clear typography, and minimal decoration.426It should look like a real, well-designed, beautiful app for a small local market.427Place the UI mockup in an iPhone frame.428```429430Example outputs:431432433434 <figure className="m-0 min-w-0">435 <figcaption className="mb-2 min-h-10 text-sm font-semibold">436 GPT Image 2.5 Flare437 </figcaption>438 439440441442443 </figure>444 <figure className="m-0 min-w-0">445 <figcaption className="mb-2 min-h-10 text-sm font-semibold">446 GPT Image 2.5 Sunburst447 </figcaption>448 449450451452453 </figure>454455456457### Create scientific and educational visuals458459Scientific and educational visuals are strong fits for biology, chemistry, classroom explanations, flat scientific icon systems, diagrams, and learning assets. Prompt them like an instructional design brief: define the audience, lesson objective, visual format, required labels, and scientific constraints. For best results, ask for a clean, flat visual system with consistent icon style, clear arrows, readable labels, and enough white space for students to scan the concept quickly.460461When accuracy matters, list the required components explicitly and say what should not be included. Use `quality="high"` for dense labels, diagrams, or assets that will be used in slides or course materials.462463Generation settings: `size="1536x1024"`, `quality="high"`.464465```text466Create a simple biology diagram titled "Cellular Respiration at a Glance" for high school students.467468Show how glucose turns into energy inside a cell. Include glycolysis, the Krebs cycle, and the electron transport chain.469Use arrows to connect the steps, and label the main molecules: glucose, pyruvate, ATP, NADH, FADH2, CO2, O2, and H2O.470Make it look like a clean classroom handout or slide, with a white background, simple icons, clear labels, and easy-to-read text.471472Avoid tiny text, extra decoration, or anything that makes the diagram hard to understand.473```474475Example outputs:476477478479 <figure className="m-0 min-w-0">480 <figcaption className="mb-2 min-h-10 text-sm font-semibold">481 GPT Image 2.5 Flare482 </figcaption>483 484485486487488 </figure>489 <figure className="m-0 min-w-0">490 <figcaption className="mb-2 min-h-10 text-sm font-semibold">491 GPT Image 2.5 Sunburst492 </figcaption>493 494495496497498 </figure>499500501502### Build slides, diagrams, and charts503504Productivity visuals work best when the prompt is written like an artifact spec rather than an illustration request. Name the exact deliverable (slide, workflow diagram, chart, page image), define the canvas and hierarchy, provide the real text or data, and describe the visual language. These prompts should include practical constraints: readable typography, polished spacing, no decorative clutter, and no generic stock-photo treatment.505506For slides, charts, and diagram-heavy assets, include the numbers and labels directly in the prompt. Use a landscape size for deck-style outputs and `quality="high"` when the image contains small text, legends, axes, or footnotes.507508The sample market figures and citations below are fictional design inputs. Replace them with verified data before using the slide.509510Generation settings: `size="1536x864"`, `quality="high"`.511512```text513Create one pitch-deck slide titled **"Market Opportunity"** that feels like a real Series A fundraising slide from a YC-backed startup.514515Use a clean white background, modern sans-serif typography like Inter, and a crisp, minimal layout. The slide should include:516517* A TAM/SAM/SOM concentric-circle diagram in muted blues and grays518* Specific, believable market sizing numbers:519520 * **TAM:** $42B521 * **SAM:** $8.7B522 * **SOM:** $340M523* A clean bar chart below showing market growth from **2021 to 2026**, with a subtle upward trend524* Small footnotes: **"AGI Research, 2024"** and **"Internal analysis"**525* A company logo placeholder in the bottom-right corner526527The design should look like it belongs in a deck that actually raised money: highly readable text, clear data hierarchy, polished spacing, and professional startup-style visual language.528529Avoid clip art, stock photography, gradients, shadows, decorative elements, or anything that feels generic or overdesigned.530```531532Example outputs:533534535536 <figure className="m-0 min-w-0">537 <figcaption className="mb-2 min-h-10 text-sm font-semibold">538 GPT Image 2.5 Flare539 </figcaption>540 541542543544545 </figure>546 <figure className="m-0 min-w-0">547 <figcaption className="mb-2 min-h-10 text-sm font-semibold">548 GPT Image 2.5 Sunburst549 </figcaption>550 551552553554555 </figure>556557558559## Edit images560561Use `client.images.edit` with the referenced input images. For local edits that require a mask, see [editing with a mask](https://developers.openai.com/api/docs/guides/image-generation#edit-an-image-using-a-mask).562563### Translate while preserving layout564565Use each model's coffee-machine diagram from [Explain a process visually](#explain-a-process-visually) as the input. Ask to replace its text while keeping the design unchanged, then check the translation and any words left in the original language.566567Edit settings: `size="1024x1536"`, `quality="high"`.568569```text570Translate the text in the infographic to Spanish. Do not change any other aspect of the image.571```572573Example outputs:574575576577 <figure className="m-0 min-w-0">578 <figcaption className="mb-2 min-h-10 text-sm font-semibold">579 GPT Image 2.5 Flare580 </figcaption>581 582583584585586 </figure>587 <figure className="m-0 min-w-0">588 <figcaption className="mb-2 min-h-10 text-sm font-semibold">589 GPT Image 2.5 Sunburst590 </figcaption>591 592593594595596 </figure>597598599600### Transfer a visual style601602Assign the reference image a specific role: its palette, texture, or visual medium. Describe the new subject separately. Use the pixel-art image below as the input.603604Edit settings: `size="1024x1536"`, `quality="medium"`.605606```text607Use the same style from the input image and generate a man riding a motorcycle on a white background.608```609610Input image:611612613614615616617618Example outputs:619620621622 <figure className="m-0 min-w-0">623 <figcaption className="mb-2 min-h-10 text-sm font-semibold">624 GPT Image 2.5 Flare625 </figcaption>626 627628629630631 </figure>632 <figure className="m-0 min-w-0">633 <figcaption className="mb-2 min-h-10 text-sm font-semibold">634 GPT Image 2.5 Sunburst635 </figcaption>636 637638639640641 </figure>642643644645### Preserve identity and change clothing646647Use the person photograph and three clothing references below as inputs. State which aspects of the person must remain fixed, and allow only the clothing to change. This pattern also applies to edits where a product or object must remain recognizable.648649Edit settings: `size="1024x1536"`, `quality="medium"`.650651```text652Edit the image to dress the woman using the provided clothing images. Do not change her face, facial features, skin tone, body shape, pose, or identity in any way. Preserve her exact likeness, expression, hairstyle, and proportions. Replace only the clothing, fitting the garments naturally to her existing pose and body geometry with realistic fabric behavior. Match lighting, shadows, and color temperature to the original photo so the outfit integrates photorealistically, without looking pasted on. Do not change the background, camera angle, framing, or image quality, and do not add accessories, text, logos, or watermarks.653```654655Input images:656657658659 660661662663664 665666667668669 670671672673674 675676677678679680681682Example outputs:683684685686 <figure className="m-0 min-w-0">687 <figcaption className="mb-2 min-h-10 text-sm font-semibold">688 GPT Image 2.5 Flare689 </figcaption>690 691692693694695 </figure>696 <figure className="m-0 min-w-0">697 <figcaption className="mb-2 min-h-10 text-sm font-semibold">698 GPT Image 2.5 Sunburst699 </figcaption>700 701702703704705 </figure>706707708709### Combine references710711Pass the scene photograph as image 1 and the dog photograph as image 2. Specify which element to move, its destination, and what must remain unchanged.712713Edit settings: `size="1024x1536"`, `quality="medium"`.714715```text716Place the dog from the second image into the setting of image 1, right next to the woman, use the same style of lighting, composition and background. Do not change anything else.717```718719Input images:720721722723 724725726727728 729730731732733734735736Example outputs:737738739740 <figure className="m-0 min-w-0">741 <figcaption className="mb-2 min-h-10 text-sm font-semibold">742 GPT Image 2.5 Flare743 </figcaption>744 745746747748749 </figure>750 <figure className="m-0 min-w-0">751 <figcaption className="mb-2 min-h-10 text-sm font-semibold">752 GPT Image 2.5 Sunburst753 </figcaption>754 755756757758759 </figure>760761762763764765766### Create a transparent product cutout767768Request both an isolated subject in the prompt and `background="transparent"` in the API. Use PNG or WebP, preserve the returned alpha channel, and omit `output_compression` for PNG. A drawn checkerboard is not transparency. For subsequent edits, repeat the requirement to preserve the transparent background. Use the product photograph below as the input.769770Edit settings: `size="1024x1536"`, `quality="medium"`, `background="transparent"`, `output_format="png"`.771772```text773Extract the product from the input image and isolate it on a fully transparent background.774Output: centered product, crisp silhouette, no halos/fringing.775Preserve product geometry and label legibility exactly.776Add only light polishing. Do not add a solid backdrop, checkerboard, scenery, or shadow.777Do not restyle the product; remove the background and preserve clean alpha transparency.778```779780Input image:781782783784785786787788Example outputs:789790791792 <figure className="m-0 min-w-0">793 <figcaption className="mb-2 min-h-10 text-sm font-semibold">794 GPT Image 2.5 Flare795 </figcaption>796 797798799800801 </figure>802 <figure className="m-0 min-w-0">803 <figcaption className="mb-2 min-h-10 text-sm font-semibold">804 GPT Image 2.5 Sunburst805 </figcaption>806 807808809810811 </figure>812813814815### Turn a drawing into a realistic image816817Sketch-to-render workflows are great for turning rough drawings into photorealistic concepts while keeping the original intent. Treat the prompt like a spec: preserve layout and perspective, then _add realism_ by specifying plausible materials, lighting, and environment. Include "do not add new elements/text" to avoid creative reinterpretations.818819Edit settings: `size="1024x1536"`, `quality="medium"`.820821```text822Turn this drawing into a photorealistic image.823Preserve the exact layout, proportions, and perspective.824Choose realistic materials and lighting consistent with the sketch intent.825Do not add new elements or text.826```827828Input image:829830831832833834835836Example outputs:837838839840 <figure className="m-0 min-w-0">841 <figcaption className="mb-2 min-h-10 text-sm font-semibold">842 GPT Image 2.5 Flare843 </figcaption>844 845846847848849 </figure>850 <figure className="m-0 min-w-0">851 <figcaption className="mb-2 min-h-10 text-sm font-semibold">852 GPT Image 2.5 Sunburst853 </figcaption>854 855856857858859 </figure>860861862863### Remove an object864865Remove one object by naming it explicitly and preserving everything around it. Keep the person, pose, lighting, and composition unchanged so the edit stays local.866867Edit settings: `size="1024x1536"`, `quality="medium"`.868869```text870Remove the flower from man's hand. Do not change anything else.871```872873Input image:874875876877878879880881Example outputs:882883884885 <figure className="m-0 min-w-0">886 <figcaption className="mb-2 min-h-10 text-sm font-semibold">887 GPT Image 2.5 Flare888 </figcaption>889 890891892893894 </figure>895 <figure className="m-0 min-w-0">896 <figcaption className="mb-2 min-h-10 text-sm font-semibold">897 GPT Image 2.5 Sunburst898 </figcaption>899 900901902903904 </figure>905906907908### Insert a person into a scene909910Insert a person into a new scene while preserving their identity. Specify natural lighting, believable detail, body framing, gaze, and interaction with the scene. State which facial features and proportions must remain unchanged. For `gpt-image-2`, omit `input_fidelity`; image inputs are always processed at high fidelity.911912Use the [woman in the museum](https://developers.openai.com/images/platform/guides/image-prompting/woman-in-museum.webp) as the input image.913914Edit settings: `size="1024x1536"`, `quality="medium"`.915916```text917Generate a highly realistic action scene where this person is running away from a large, realistic brown bear attacking a campsite. The image should look like a real photograph someone could have taken, not an overly enhanced or cinematic movie-poster image.918She is centered in the image but looking away from the camera, wearing outdoorsy camping attire, with dirt on her face and tears in her clothing. She is clearly afraid but focused on escaping, running away from the bear as it destroys the campsite behind her.919The campsite is in Yosemite National Park, with believable natural details. The time of day is dusk, with natural lighting and realistic colors. Everything should feel grounded, authentic, and unstyled, as if captured in a real moment. Avoid cinematic lighting, dramatic color grading, or stylized composition.920```921922Example outputs:923924925926 <figure className="m-0 min-w-0">927 <figcaption className="mb-2 min-h-10 text-sm font-semibold">928 GPT Image 2.5 Flare929 </figcaption>930 931932933934935 </figure>936 <figure className="m-0 min-w-0">937 <figcaption className="mb-2 min-h-10 text-sm font-semibold">938 GPT Image 2.5 Sunburst939 </figcaption>940 941942943944945 </figure>946947948949## Refine an image across turns950951Start with one output, inspect it, and use it as the next input. Keep each follow-up narrow so you can see which change helped.952953### Create the starting image954955Use the shampoo photograph from [Create a transparent product cutout](#create-a-transparent-product-cutout) as the input for this billboard scene. Quote the label text exactly.956957Edit settings: `size="1024x1536"`, `quality="medium"`.958959```text960Create a realistic billboard mockup of the shampoo on a highway scene during sunset.961Billboard text (EXACT, verbatim, no extra characters):962"Fresh and clean"963Typography: bold sans-serif, high contrast, centered, clean kerning.964Ensure text appears once and is perfectly legible.965No watermarks, no logos.966```967968Input image:969970971972973974975976Example outputs:977978979980 <figure className="m-0 min-w-0">981 <figcaption className="mb-2 min-h-10 text-sm font-semibold">982 GPT Image 2.5 Flare983 </figcaption>984 985986987988989 </figure>990 <figure className="m-0 min-w-0">991 <figcaption className="mb-2 min-h-10 text-sm font-semibold">992 GPT Image 2.5 Sunburst993 </figcaption>994 995996997998999 </figure>1000100110021003### Change one condition10041005Pass each model's billboard output from the previous step into its next edit request. This short follow-up changes the weather while retaining the existing scene.10061007Edit settings: `size="1024x1536"`, `quality="medium"`.10081009```text1010Make it look like a winter evening with snowfall.1011```10121013Example outputs:1014101510161017 <figure className="m-0 min-w-0">1018 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1019 GPT Image 2.5 Flare1020 </figcaption>1021 10221023102410251026 </figure>1027 <figure className="m-0 min-w-0">1028 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1029 GPT Image 2.5 Sunburst1030 </figcaption>1031 10321033103410351036 </figure>1037103810391040### Keep a character consistent10411042For a book with multiple illustrations, create a reusable character reference to help preserve appearance across scenes, poses, and pages. Change the environment and story while repeating the character’s defining details.10431044#### Establish the character10451046Define the character’s appearance, proportions, outfit, and tone.10471048Generation settings: `size="1024x1536"`, `quality="medium"`.10491050```text1051Create a children’s book illustration introducing a main character.10521053Character:1054A young, storybook-style hero inspired by a little forest outlaw,1055wearing a simple green hooded tunic, soft brown boots, and a small belt pouch.1056The character has a kind expression, gentle eyes, and a brave but warm demeanor.1057Carries a small wooden bow used only for helping, never harming.10581059Theme:1060The character protects and rescues small forest animals like squirrels, birds, and rabbits.10611062Style:1063Children’s book illustration, hand-painted watercolor look,1064soft outlines, warm earthy colors, whimsical and friendly.1065Proportions suitable for picture books (slightly oversized head, expressive face).10661067Constraints:1068- Original character (no copyrighted characters)1069- No text1070- No watermarks1071- Plain forest background to clearly showcase the character1072```10731074Example outputs:1075107610771078 <figure className="m-0 min-w-0">1079 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1080 GPT Image 2.5 Flare1081 </figcaption>1082 10831084108510861087 </figure>1088 <figure className="m-0 min-w-0">1089 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1090 GPT Image 2.5 Sunburst1091 </figcaption>1092 10931094109510961097 </figure>1098109911001101#### Continue the story11021103Reuse each model's generated character image and describe a new scene. Repeat the appearance constraints so the character stays consistent.11041105Edit settings: `size="1024x1536"`, `quality="medium"`.11061107```text1108Continue the children’s book story using the same character.11091110Scene:1111The same young forest hero is gently helping a frightened squirrel1112out of a fallen tree after a winter storm.1113The character kneels beside the squirrel, offering reassurance.11141115Character Consistency:1116- Same green hooded tunic1117- Same facial features, proportions, and color palette1118- Same gentle, heroic personality11191120Style:1121Children’s book watercolor illustration,1122soft lighting, snowy forest environment,1123warm and comforting mood.11241125Constraints:1126- Do not redesign the character1127- No text1128- No watermarks1129```11301131Example outputs:1132113311341135 <figure className="m-0 min-w-0">1136 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1137 GPT Image 2.5 Flare1138 </figcaption>1139 11401141114211431144 </figure>1145 <figure className="m-0 min-w-0">1146 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1147 GPT Image 2.5 Sunburst1148 </figcaption>1149 11501151115211531154 </figure>1155115611571158## More workflows11591160### Change furniture in a room11611162Visualize furniture or décor changes in real spaces without recreating the entire scene. The goal is surgical realism: swap a single object while preserving camera angle, lighting, shadows, and surrounding context so the edit looks like a real photograph, not a redesign.11631164Edit settings: `size="1536x1024"`, `quality="medium"`.11651166```text1167In this room photo, replace ONLY the white chairs with chairs made of wood.1168Preserve camera angle, room lighting, floor shadows, and surrounding objects.1169Keep all other aspects of the image unchanged.1170Photorealistic contact shadows and fabric texture.1171```11721173Input image:11741175117611771178117911801181Example outputs:1182118311841185 <figure className="m-0 min-w-0">1186 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1187 GPT Image 2.5 Flare1188 </figcaption>1189 11901191119211931194 </figure>1195 <figure className="m-0 min-w-0">1196 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1197 GPT Image 2.5 Sunburst1198 </figcaption>1199 12001201120212031204 </figure>1205120612071208### Design a holiday card12091210For seasonal card concepts, describe the scene, emotional tone, materials, lighting, and exact copy. For a 3D pop-up or photographed-card treatment, specify paper layers, fibers, folds, and soft studio lighting. The example below uses a nostalgic teddy-bear scene.12111212Generation settings: `size="1024x1536"`, `quality="medium"`.12131214```text1215Create a Christmas holiday card illustration.12161217Scene:1218a cozy Christmas scene with an old teddy bear sitting inside a keepsake box, slightly worn fur, soft stitching repairs, placed near a window with falling snow outside. The scene suggests the child has grown up, but the memories remain.12191220Mood:1221Warm, nostalgic, gentle, emotional.12221223Style:1224Premium holiday card photography, soft cinematic lighting,1225realistic textures, shallow depth of field,1226tasteful bokeh lights, high print-quality composition.12271228Constraints:1229- Original artwork only1230- No trademarks1231- No watermarks1232- No logos12331234Include ONLY this card text (verbatim):1235"Merry Christmas — some memories never fade."1236```12371238Example outputs:1239124012411242 <figure className="m-0 min-w-0">1243 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1244 GPT Image 2.5 Flare1245 </figcaption>1246 12471248124912501251 </figure>1252 <figure className="m-0 min-w-0">1253 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1254 GPT Image 2.5 Sunburst1255 </figcaption>1256 12571258125912601261 </figure>1262126312641265### Design collectible merchandise12661267Explore merchandise and packaging concepts using product photography cues: materials, packaging, and print clarity. Keep designs original and non-infringing, and compare multiple character or packaging variants.12681269Generation settings: `size="1024x1536"`, `quality="medium"`.12701271```text1272Create a collectible action figure of a vintage-style toy propeller airplane with rounded wings, a front-mounted spinning propeller, slightly worn paint edges, classic childhood proportions, designed as a nostalgic holiday collectible, in blister packaging.12731274Concept:1275A nostalgic holiday collectible inspired by the simple toy airplanes1276children used to play with during winter holidays.1277Evokes warmth, imagination, and childhood wonder.12781279Style:1280Premium toy photography, realistic plastic and painted metal textures,1281studio lighting, shallow depth of field,1282sharp label printing, high-end retail presentation.12831284Constraints:1285- Original design only1286- No trademarks1287- No watermarks1288- No logos12891290Include ONLY this packaging text (verbatim):1291"Christmas Memories Edition"1292```12931294Example outputs:1295129612971298 <figure className="m-0 min-w-0">1299 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1300 GPT Image 2.5 Flare1301 </figcaption>1302 13031304130513061307 </figure>1308 <figure className="m-0 min-w-0">1309 <figcaption className="mb-2 min-h-10 text-sm font-semibold">1310 GPT Image 2.5 Sunburst1311 </figcaption>1312 13131314131513161317 </figure>1318131913201321## Run a complete example13221323This runnable example remains pinned to `gpt-image-2`. Use it as a baseline, then choose an available model and its supported request settings for your evaluation.13241325The examples below generate four logo variations and extract a product onto a transparent background. Install the [OpenAI SDK](https://developers.openai.com/api/docs/libraries#install-an-official-sdk) with `pip install openai` for Python or `gem install openai` for Ruby. Set `OPENAI_API_KEY` and save the [product photograph](https://developers.openai.com/images/platform/guides/image-prompting/shampoo.webp) as `input_images/shampoo.webp`. Live requests incur API usage charges.1326132713281329### View the complete example133013311332 Generate and edit transparent assets13331334```python1335import base641336from pathlib import Path13371338from openai import OpenAI13391340client = OpenAI()134113421343prompt = """1344Create an original, non-infringing logo for a company called Field & Flour, a local bakery.1345The logo should feel warm, simple, and timeless. Use clean, vector-like shapes, a strong silhouette, and balanced negative space.1346Favor simplicity over detail so it reads clearly at small and large sizes. Flat design, minimal strokes, no gradients unless essential.1347Fully transparent background. Deliver a single centered logo with generous padding, clean alpha edges, and no solid backdrop, scenery, checkerboard, or watermark.1348"""13491350result = client.images.generate(1351 model="gpt-image-2",1352 prompt=prompt,1353 size="1024x1536",1354 quality="medium",1355 background="transparent",1356 output_format="png",1357 n=4, # Generate 4 versions of the logo1358)13591360# Preserve the returned PNG bytes, including the alpha channel.1361for index, item in enumerate(result.data, start=1):1362 Path(f"logo-generation-{index}-gpt-image-2.png").write_bytes(1363 base64.b64decode(item.b64_json)1364 )13651366# Extract a product from a reference image.1367prompt = """1368Extract the product from the input image and isolate it on a fully transparent background.1369Output: centered product, crisp silhouette, no halos/fringing.1370Preserve product geometry and label legibility exactly.1371Add only light polishing. Do not add a solid backdrop, checkerboard, scenery, or shadow.1372Do not restyle the product; remove the background and preserve clean alpha transparency.1373"""13741375result = client.images.edit(1376 model="gpt-image-2",1377 image=[1378 Path("input_images/shampoo.webp"),1379 ],1380 prompt=prompt,1381 size="1024x1536",1382 quality="medium",1383 background="transparent",1384 output_format="png",1385)13861387Path("extract-product-gpt-image-2.png").write_bytes(1388 base64.b64decode(result.data[0].b64_json)1389)1390```13911392```ruby1393require "base64"1394require "openai"1395require "pathname"13961397client = OpenAI::Client.new1398result = client.images.generate(1399 model: "gpt-image-2",1400 prompt: "Create an original logo for Field & Flour, a local bakery. Use warm, simple shapes on a fully transparent background, with clean alpha edges and no shadow or checkerboard.",1401 size: "1024x1536", quality: :medium, background: :transparent, output_format: :png, n: 41402)1403Array(result.data).each_with_index do |item, index|1404 File.binwrite("logo-generation-#{index + 1}-gpt-image-2.png", Base64.strict_decode64(item.b64_json || raise("No PNG returned")))1405end1406result = client.images.edit(1407 model: "gpt-image-2", image: OpenAI::FilePart.new(Pathname("input_images/shampoo.webp"), content_type: "image/webp"),1408 prompt: "Extract the product onto a fully transparent background. Preserve its geometry and label, with clean edges and no shadow or restyling.",1409 size: "1024x1536", quality: :medium, background: :transparent, output_format: :png1410)1411File.binwrite("extract-product-gpt-image-2.png", Base64.strict_decode64(Array(result.data).fetch(0).b64_json || raise("No PNG returned")))1412```141314141415141614171418For additional prompts and complete workflows, see the [original notebook](https://github.com/openai/openai-cookbook/blob/d310dfa05d20fb653caa9c1c4b89ac1a4aeeeae4/examples/multimodal/image-gen-models-prompting-guide.ipynb).14191420## Check the result14211422Check the output against the requirements before using it:14231424- Is required text accurate and legible? Are diagram labels and relationships correct?1425- Do identities, product shapes, labels, and reference details remain intact?1426- Did the edit change only what you requested?1427- If transparency is required, does the file contain an alpha channel rather than a painted background?14281429Compare quality, latency, and cost on representative inputs when changing prompts or models. See [image generation pricing](https://developers.openai.com/api/docs/pricing#image-generation) for current costs.143014311432 14331434 14351436 <header className="not-prose mb-8">1437 <h21438 id="gpt-image-2-guide"1439 className="m-0 text-3xl font-semibold text-default"1440 >1441 {"GPT Image 2 reference"}1442 </h2>1443 14441445 Overview and request settings for existing GPT Image 2 workflows.1446 14471448 </header>1449 14501451## Overview14521453GPT Image 2 supports image generation and editing, including text rendering, reference-based edits, and flexible output sizes. Use this reference to maintain existing integrations. The [prompting guide](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5) covers shared techniques for composition, text, reference images, and preserving details during edits. Its illustrated examples use GPT Image 2.5 Flare and GPT Image 2.5 Sunburst; outputs can differ across models. For migration, use the guide's [model selection](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5#choose-a-model) and [evaluation workflow](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5#migrate-an-existing-workflow).14541455## Model parameters14561457Use `client.images.generate` for generation and `client.images.edit` for edits. See the [image generation guide](https://developers.openai.com/api/docs/guides/image-generation) for API setup and request examples.14581459| Parameter | GPT Image 2 |1460| -------------------- | -------------------------------------------------------------------------------------------------------------------- |1461| `model` | `gpt-image-2` |1462| `quality` | `low`, `medium`, `high`, or `auto` |1463| `size` | `auto` or a supported resolution; see [size constraints](https://developers.openai.com/api/docs/guides/image-generation#size-and-quality-options) |1464| `input_fidelity` | Omit it. Image inputs are always processed at high fidelity. |1465| `output_format` | `png`, `jpeg`, or `webp` |1466| `background` | For transparent output, explicitly set `transparent` and use PNG or WebP. |1467| `output_compression` | Use only for JPEG or WebP output, not PNG. |14681469Transparent backgrounds are available in preview for `gpt-image-2`.14701471For the original prompts, inputs, and runnable workflows, see the pinned [GPT Image 2 notebook](https://github.com/openai/openai-cookbook/blob/d310dfa05d20fb653caa9c1c4b89ac1a4aeeeae4/examples/multimodal/image-gen-models-prompting-guide.ipynb).147214731474 14751476 14771478 <header className="not-prose mb-8">1479 <h21480 id="gpt-image-1.5-guide"1481 className="m-0 text-3xl font-semibold text-default"1482 >1483 {"GPT Image 1.5 reference"}1484 </h2>1485 14861487 Overview and request settings for existing GPT Image 1.5 workflows.1488 14891490 </header>1491 14921493## Overview14941495**Deprecated model.** `gpt-image-1.5` is scheduled to shut down on December 1,1496 2026. See the [deprecation1497 notice](https://developers.openai.com/api/docs/deprecations#2026-06-02-gpt-image-model-deprecations) and1498 validate existing workflows with `gpt-image-2` before migrating.14991500GPT Image 1.5 supports image generation and editing, including text rendering, photorealistic images, and reference-based edits. Use this reference to maintain existing integrations. The [prompting guide](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5) covers shared techniques for composition, text, reference images, and preserving details during edits. Test those techniques with your model and inputs; outputs can differ across models.15011502## Model parameters15031504Use `client.images.generate` for generation and `client.images.edit` for edits. See the [image generation guide](https://developers.openai.com/api/docs/guides/image-generation) for API setup and request examples.15051506| Parameter | GPT Image 1.5 |1507| -------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |1508| `model` | `gpt-image-1.5` |1509| `quality` | `low`, `medium`, `high`, or `auto` |1510| `size` | `1024x1024`, `1024x1536`, `1536x1024`, or `auto` |1511| `output_format` | `png`, `jpeg`, or `webp` |1512| `output_compression` | 0 to 100, for JPEG or WebP output only |1513| `background` | Set `transparent` explicitly for transparent output; use PNG or WebP |1514| `input_fidelity` | `low` or `high`; `high` preserves input details, while `quality` controls output generation. Omit this parameter when migrating to GPT Image 2, which always uses high input fidelity. |151515161517 15181519 15201521 <header className="not-prose mb-8">1522 <h21523 id="gpt-image-1-guide"1524 className="m-0 text-3xl font-semibold text-default"1525 >1526 {"GPT Image 1 reference"}1527 </h2>1528 15291530 Overview and request settings for existing GPT Image 1 workflows.1531 15321533 </header>1534 15351536## Overview15371538**Deprecated model.** `gpt-image-1` is scheduled to shut down on October 23,1539 2026. See the [deprecation1540 notice](https://developers.openai.com/api/docs/deprecations#2026-04-22-legacy-gpt-model-snapshots) and1541 validate existing workflows with `gpt-image-2` before migrating.15421543GPT Image 1 supports image generation and editing with reference images and masks. Use this reference to maintain existing integrations. For shared techniques such as describing a scene, preserving details, and refining an edit, see the [prompting guide](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5).15441545## Model parameters15461547Use `client.images.generate` for generation and `client.images.edit` for edits. See the [image generation guide](https://developers.openai.com/api/docs/guides/image-generation) for API setup and request examples.15481549| Parameter | GPT Image 1 |1550| -------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |1551| `model` | `gpt-image-1` |1552| `quality` | `low`, `medium`, `high`, or `auto` |1553| `size` | `1024x1024`, `1024x1536`, `1536x1024`, or `auto` |1554| `output_format` | `png`, `jpeg`, or `webp` |1555| `output_compression` | 0 to 100, for JPEG or WebP output only |1556| `background` | Set `transparent` explicitly for transparent output; use PNG or WebP |1557| `input_fidelity` | `low` or `high`; `high` preserves input details, while `quality` controls output generation. High input fidelity uses more image input tokens. Omit this parameter when migrating to GPT Image 2, which always uses high input fidelity. |