2 2
3> For the complete documentation index, see [llms.txt](/llms.txt). Markdown versions of documentation pages are available by appending `.md` to the page URL.3> For the complete documentation index, see [llms.txt](/llms.txt). Markdown versions of documentation pages are available by appending `.md` to the page URL.
4 4
5<div
6 className="image-prompting-guide [&_img]:w-auto!"
7 data-legacy-heading-model="gpt-image-2.5"
8>
9
10
5<header className="not-prose mb-8">11<header className="not-prose mb-8">
12 <h2
13 id="gpt-image-2.6-guide"
14 className="m-0 text-3xl font-semibold text-default"
15 >
16 {"GPT Image 2.6 prompting guide"}
17 </h2>
18
19
20 Choose a model, tune complexity, and guide image edits.
21
22
23 </header>
24
25
26## Overview
27
28Start 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.
29
30GPT Image 2.6 Flare is the smaller model for workflows where speed matters most. GPT Image 2.6 Sunburst is the base model for workflows that need more detail or precise edits. Both support image generation, editing, and transparent backgrounds.
31
32For API setup and request examples, see the [image generation guide](https://developers.openai.com/api/docs/guides/image-generation).
33
34## Choose a model
35
36For a new workflow, start with **GPT Image 2.6 Flare** if response time matters most. Start with **GPT Image 2.6 Sunburst** if the image must meet demanding requirements for detail, layout, or edit accuracy.
37
38| Your current workflow | Start by testing |
39| --------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- |
40| Your existing GPT Image 2 workflow meets your quality requirements, or you use GPT Image 2.5 Flare. | `gpt-image-2.6-flare`. Check quality and response time on the prompts you already use. |
41| GPT Image 2 falls short on a complex task, or you use GPT Image 2.5 Sunburst. | `gpt-image-2.6-sunburst`. Check the details that determine whether you can use the image. |
42
43If Sunburst meets your requirements, try Flare with the same prompts, reference images, dimensions, and quality setting. Compare the results and response times on your workload.
44
45## Model parameters
46
47Set these fields in the API request. Use the prompt to describe the image you want.
48
49| Parameter | GPT Image 2.6 settings |
50| --------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
51| `model` | Use `gpt-image-2.6-flare` or `gpt-image-2.6-sunburst`. |
52| `quality` | Controls how much visual detail the model can render, such as texture or small lettering. Choose `auto`, `low`, `medium`, `high`, `xhigh`, or `max`. |
53| `complexity` | Controls how much the model plans before rendering. Choose `low`, `medium`, or `high`. Both Sunburst and Flare use `high` when you omit the field. |
54| `size` | Sets the output dimensions in pixels. Use `auto` or a supported size such as `1024x1024` (square), `1536x1024` (landscape), or `1024x1536` (portrait). |
55| `background` | Use `auto`, `opaque`, or `transparent`. For transparent PNG or WebP, check the decoded alpha channel. [SVG output](#generate-svg-output) also supports transparency. |
56| `output_format` | Use `png`, `jpeg`, or `webp`. Generations also support [`svg`](#generate-svg-output) and [`gif`](#generate-a-looping-gif). Only JPEG and WebP support `output_compression`. |
57| `markup` | On edits, upload a transparent PNG overlay with the same pixel dimensions as the first input image. See [Guide an edit with markup](#guide-an-edit-with-markup). |
58
59For a complex layout, start with the default and check how the model placed labels and connected objects. To compare complexity levels, keep `quality` fixed. If the layout is correct but small text is hard to read, compare quality settings at the same complexity. See [Compare complexity](#compare-complexity).
60
61For a custom `size`, use multiples of 16 for both dimensions and an aspect ratio from 1:3 to 3:1. The landscape slide example uses `1536x864`.
62
63Omit `input_fidelity` on image edits. GPT Image 2.6 does not support `reasoning.summary`.
64
65## Migrate an existing workflow
66
671. **Save a baseline.** Collect representative prompts, reference images, and outputs from your current model. Include difficult edits, exact text, product geometry, and transparent assets where relevant.
682. **Test the corresponding model.** For a 2.5 Flare workflow, try 2.6 Flare first. For a 2.5 Sunburst workflow, try 2.6 Sunburst first. For a GPT Image 2 workflow, use the [model selection table](#choose-a-model).
693. **Keep the first comparison consistent.** Use the same prompt, reference images, size, quality, and output format. Leave complexity at the model default.
704. **Inspect the result.** Check text, layout, identity, and details that should survive an edit. For a sequence of edits, check the final image as well as each intermediate result.
715. **Tune one setting at a time.** Change either quality or complexity. Repeat each request and compare how often the output meets your criteria and how long it takes.
726. **Roll out by workflow.** Move a small share of traffic after the new model meets your criteria. Track response time, failures, retries, and cost per accepted image before you expand.
73
74When migrating from GPT Image 1 or 1.5, use its reference tab to check parameter differences and shutdown dates. If part of an image must remain pixel-identical, composite the approved edit into the original.
75
76## Prompting fundamentals
77
781. **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.
792. **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.
803. **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.
814. **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.
825. **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.
836. **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.
847. **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.
858. **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.
86
87Most examples omit `complexity` and use the model default. Adapt the prompts to your own images and requirements.
88
89## Generate images
90
91### Compare complexity
92
93When a prompt leaves design decisions open, try `high` complexity to give the model more room to plan the composition. This prompt names the scene and the material but leaves the trophy, uniforms, and lettering up to the model.
94
95> A premium handcrafted layered paper-cut illustration of a football team celebrating a championship victory inside a packed stadium.
96
97GPT Image 2.6 Sunburst generated one image at each complexity level with the same prompt, `size="1024x1536"`, `quality="medium"`, and PNG output. Compare the trophy and where the word “CHAMPIONS” appears.
98
99
100
101 <figure className="m-0 min-w-0">
102 <figcaption className="mb-2 text-sm font-semibold">
103 Low complexity
104 </figcaption>
105
106
107
108
109
110 </figure>
111 <figure className="m-0 min-w-0">
112 <figcaption className="mb-2 text-sm font-semibold">
113 Medium complexity
114 </figcaption>
115
116
117
118
119
120 </figure>
121 <figure className="m-0 min-w-0">
122 <figcaption className="mb-2 text-sm font-semibold">
123 High complexity
124 </figcaption>
125
126
127
128
129
130 </figure>
131
132
133
134In these outputs, `low` places a crown on the trophy. At `medium`, a laurel wreath decorates the cup and crests appear on the shirts. Both put a large banner in front of the team. At `high`, the sign moves into the stands, leaving the foreground for the players and the folds of their blue paper jerseys.
135
136- **Fully specified designs.** When the prompt already sets the layout, text, and style, higher complexity may make little visible difference.
137- **Targeted edits.** Start with `low` when you want to change one detail. Check that the surrounding details stay intact.
138
139### Generate SVG output
140
141For a logo you can scale or recolor by editing its paths, send the [Field & Flour bakery logo prompt](#design-a-reusable-logo) to `POST /v1/images/generations`. Set `output_format="svg"` and `n=1`. SVG also supports `background="transparent"`, so the space around the mark can take the color of the surface behind it.
142
143Omit `stream`, `partial_images`, `output_compression`, and `prompt_rewriter` from the SVG request. Decode `data[0].b64_json` from base64 and save the resulting XML as a `.svg` file.
144
145
146
147 <figure className="m-0 min-w-0 [&_[data-codex-screenshot-overlay]_img]:bg-surface">
148 <figcaption className="mb-2 text-sm font-semibold">
149 GPT Image 2.6 Flare
150 </figcaption>
151
152
153
154
155
156 </figure>
157 <figure className="m-0 min-w-0 [&_img]:bg-white">
158 <figcaption className="mb-2 text-sm font-semibold">
159 GPT Image 2.6 Sunburst
160 </figcaption>
161
162
163
164
165
166 </figure>
167
168
169
170Open the [Flare SVG](https://developers.openai.com/images/platform/guides/image-prompting/field-and-flour-gpt-image-2-6-flare.svg) or the [Sunburst SVG](https://developers.openai.com/images/platform/guides/image-prompting/field-and-flour-gpt-image-2-6-sunburst.svg) to inspect the artwork.
171
172### Generate a looping GIF
173
174For a smooth loop, give the subject a starting pose to return to and say whether the camera should move.
175
176> A little monkey in an olive-green hooded coat stands in a snowy pine forest at dusk. A wooden cabin glows in the background. The monkey holds a glowing amber paper lantern at chest height, slowly lifts it, looks up and blinks, then lowers it back to its starting pose. Keep the lantern in its hands and the camera still. Make the motion loop smoothly.
177
178
179
180
181
182
183
184Send the prompt to `POST /v1/images/generations` with `output_format="gif"`. This example uses `gpt-image-2.6-sunburst`, `size="512x512"`, `quality="high"`, and the default `high` complexity. With `frame_count=9` and `frame_duration_ms=200`, the loop lasts 1.8 seconds. See [GIF settings and limits](https://developers.openai.com/api/docs/guides/image-generation#generate-animated-gif-output).
185
186### Control style and lighting
187
188Describe a photograph through its subject, framing, light, and texture. This example specifies a candid composition and explicitly excludes heavy retouching.
189
190The example uses `size="1024x1536"`, `quality="medium"`.
191
192```text
193Create a photorealistic candid photograph of an elderly sailor standing on a small fishing boat.
194He has weathered skin with visible wrinkles, pores, and sun texture, and a few faded traditional sailor tattoos on his arms.
195He 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.
196Soft coastal daylight, shallow depth of field, subtle film grain, natural color balance.
197The image should feel honest and unposed, with real skin texture, worn materials, and everyday detail. No glamorization, no heavy retouching.
198```
199
200
201
202 <figure className="m-0 min-w-0">
203 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
204 GPT Image 2.6 Flare
205 </figcaption>
206
207
208
209
210
211 </figure>
212 <figure className="m-0 min-w-0">
213 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
214 GPT Image 2.6 Sunburst
215 </figcaption>
216
217
218
219
220
221 </figure>
222
223
224
225### Explain a process visually
226
227Name the process, audience, and information the image should communicate. For diagrams and information graphics, verify labels and factual relationships as well as appearance.
228
229The example uses `size="1024x1536"`, `quality="medium"`.
230
231```text
232Create a detailed Infographic of the functioning and flow of an automatic coffee machine like a Jura.
233From bean basket, to grinding, to scale, water tank, boiler, etc.
234I'd like to understand technically and visually the flow.
235```
236
237
238
239 <figure className="m-0 min-w-0">
240 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
241 GPT Image 2.6 Flare
242 </figcaption>
243
244
245
246
247
248 </figure>
249 <figure className="m-0 min-w-0">
250 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
251 GPT Image 2.6 Sunburst
252 </figcaption>
253
254
255
256
257
258 </figure>
259
260
261
262### Render exact text
263
264Quote the required copy and tell the model how many times it should appear. Specify the audience and visual treatment without adding unrelated instructions.
265
266The example uses `size="1024x1536"`, `quality="medium"`.
267
268```text
269Give me a cool in culture ad / fashion shot for a brand called Thread.
270It's a hip young street brand. The ad shows a group of friends hanging out together with the tagline "Yours to Create."
271Make it feel like a polished campaign image for a youth streetwear audience: stylish, contemporary, energetic, and tasteful.
272Use clean composition, strong color direction, natural poses, and premium fashion photography cues.
273Render the tagline exactly once, clearly and legibly, integrated into the ad layout.
274No extra text, no watermarks, no unrelated logos.
275```
276
277
278
279 <figure className="m-0 min-w-0">
280 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
281 GPT Image 2.6 Flare
282 </figcaption>
283
284
285
286
287
288 </figure>
289 <figure className="m-0 min-w-0">
290 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
291 GPT Image 2.6 Sunburst
292 </figcaption>
293
294
295
296
297
298 </figure>
299
300
301
302### Design a reusable logo
303
304Describe the brand and the shapes that define the mark. Ask for a silhouette that stays recognizable on a small label or a large sign.
305
306```text
307Create an original, non-infringing logo for a company called Field & Flour, a local bakery.
308The logo should feel warm, simple, and timeless. Use clean, vector-like shapes, a strong silhouette, and balanced negative space.
309Favor simplicity over detail so it reads clearly at small and large sizes. Flat design, minimal strokes, no gradients unless essential.
310Fully transparent background. Deliver a single centered logo with generous padding, clean alpha edges, and no solid backdrop, scenery, checkerboard, or watermark.
311```
312
313For a raster logo you can place on another background, use `output_format="png"` and `background="transparent"`. See the [Cookbook example on generating transparent image assets](https://developers.openai.com/cookbook/examples/multimodal/transparent-image-assets-for-campaigns-and-presentations).
314
315Each model used `size="1024x1536"`, `quality="medium"`, and `n=1` for the prompt above. Increase `n` if you want more raster variations.
316
317
318
319 <figure className="m-0 min-w-0">
320 <figcaption className="mb-2 text-sm font-semibold">
321 GPT Image 2.6 Flare
322 </figcaption>
323
324
325
326
327
328 </figure>
329 <figure className="m-0 min-w-0">
330 <figcaption className="mb-2 text-sm font-semibold">
331 GPT Image 2.6 Sunburst
332 </figcaption>
333
334
335
336
337
338 </figure>
339
340
341
342### Use historical and real-world context
343
344Name the place and date to establish a historical setting. The model can infer contextual details, but inspect clothing, staging, and surroundings for historical accuracy.
345
346The example uses `size="1024x1536"`, `quality="medium"`.
347
348```text
349Create a realistic outdoor crowd scene in Bethel, New York on August 16, 1969.
350Photorealistic, period-accurate clothing, staging, and environment.
351```
352
353
354
355 <figure className="m-0 min-w-0">
356 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
357 GPT Image 2.6 Flare
358 </figcaption>
359
360
361
362
363
364 </figure>
365 <figure className="m-0 min-w-0">
366 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
367 GPT Image 2.6 Sunburst
368 </figcaption>
369
370
371
372
373
374 </figure>
375
376
377
378### Turn a story into a comic strip
379
380For 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.
381
382The example uses `size="1024x1536"`, `quality="medium"`.
383
384```text
385Create a short vertical comic-style reel with 4 panels.
386Panel 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.
387Panel 2: The door clicks shut. Silence breaks. The pet slowly turns toward the empty house, posture shifting, eyes sharp with possibility.
388Panel 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.
389Panel 4: The door opens. The pet is seated perfectly by the entrance, alert and composed, as if nothing happened.
390```
391
392
393
394 <figure className="m-0 min-w-0">
395 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
396 GPT Image 2.6 Flare
397 </figcaption>
398
399
400
401
402
403 </figure>
404 <figure className="m-0 min-w-0">
405 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
406 GPT Image 2.6 Sunburst
407 </figcaption>
408
409
410
411
412
413 </figure>
414
415
416
417### Create an interface preview
418
419Interface 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.
420
421The example uses `size="1024x1536"`, `quality="medium"`.
422
423```text
424Create a realistic mobile app UI mockup for a local farmers market.
425Show 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.
426Design it to be practical, and easy to use. White background, subtle natural accent colors, clear typography, and minimal decoration.
427It should look like a real, well-designed, beautiful app for a small local market.
428Place the UI mockup in an iPhone frame.
429```
430
431
432
433 <figure className="m-0 min-w-0">
434 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
435 GPT Image 2.6 Flare
436 </figcaption>
437
438
439
440
441
442 </figure>
443 <figure className="m-0 min-w-0">
444 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
445 GPT Image 2.6 Sunburst
446 </figcaption>
447
448
449
450
451
452 </figure>
453
454
455
456### Create scientific and educational visuals
457
458Scientific 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.
459
460When 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.
461
462The example uses `size="1536x1024"`, `quality="high"`.
463
464```text
465Create a simple biology diagram titled "Cellular Respiration at a Glance" for high school students.
466
467Show how glucose turns into energy inside a cell. Include glycolysis, the Krebs cycle, and the electron transport chain.
468Use arrows to connect the steps, and label the main molecules: glucose, pyruvate, ATP, NADH, FADH2, CO2, O2, and H2O.
469Make it look like a clean classroom handout or slide, with a white background, simple icons, clear labels, and easy-to-read text.
470
471Avoid tiny text, extra decoration, or anything that makes the diagram hard to understand.
472```
473
474
475
476 <figure className="m-0 min-w-0">
477 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
478 GPT Image 2.6 Flare
479 </figcaption>
480
481
482
483
484
485 </figure>
486 <figure className="m-0 min-w-0">
487 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
488 GPT Image 2.6 Sunburst
489 </figcaption>
490
491
492
493
494
495 </figure>
496
497
498
499### Build slides, diagrams, and charts
500
501Productivity 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.
502
503For 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.
504
505The figures and citations in this prompt are fictional. Check each number and any extra percentage the model adds before using a generated slide.
506
507The example uses `size="1536x864"`, `quality="high"`.
508
509```text
510Create one pitch-deck slide titled **"Market Opportunity"** that feels like a real Series A fundraising slide from a YC-backed startup.
511
512Use a clean white background, modern sans-serif typography like Inter, and a crisp, minimal layout. The slide should include:
513
514* A TAM/SAM/SOM concentric-circle diagram in muted blues and grays
515* Specific, believable market sizing numbers:
516
517 * **TAM:** $42B
518 * **SAM:** $8.7B
519 * **SOM:** $340M
520* A clean bar chart below showing market growth from **2021 to 2026**, with a subtle upward trend
521* Small footnotes: **"AGI Research, 2024"** and **"Internal analysis"**
522* A company logo placeholder in the bottom-right corner
523
524The 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.
525
526Avoid clip art, stock photography, gradients, shadows, decorative elements, or anything that feels generic or overdesigned.
527```
528
529
530
531 <figure className="m-0 min-w-0">
532 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
533 GPT Image 2.6 Flare
534 </figcaption>
535
536
537
538
539
540 </figure>
541 <figure className="m-0 min-w-0">
542 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
543 GPT Image 2.6 Sunburst
544 </figcaption>
545
546
547
548
549
550 </figure>
551
552
553
554## Edit images
555
556Use `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).
557
558### Guide an edit with markup
559
560Draw a circle around the object you want to change, then describe the change in the prompt. This example targets the striped vase on a kitchen table.
561
562Send two files in one multipart `POST /v1/images/edits` request. Use the [original kitchen photo](https://developers.openai.com/images/platform/guides/image-prompting/kitchen.webp) for `image` and the [transparent circle overlay](https://developers.openai.com/images/platform/guides/image-prompting/markup-vase-overlay-gpt-image-2-6.png) for `markup`. Both files are 1200 × 900 pixels. The preview shows the two files together.
563
564The markup overlay applies only to the first input image. Any additional images serve as references. Omit `mask` and `prompt_rewriter` when you send `markup`.
565
566
567
568
569
570
571
572Both edits use `quality="medium"`, `size="auto"`, and PNG output.
573
574```text
575Recolor the striped vase inside the red circle to solid cobalt blue ceramic. Keep its shape, the dried stems, and the rest of the kitchen unchanged. Remove the red circle from the finished photo.
576```
577
578
579
580 <figure className="m-0 min-w-0">
581 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
582 GPT Image 2.6 Flare
583 </figcaption>
584
585
586
587
588
589 </figure>
590 <figure className="m-0 min-w-0">
591 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
592 GPT Image 2.6 Sunburst
593 </figcaption>
594
595
596
597
598
599 </figure>
600
601
602
603Both models turned the vase cobalt blue, kept the dried stems, and removed the red circle. Compare the rest of each image with the original to check for other changes.
604
605### Translate while preserving layout
606
607Use 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.
608
609The example uses `size="1024x1536"`, `quality="high"`.
610
611```text
612Translate the text in the infographic to Spanish. Do not change any other aspect of the image.
613```
614
615
616
617 <figure className="m-0 min-w-0">
618 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
619 GPT Image 2.6 Flare
620 </figcaption>
621
622
623
624
625
626 </figure>
627 <figure className="m-0 min-w-0">
628 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
629 GPT Image 2.6 Sunburst
630 </figcaption>
631
632
633
634
635
636 </figure>
637
638
639
640### Transfer a visual style
641
642Assign 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.
643
644The example uses `size="1024x1536"`, `quality="medium"`.
645
646```text
647Use the same style from the input image and generate a man riding a motorcycle on a white background.
648```
649
650Reference image
651
652
653
654
655
656
657
658
659
660 <figure className="m-0 min-w-0">
661 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
662 GPT Image 2.6 Flare
663 </figcaption>
664
665
666
667
668
669 </figure>
670 <figure className="m-0 min-w-0">
671 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
672 GPT Image 2.6 Sunburst
673 </figcaption>
674
675
676
677
678
679 </figure>
680
681
682
683### Preserve identity and change clothing
684
685Use 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.
686
687The example uses `size="1024x1536"`, `quality="medium"`.
688
689```text
690Edit 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.
691```
692
693Reference images
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722 <figure className="m-0 min-w-0">
723 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
724 GPT Image 2.6 Flare
725 </figcaption>
726
727
728
729
730
731 </figure>
732 <figure className="m-0 min-w-0">
733 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
734 GPT Image 2.6 Sunburst
735 </figcaption>
736
737
738
739
740
741 </figure>
742
743
744
745### Combine references
746
747Pass 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.
748
749The example uses `size="1024x1536"`, `quality="medium"`.
750
751```text
752Place 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.
753```
754
755Reference images
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774 <figure className="m-0 min-w-0">
775 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
776 GPT Image 2.6 Flare
777 </figcaption>
778
779
780
781
782
783 </figure>
784 <figure className="m-0 min-w-0">
785 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
786 GPT Image 2.6 Sunburst
787 </figcaption>
788
789
790
791
792
793 </figure>
794
795
796
797### Create a transparent product cutout
798
799Request 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.
800
801The example uses `size="1024x1536"`, `quality="medium"`, `background="transparent"`, `output_format="png"`.
802
803```text
804Extract the product from the input image and isolate it on a fully transparent background.
805Output: centered product, crisp silhouette, no halos/fringing.
806Preserve product geometry and label legibility exactly.
807Add only light polishing. Do not add a solid backdrop, checkerboard, scenery, or shadow.
808Do not restyle the product; remove the background and preserve clean alpha transparency.
809```
810
811Reference image
812
813
814
815
816
817
818
819
820
821 <figure className="m-0 min-w-0">
822 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
823 GPT Image 2.6 Flare
824 </figcaption>
825
826
827
828
829
830 </figure>
831 <figure className="m-0 min-w-0">
832 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
833 GPT Image 2.6 Sunburst
834 </figcaption>
835
836
837
838
839
840 </figure>
841
842
843
844### Turn a drawing into a realistic image
845
846Sketch-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.
847
848The example uses `size="1024x1536"`, `quality="medium"`.
849
850```text
851Turn this drawing into a photorealistic image.
852Preserve the exact layout, proportions, and perspective.
853Choose realistic materials and lighting consistent with the sketch intent.
854Do not add new elements or text.
855```
856
857Reference image
858
859
860
861
862
863
864
865
866
867 <figure className="m-0 min-w-0">
868 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
869 GPT Image 2.6 Flare
870 </figcaption>
871
872
873
874
875
876 </figure>
877 <figure className="m-0 min-w-0">
878 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
879 GPT Image 2.6 Sunburst
880 </figcaption>
881
882
883
884
885
886 </figure>
887
888
889
890### Remove an object
891
892Remove one object by naming it explicitly and preserving everything around it. Keep the person, pose, lighting, and composition unchanged so the edit stays local.
893
894The example uses `size="1024x1536"`, `quality="medium"`.
895
896```text
897Remove the flower from man's hand. Do not change anything else.
898```
899
900Reference image
901
902
903
904
905
906
907
908
909
910 <figure className="m-0 min-w-0">
911 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
912 GPT Image 2.6 Flare
913 </figcaption>
914
915
916
917
918
919 </figure>
920 <figure className="m-0 min-w-0">
921 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
922 GPT Image 2.6 Sunburst
923 </figcaption>
924
925
926
927
928
929 </figure>
930
931
932
933### Insert a person into a scene
934
935Insert 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. Omit `input_fidelity` from the edit request.
936
937Use the [woman in the museum](https://developers.openai.com/images/platform/guides/image-prompting/woman-in-museum.webp) as the input image.
938
939The example uses `size="1024x1536"`, `quality="medium"`.
940
941```text
942Generate 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.
943She 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.
944The 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.
945```
946
947
948
949 <figure className="m-0 min-w-0">
950 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
951 GPT Image 2.6 Flare
952 </figcaption>
953
954
955
956
957
958 </figure>
959 <figure className="m-0 min-w-0">
960 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
961 GPT Image 2.6 Sunburst
962 </figcaption>
963
964
965
966
967
968 </figure>
969
970
971
972## Refine an image across turns
973
974Start with one output, inspect it, and use it as the next input. Keep each follow-up narrow so you can see which change helped.
975
976### Create the starting image
977
978Use the shampoo photograph from [Create a transparent product cutout](#create-a-transparent-product-cutout) as the input for this billboard scene. Quote the billboard headline exactly.
979
980The example uses `size="1024x1536"`, `quality="medium"`.
981
982```text
983Create a realistic billboard mockup of the shampoo on a highway scene during sunset.
984Billboard text (EXACT, verbatim, no extra characters):
985"Fresh and clean"
986Typography: bold sans-serif, high contrast, centered, clean kerning.
987Ensure text appears once and is perfectly legible.
988No watermarks, no logos.
989```
990
991Reference image
992
993
994
995
996
997
998
999
1000
1001 <figure className="m-0 min-w-0">
1002 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1003 GPT Image 2.6 Flare
1004 </figcaption>
1005
1006
1007
1008
1009
1010 </figure>
1011 <figure className="m-0 min-w-0">
1012 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1013 GPT Image 2.6 Sunburst
1014 </figcaption>
1015
1016
1017
1018
1019
1020 </figure>
1021
1022
1023
1024### Change one condition
1025
1026Pass 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.
1027
1028The example uses `size="1024x1536"`, `quality="medium"`.
1029
1030```text
1031Make it look like a winter evening with snowfall.
1032```
1033
1034
1035
1036 <figure className="m-0 min-w-0">
1037 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1038 GPT Image 2.6 Flare
1039 </figcaption>
1040
1041
1042
1043
1044
1045 </figure>
1046 <figure className="m-0 min-w-0">
1047 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1048 GPT Image 2.6 Sunburst
1049 </figcaption>
1050
1051
1052
1053
1054
1055 </figure>
1056
1057
1058
1059### Keep a character consistent
1060
1061For 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.
1062
1063#### Establish the character
1064
1065Define the character’s appearance, proportions, outfit, and tone.
1066
1067The example uses `size="1024x1536"`, `quality="medium"`.
1068
1069```text
1070Create a children’s book illustration introducing a main character.
1071
1072Character:
1073A young, storybook-style hero inspired by a little forest outlaw,
1074wearing a simple green hooded tunic, soft brown boots, and a small belt pouch.
1075The character has a kind expression, gentle eyes, and a brave but warm demeanor.
1076Carries a small wooden bow used only for helping, never harming.
1077
1078Theme:
1079The character protects and rescues small forest animals like squirrels, birds, and rabbits.
1080
1081Style:
1082Children’s book illustration, hand-painted watercolor look,
1083soft outlines, warm earthy colors, whimsical and friendly.
1084Proportions suitable for picture books (slightly oversized head, expressive face).
1085
1086Constraints:
1087- Original character (no copyrighted characters)
1088- No text
1089- No watermarks
1090- Plain forest background to clearly showcase the character
1091```
1092
1093
1094
1095 <figure className="m-0 min-w-0">
1096 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1097 GPT Image 2.6 Flare
1098 </figcaption>
1099
1100
1101
1102
1103
1104 </figure>
1105 <figure className="m-0 min-w-0">
1106 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1107 GPT Image 2.6 Sunburst
1108 </figcaption>
1109
1110
1111
1112
1113
1114 </figure>
1115
1116
1117
1118#### Continue the story
1119
1120Reuse each model's generated character image and describe a new scene. Repeat the appearance constraints so the character stays consistent.
1121
1122The example uses `size="1024x1536"`, `quality="medium"`.
1123
1124```text
1125Continue the children’s book story using the same character.
1126
1127Scene:
1128The same young forest hero is gently helping a frightened squirrel
1129out of a fallen tree after a winter storm.
1130The character kneels beside the squirrel, offering reassurance.
1131
1132Character Consistency:
1133- Same green hooded tunic
1134- Same facial features, proportions, and color palette
1135- Same gentle, heroic personality
1136
1137Style:
1138Children’s book watercolor illustration,
1139soft lighting, snowy forest environment,
1140warm and comforting mood.
1141
1142Constraints:
1143- Do not redesign the character
1144- No text
1145- No watermarks
1146```
1147
1148
1149
1150 <figure className="m-0 min-w-0">
1151 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1152 GPT Image 2.6 Flare
1153 </figcaption>
1154
1155
1156
1157
1158
1159 </figure>
1160 <figure className="m-0 min-w-0">
1161 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1162 GPT Image 2.6 Sunburst
1163 </figcaption>
1164
1165
1166
1167
1168
1169 </figure>
1170
1171
1172
1173## More workflows
1174
1175### Change furniture in a room
1176
1177Visualize 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.
1178
1179The example uses `size="1536x1024"`, `quality="medium"`.
1180
1181```text
1182In this room photo, replace ONLY the white chairs with chairs made of wood.
1183Preserve camera angle, room lighting, floor shadows, and surrounding objects.
1184Keep all other aspects of the image unchanged.
1185Photorealistic contact shadows and fabric texture.
1186```
1187
1188Reference image
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198 <figure className="m-0 min-w-0">
1199 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1200 GPT Image 2.6 Flare
1201 </figcaption>
1202
1203
1204
1205
1206
1207 </figure>
1208 <figure className="m-0 min-w-0">
1209 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1210 GPT Image 2.6 Sunburst
1211 </figcaption>
1212
1213
1214
1215
1216
1217 </figure>
1218
1219
1220
1221### Design a holiday card
1222
1223For 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.
1224
1225The example uses `size="1024x1536"`, `quality="medium"`.
1226
1227```text
1228Create a Christmas holiday card illustration.
1229
1230Scene:
1231a 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.
1232
1233Mood:
1234Warm, nostalgic, gentle, emotional.
1235
1236Style:
1237Premium holiday card photography, soft cinematic lighting,
1238realistic textures, shallow depth of field,
1239tasteful bokeh lights, high print-quality composition.
1240
1241Constraints:
1242- Original artwork only
1243- No trademarks
1244- No watermarks
1245- No logos
1246
1247Include ONLY this card text (verbatim):
1248"Merry Christmas — some memories never fade."
1249```
1250
1251
1252
1253 <figure className="m-0 min-w-0">
1254 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1255 GPT Image 2.6 Flare
1256 </figcaption>
1257
1258
1259
1260
1261
1262 </figure>
1263 <figure className="m-0 min-w-0">
1264 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1265 GPT Image 2.6 Sunburst
1266 </figcaption>
1267
1268
1269
1270
1271
1272 </figure>
1273
1274
1275
1276### Design collectible merchandise
1277
1278Explore 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.
1279
1280The example uses `size="1024x1536"`, `quality="medium"`.
1281
1282```text
1283Create 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.
1284
1285Concept:
1286A nostalgic holiday collectible inspired by the simple toy airplanes
1287children used to play with during winter holidays.
1288Evokes warmth, imagination, and childhood wonder.
1289
1290Style:
1291Premium toy photography, realistic plastic and painted metal textures,
1292studio lighting, shallow depth of field,
1293sharp label printing, high-end retail presentation.
1294
1295Constraints:
1296- Original design only
1297- No trademarks
1298- No watermarks
1299- No logos
1300
1301Include ONLY this packaging text (verbatim):
1302"Christmas Memories Edition"
1303```
1304
1305
1306
1307 <figure className="m-0 min-w-0">
1308 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1309 GPT Image 2.6 Flare
1310 </figcaption>
1311
1312
1313
1314
1315
1316 </figure>
1317 <figure className="m-0 min-w-0">
1318 <figcaption className="mb-2 min-h-10 text-sm font-semibold">
1319 GPT Image 2.6 Sunburst
1320 </figcaption>
1321
1322
1323
1324
1325
1326 </figure>
1327
1328
1329
1330## Run a complete example
1331
1332Generate four logo variations and extract a product onto a transparent background with the 2.6 model you want to test.
1333
1334Install 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`.
1335
1336Save the sample as `demo.py` or `demo.rb`. To try Flare, run `python demo.py gpt-image-2.6-flare` or `ruby demo.rb gpt-image-2.6-flare`. To try Sunburst, pass `gpt-image-2.6-sunburst` instead. The script writes four logo files in PNG format and `extract-product.png`.
1337
1338
1339
1340### View the complete example
1341
1342
1343 Generate and edit transparent assets
1344
1345```python
1346import argparse
1347import base64
1348from pathlib import Path
1349
1350from openai import OpenAI
1351
1352parser = argparse.ArgumentParser()
1353parser.add_argument("model", help="Image model ID enabled for your project")
1354model = parser.parse_args().model
1355client = OpenAI()
1356
1357prompt = """
1358Create an original, non-infringing logo for a company called Field & Flour, a local bakery.
1359The logo should feel warm, simple, and timeless. Use clean, vector-like shapes, a strong silhouette, and balanced negative space.
1360Favor simplicity over detail so it reads clearly at small and large sizes. Flat design, minimal strokes, no gradients unless essential.
1361Fully transparent background. Deliver a single centered logo with generous padding, clean alpha edges, and no solid backdrop, scenery, checkerboard, or watermark.
1362""".strip()
1363logos = client.images.generate(
1364 model=model,
1365 prompt=prompt,
1366 size="1024x1536",
1367 quality="medium",
1368 background="transparent",
1369 output_format="png",
1370 n=4,
1371)
1372for index, image in enumerate(logos.data, start=1):
1373 Path(f"logo-generation-{index}.png").write_bytes(base64.b64decode(image.b64_json))
1374
1375prompt = """
1376Extract the product from the input image and isolate it on a fully transparent background.
1377Output: centered product, crisp silhouette, no halos/fringing.
1378Preserve product geometry and label legibility exactly.
1379Add only light polishing. Do not add a solid backdrop, checkerboard, scenery, or shadow.
1380Do not restyle the product; remove the background and preserve clean alpha transparency.
1381""".strip()
1382cutout = client.images.edit(
1383 model=model,
1384 image=[("shampoo.webp", Path("input_images/shampoo.webp"), "image/webp")],
1385 prompt=prompt,
1386 size="1024x1536",
1387 quality="medium",
1388 background="transparent",
1389 output_format="png",
1390)
1391Path("extract-product.png").write_bytes(base64.b64decode(cutout.data[0].b64_json))
1392```
1393
1394```ruby
1395require "base64"
1396require "openai"
1397require "pathname"
1398
1399model = ARGV.fetch(0) { abort "Usage: ruby #{$PROGRAM_NAME} MODEL_ID" }
1400client = OpenAI::Client.new
1401
1402prompt = <<~PROMPT.strip
1403 Create an original, non-infringing logo for a company called Field & Flour, a local bakery.
1404 The logo should feel warm, simple, and timeless. Use clean, vector-like shapes, a strong silhouette, and balanced negative space.
1405 Favor simplicity over detail so it reads clearly at small and large sizes. Flat design, minimal strokes, no gradients unless essential.
1406 Fully transparent background. Deliver a single centered logo with generous padding, clean alpha edges, and no solid backdrop, scenery, checkerboard, or watermark.
1407PROMPT
1408logos = client.images.generate(
1409 model: model,
1410 prompt: prompt,
1411 size: "1024x1536",
1412 quality: :medium,
1413 background: :transparent,
1414 output_format: :png,
1415 n: 4
1416)
1417Array(logos.data).each_with_index do |image, index|
1418 File.binwrite("logo-generation-#{index + 1}.png", Base64.strict_decode64(image.b64_json || raise("No PNG returned")))
1419end
1420
1421prompt = <<~PROMPT.strip
1422 Extract the product from the input image and isolate it on a fully transparent background.
1423 Output: centered product, crisp silhouette, no halos/fringing.
1424 Preserve product geometry and label legibility exactly.
1425 Add only light polishing. Do not add a solid backdrop, checkerboard, scenery, or shadow.
1426 Do not restyle the product; remove the background and preserve clean alpha transparency.
1427PROMPT
1428cutout = client.images.edit(
1429 model: model,
1430 image: OpenAI::FilePart.new(Pathname("input_images/shampoo.webp"), content_type: "image/webp"),
1431 prompt: prompt,
1432 size: "1024x1536",
1433 quality: :medium,
1434 background: :transparent,
1435 output_format: :png
1436)
1437File.binwrite("extract-product.png", Base64.strict_decode64(Array(cutout.data).fetch(0).b64_json || raise("No PNG returned")))
1438```
1439
1440
1441
1442
1443
1444For 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).
1445
1446## Check the result
1447
1448Before you use an image, check it against your requirements.
1449
1450- Can you read every word, and does it match the requested text? Are diagram labels and relationships correct?
1451- Do identities, product shapes, labels, and reference details remain intact?
1452- Did the edit change only what you requested?
1453- If you need transparency, does the decoded file contain transparent pixels?
1454
1455Compare 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.
1456
1457
1458
1459
1460
1461
1462 <header className="not-prose mb-8">
6 <h21463 <h2
7 id="gpt-image-2.5-guide"1464 id="gpt-image-2.5-guide"
8 className="m-0 text-3xl font-semibold text-default"1465 className="m-0 text-3xl font-semibold text-default"
