GPT Image 2.5 Official Prompting Handbook 1

What this handbook covers

OpenAI's official image prompting guide splits GPT Image 2.5 into two layers. One layer is the model and the parameters you set; the other is how you write the prompt. They are deliberately separate: parameters live in the request, and the prompt only describes the picture. Most unstable results trace back to picking the wrong tier, or to writing a parameter into the prompt as if it were scene description.

Part one follows the generate-from-scratch path only. Confirm the difference between the two tiers, confirm the parameter limits, then work through nine official scenarios. Image editing, multi-turn refinement and the remaining workflows are in part two.

• Every official example states its generation settings. Copy the settings along with the prompt, or the comparison you are looking at stops meaning anything.
• All prompts below are the official text verbatim. Paste them into the workspace and replace the subject with your own.
• Each scenario ships official output from both tiers, so the difference between Flare and Sunburst under one identical prompt is the material you choose from.

Two tiers: choosing between Flare and Sunburst

GPT Image 2.5 is not one model. GPT Image 2.5 Flare is the small model, optimized for speed, with image quality the guide describes as 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 improve on precise editing and subject preservation.

The fork is not "which one is better" but "does this workflow have a quality floor". The guide gives a routing table based on where you are starting from.

• The guide insists on measuring response time and quality on your own workload. Results depend on prompts, reference images, output dimensions and quality settings, so a speed improvement on one workload does not establish a fixed improvement on another.
• Reduce latency only after quality passes. Doing it the other way round gives you a fast default you cannot ship.

Parameters and custom sizes: where they go, where they stop

Parameters go in the request, not in the prompt. The guide separates them explicitly, and that separation decides what a prompt should look like: describe the picture, and never smuggle "4K" or "transparent background" into the sentence as if it were scene content.

The guide also fixes the order for tuning quality: choose the model from the table above first, then touch quality. For a first two-tier comparison, keep an explicitly selected quality setting unchanged along with the prompt, reference images and output dimensions, because the same quality label does not imply the same image quality or response time across models.

If the output falls short, test a higher setting. Once it meets requirements, test lower settings to see whether they preserve acceptable quality while reducing latency. Use xhigh or max only when they fix an unmet quality requirement inside your latency budget; the guide states plainly that a higher setting does not guarantee a better result for every prompt.

Four hard constraints on a custom size: each edge no more than 3,840 px, both edges multiples of 16 px, a long-to-short edge ratio no greater than 3:1, and a total pixel count between 655,360 and 8,294,400.

• Transparent assets need both halves: a prompt that isolates the subject, and background="transparent" in the request, writing PNG or WebP.
• Check the decoded alpha channel once the file comes back, especially hair, glass, shadows and object edges.
• output_compression applies only to JPEG and WebP output, never to PNG.
• The guide marks outputs above 3,686,400 total pixels (2560x1440) as experimental, so keep them out of a production path.

Migrating an existing workflow: the official six steps

If you already run GPT Image 2, or the older gpt-image-1 and gpt-image-1.5, the guide does not recommend swapping the model name and moving on. It prescribes six steps, all built on one rule: change one variable at a time, and save a baseline before you talk about speed.

1. Save a baseline. Collect representative production prompts and reference images, including difficult edits, exact text, faces, product geometry and transparent assets, then record the current model, request settings and results.
2. Choose the first candidate. If GPT Image 2 already meets your quality requirements, start with Flare and test for a latency improvement. If it falls short on a complex use case, start with Sunburst and first establish that it meets your requirements. Keep the prompt, references, dimensions and output format unchanged for the first comparison.
3. Check the complete result. Compare instruction following, identity and product preservation, text accuracy, unwanted changes and transparency. Repeat requests to measure consistency, and for editing workflows test the complete sequence of edits as well as individual steps.
4. Test for a latency gain after quality passes. If you started with Sunburst, evaluate Flare against the same requirements and switch only if quality remains acceptable and latency improves.
5. 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, and confirm current pricing rather than assuming the faster model costs less.
6. Roll out by workflow. Once the released model passes your acceptance criteria, move a small share of traffic, monitor the same measures, then expand gradually, keeping the previous model available for rollback while it remains supported.

• Migrating from gpt-image-1 or gpt-image-1.5 means checking parameter differences and shutdown dates on the reference tabs rather than copying older settings unchanged. gpt-image-1 is scheduled to shut down on October 23, 2026, and gpt-image-1.5 on December 1, 2026, so validate existing workflows with gpt-image-2 before migrating.
• Repeated edits can still change details you intended to preserve. Restate those constraints and inspect each result; where a region must stay pixel-identical, the guide recommends compositing the approved edit into the original image instead of relying on prompting alone.

The eight prompting fundamentals

The guide compresses prompting into eight principles. They are not syntax rules; they are a checklist of what has to be written down. For complex requests it recommends organising the prompt as scene, subject, details and constraints, using labelled sections.

• Define the result. Name the subject and the intended use, such as a product photograph, advertisement or diagram. Specify composition, aspect ratio and placement constraints; for complex requests organise the prompt as scene, subject, details and constraints.
• Choose a maintainable format. Short prompts, descriptive paragraphs, JSON-like structures, instructions and tags express the same intent. Pick the one that makes requirements easiest to read and update rather than relying on special syntax.
• Describe visible details. Name materials, lighting, colours 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 physical simulation; for wide, cinematic, low-light, rainy or neon scenes, specify scale, atmosphere and colour instead of relying on mood words.
• 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 pose and action unambiguous.
• Specify exact text. Put required wording in quotes, 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.
• Separate changes from constraints. For edits, say "change only X" and list the details to preserve: identity, geometry, layout, lighting, labels. State exclusions too, such as unwanted text, logos or watermarks. For precise local edits, also identify the saturation, contrast, arrows, camera angle and surrounding objects that must remain unchanged.
• 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.
• 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, and compare results before adding more instructions.

Nine text-to-image scenarios with the official prompts

The guide demonstrates image generation through nine scenarios, each one showing a different technique with a complete prompt. Below, the nine official prompts verbatim, followed by the official output from both tiers. Each prompt can be edited to your own subject.

Scenario one: control style and lighting

This prompt demonstrates a tone constraint across a whole image: subject and surface detail first, then framing and lens, then light, depth of field and grain, closing with an explicit exclusion of glamorisation and heavy retouching. It is the pattern for documentary realism rather than studio polish.

Official generation settings: size="1024x1536", quality="medium".

Create a photorealistic candid photograph of an elderly sailor standing on a small fishing boat.
He has weathered skin with visible wrinkles, pores, and sun texture, and a few faded traditional sailor tattoos on his arms.
He 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.
Soft coastal daylight, shallow depth of field, subtle film grain, natural color balance.
The image should feel honest and unposed, with real skin texture, worn materials, and everyday detail. No glamorization, no heavy retouching.

Scenario two: explain a process visually

An infographic is an information checklist, not an adjective exercise. This prompt names three things only: which process to draw, in which order, and what the reader should get out of it, then lists the stages by name. The guide adds that for diagrams you verify labels and factual relationships as well as appearance.

Official generation settings: size="1024x1536", quality="medium".

Create a detailed Infographic of the functioning and flow of an automatic coffee machine like a Jura.
From bean basket, to grinding, to scale, water tank, boiler, etc.
I'd like to understand technically and visually the flow.

Scenario three: render exact text

When wording has to appear in the image, three things need stating at once: the exact copy, how many times it appears, and how it is set. This prompt quotes the tagline, requires it to appear exactly once and legibly, and closes by excluding extra text, watermarks and unrelated logos.

The fundamentals add one more lever: spell unusual words or brand names letter by letter when needed, and compare medium against high quality for small text, dense information or multiple fonts.

Official generation settings: size="1024x1536", quality="medium".

Give me a cool in culture ad / fashion shot for a brand called Thread.
It's a hip young street brand. The ad shows a group of friends hanging out together with the tagline "Yours to Create."
Make it feel like a polished campaign image for a youth streetwear audience: stylish, contemporary, energetic, and tasteful.
Use clean composition, strong color direction, natural poses, and premium fashion photography cues.
Render the tagline exactly once, clearly and legibly, integrated into the ad layout.
No extra text, no watermarks, no unrelated logos.

Scenario four: design a reusable logo

A logo's test is not whether it looks good but whether it survives being scaled down. The official prompt works in this order: describe the brand's feel, prescribe the construction technique (vector-like shapes, strong silhouette, balanced negative space), write "favor simplicity" as an explicit requirement, then handle the transparent background and clean alpha edges.

This scenario also uses the n parameter to request several variations in one call. The official example compares four variations across both tiers.

Official generation settings: size="1024x1536", quality="medium", background="transparent", output_format="png", n=1 (the runnable example uses n=4 for four variations in one request).

Create an original, non-infringing logo for a company called Field & Flour, a local bakery.
The logo should feel warm, simple, and timeless. Use clean, vector-like shapes, a strong silhouette, and balanced negative space.
Favor simplicity over detail so it reads clearly at small and large sizes. Flat design, minimal strokes, no gradients unless essential.
Fully transparent background. Deliver a single centered logo with generous padding, clean alpha edges, and no solid backdrop, scenery, checkerboard, or watermark.

Scenario five: use historical and real-world context

For photorealistic historical or real-world scenes, the guide recommends naming the place and the date and letting the model infer period clothing, staging and environment. The prompt is short, but it leans on the model's world knowledge, so clothing, props and surroundings have to be checked for historical accuracy afterwards.

The official example is two lines: one sets the scene, one constrains realism and period accuracy.

Official generation settings: size="1024x1536", quality="medium".

Create a realistic outdoor crowd scene in Bethel, New York on August 16, 1969.
Photorealistic, period-accurate clothing, staging, and environment.

Scenario six: turn a story into a comic strip

For story-to-comic generation the guide defines the narrative as a sequence of clear visual beats, one per panel, keeping descriptions concrete and action-focused so the model can translate them into readable, well-paced panels. Write the door clicking shut and the pet turning towards the empty house; do not write that the mood is cosy.

Official generation settings: size="1024x1536", quality="medium".

Create a short vertical comic-style reel with 4 panels.
Panel 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.
Panel 2: The door clicks shut. Silence breaks. The pet slowly turns toward the empty house, posture shifting, eyes sharp with possibility.
Panel 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.
Panel 4: The door opens. The pet is seated perfectly by the entrance, alert and composed, as if nothing happened.

Scenario seven: create an interface preview

Interface mockups fail in one specific way: they come back looking like concept art instead of a shipped product. The guide's remedy is to describe the product as if it already exists, focus on layout, hierarchy, spacing and real interface elements, and avoid concept-art language altogether.

Official generation settings: size="1024x1536", quality="medium".

Create a realistic mobile app UI mockup for a local farmers market.
Show 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.
Design it to be practical, and easy to use. White background, subtle natural accent colors, clear typography, and minimal decoration.
It should look like a real, well-designed, beautiful app for a small local market.
Place the UI mockup in an iPhone frame.

Scenario eight: create scientific and educational visuals

Biology, chemistry, classroom explanations, flat scientific icon systems, diagrams and learning assets all sit here. The guide recommends prompting them like an instructional design brief: define the audience, lesson objective, visual format, required labels and scientific constraints, and ask for a clean, flat visual system with a consistent icon style, clear arrows, readable labels and enough white space to scan.

When accuracy matters, list the required components explicitly and state what should not be included. For dense labels, diagrams or assets destined for slides and course materials, the guide recommends quality="high".

Official generation settings: size="1536x1024", quality="high".

Create a simple biology diagram titled "Cellular Respiration at a Glance" for high school students.

Show how glucose turns into energy inside a cell. Include glycolysis, the Krebs cycle, and the electron transport chain.
Use arrows to connect the steps, and label the main molecules: glucose, pyruvate, ATP, NADH, FADH2, CO2, O2, and H2O.
Make it look like a clean classroom handout or slide, with a white background, simple icons, clear labels, and easy-to-read text.

Avoid tiny text, extra decoration, or anything that makes the diagram hard to understand.

Scenario nine: build slides, diagrams, and charts

For productivity visuals the guide writes the prompt like an artifact spec rather than an illustration request: name the exact deliverable (slide, workflow diagram, chart, page image), define the canvas and hierarchy, put the real text or data into the prompt, then describe the visual language. The constraints stay practical too: readable typography, polished spacing, no decorative clutter, no generic stock-photo treatment.

The official example deliberately flags its market figures as fictional design inputs. Replace them with verified data before the slide goes anywhere real.

Official generation settings: size="1536x864", quality="high".

Create one pitch-deck slide titled **"Market Opportunity"** that feels like a real Series A fundraising slide from a YC-backed startup.

Use a clean white background, modern sans-serif typography like Inter, and a crisp, minimal layout. The slide should include:

* A TAM/SAM/SOM concentric-circle diagram in muted blues and grays
* Specific, believable market sizing numbers:

  * **TAM:** $42B
  * **SAM:** $8.7B
  * **SOM:** $340M
* A clean bar chart below showing market growth from **2021 to 2026**, with a subtle upward trend
* Small footnotes: **"AGI Research, 2024"** and **"Internal analysis"**
* A company logo placeholder in the bottom-right corner

The 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.

Avoid clip art, stock photography, gradients, shadows, decorative elements, or anything that feels generic or overdesigned.

Check the result before you ship it

The guide closes with a checklist to run before an image leaves your hands. Its logic is facts first, appearance second.

• Is required text accurate and legible? Are diagram labels and relationships correct?
• Do identities, product shapes, labels and reference details remain intact?
• Did the edit change only what you requested?
• If transparency is required, does the file contain an alpha channel rather than a painted background?
• When changing prompts or models, compare quality, latency and cost on representative inputs.