GPT Image 2.5

Sign in

Create, review and download

GPT Image 2.5 Generator and Editor

Describe what to change and what to preserve. Create and edit with Flare, or generate from text with Sunburst, from 3 credits per image.

Loading image workspace…
GPT Image 2.5 Flare: supported inputs and settings

text to image

  • Aspect ratio: auto, 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9, 27:16, 16:27, 9:8, 8:9
  • Resolution: 1K, 2K, 4K

image to image

  • Reference images (required): Up to 16 images. 30MB per image; Files are uploaded privately.
  • Aspect ratio: auto, 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9, 27:16, 16:27, 9:8, 8:9
  • Resolution: 1K, 2K, 4K
Input
Start with text or up to 16 reference images. Upload JPEG, PNG, or WebP files up to 30 MB each, including an exported sketch. Assign each reference a role, then describe the details to change and preserve. Sign in to generate.
Output
Review the finished image, download it, or use it as a reference for your next edit. Flare generation and reference editing, and Sunburst text generation, cost 3 credits at 1K, 10 at 2K, or 16 at 4K. Sunburst reference editing is temporarily unavailable. Each new edit is a separately confirmed generation.

GPT Image 2.5 examples

Review the input type and core settings before starting your own version.

Real GPT Image 2.5 outputs recorded on VibeAha. Each card carries the prompt and settings that produced it, and the two mug cards can be read as a before-and-after reference edit.

Prepare the input

Start with a written prompt. To change one part of an existing picture, add a reference image and describe both what to change and what to keep.

What to compare

Compare the subject, materials, lettering and lighting against the original. For an edit, inspect the areas you asked to preserve: a single pass can still drift on fine detail.

Example boundaries

These are recorded runs at 1K with automatic aspect ratio. Reusing a prompt does not guarantee the same result, and 2K or 4K output may differ.

Source review: 2026-09-12. Preview dimensions are measured from the original files.

Nine-character portrait grid

Model example

Nine-character portrait grid

GPT Image 2.5 Flare text-to-image output recorded on VibeAha.

1024 × 1536

Prompt and settings

Prompt submitted

A 3x3 character portrait grid on a light gray textured background. Each portrait sits in a white rounded-rectangle frame with a text label below it. Art style: 2D flat illustration, clean line art, comic style, profile view facing right, pale skin with blush. Characters: Jon Snow (black hair half-up, beard, dark fur cloak); Daenerys Targaryen (silver braided hair, blue dress); Tyrion Lannister (curly brown hair, beard, dark tunic with gold pin); Cersei Lannister (blonde braided hair, ornate red and gold dress); Ned Stark (brown hair half-up, beard, dark fur cloak); Arya Stark (short dark hair half-up, brown tunic, sword hilt); Jaime Lannister (short blonde hair, gold armor with lion motif); Sansa Stark (red braided hair, blue dress with fur collar); Theon Greyjoy (short dark curly hair, dark tunic with kraken pin).

Submitted in the VibeAha workspace. The image shown is the returned output, unchanged.

Published example settings

  • GPT Image 2.5 Flare
  • Text to image
  • 1K
  • Automatic aspect ratio
  • 3 credits
Published source record

Matte red mug with the word VIBE on a warm grey studio background

Before · Model example

Matte red mug in a studio setting

GPT Image 2.5 Flare text-to-image output recorded on VibeAha.

1024 × 1024

Compare the after frame
Prompt and settings

Prompt submitted

Create a photorealistic studio product photograph of one matte red ceramic mug, with a rounded handle on the right and the word VIBE in small white uppercase letters centered on its front. The mug sits on a pale warm-gray tabletop against a seamless pale warm-gray background. Soft light comes from the upper left, with one natural shadow to the lower right. Three-quarter front view, eye-level camera, centered composition, entire mug and handle visible. No other objects, no steam, no extra text. Square product photograph.

Submitted in the VibeAha workspace during the 2026-09-09 acceptance run. The image shown is the returned output, unchanged.

Published example settings

  • GPT Image 2.5 Flare
  • Text to image
  • 1K
  • Automatic aspect ratio
  • 3 credits
Published source record

The same red mug with the background replaced by a soft pastel blue

After · Model example

The same mug with a pastel blue background

GPT Image 2.5 Flare reference edit of the card beside it, recorded on VibeAha.

1024 × 1024

Compare the before frame
Prompt and settings

Prompt submitted

Edit only the background color I specify: change the seamless pale warm-gray background to a soft pastel blue. Keep the surrounding anatomy, material, lighting and angle untouched. Nothing else in the frame should change. Preserve the red mug, its exact silhouette, rounded handle, white VIBE lettering, the pale warm-gray tabletop and the existing shadow. Only the background behind the mug becomes pastel blue.

Submitted in the VibeAha workspace during the 2026-09-09 acceptance run, with the first mug as the reference image. The image shown is the returned output, unchanged.

Published example settings

  • GPT Image 2.5 Flare
  • Reference edit
  • 1K
  • Automatic aspect ratio
  • 3 credits
Published source record

Example 01

One product, three campaign images

Sequence: original product photo, lifestyle scene, then promotional version. Review the label, silhouette, and material at every step.

Input:
Your own product photo
Settings:
Suggested: Flare, reference editing, 2K, 10 credits per image.
View original prompt

Keep the product shape, label, and material from my reference. Place it on a pale stone shelf in soft daylight. Leave space on the right for campaign copy. Do not add text or change the packaging.

Example 02

Change a portrait background

Prepare an authorized portrait with a clear face. Change the background first, then request a clothing edit separately. Compare facial features and hair boundaries with the original; identity preservation is a review task.

Input:
A portrait with permission to edit
Settings:
Suggested: Flare, reference editing, 2K, 10 credits.
View original prompt

Preserve this person’s facial features, expression, and hairstyle. Replace only the background with a softly lit studio. Keep the original framing and natural skin texture. Do not retouch the face.

Example 03

A poster with exact event text

Use a fictional event brief to check layout and lettering. The case will include a transcription of the output beside the requested title, date, and price. No text-accuracy result is claimed before that check.

Input:
Text prompt
Settings:
Suggested: Flare, text to image, 3:4, 2K, 10 credits.
View original prompt

Create a clean event poster with this exact text: “STUDIO OPEN DAY”, “18 OCTOBER”, “ENTRY $12”. Use a clear headline, generous spacing, and a simple two-color layout. Add no other words, logos, or dates.

Example 04

Background, then lighting

Start from an owned still-life photograph. Replace the background, then use that completed output for a lighting change. Compare object edges and surface detail across all three frames to identify accumulated drift.

Input:
Your own still-life photo
Settings:
Suggested: Flare, 2K, two separately confirmed edits, 10 credits each.
View original prompt

Keep the objects, their positions, and the camera angle unchanged. Replace the background with a plain warm-gray wall. Preserve the original lighting for this edit. Do not add props or alter the object surfaces.

15 prompts for controlled image editing

Choose a task, replace the bracketed details, then copy the prompt or load it into your draft. These templates express editing intent; unchanged details still need review. Each generation or edit requires a separate confirmation at the displayed credit price.

01Assign each reference a role

Separate identity, outfit, setting, and framing.

Upload the four references in the same order as their roles.

Image 1 = subject identity. Image 2 = outfit. Image 3 = setting. Image 4 = framing. Use each source only for its assigned role and merge them into one realistic scene.

02Establish a fixed base image

Name what stays fixed before starting a series.

Lock this image as the base version. Every attribute stays fixed unless I name it directly. On future edits, touch only the variable I mention, leave everything else exactly as is.

03Edit one local element

Identify a single change and its surrounding constraints.

Edit only the [ELEMENT] I specify. Keep the surrounding anatomy, material, lighting and angle untouched. Nothing else in the frame should change.

04Place a product in a scene

Preserve product details while changing the presentation.

Treat the uploaded product shot as fixed, exact shape, materials, color and branding. Drop it into [SCENE], only adjust the backdrop, light and camera angle.

05Prepare controlled A/B variations

Change one test variable per separately confirmed task.

Batch generation is not included. Run and review each variation separately.

Produce [N] versions of this creative. Keep product, layout, type and lighting identical across all of them. Vary only [ONE VARIABLE].

06Rebuild for a new aspect ratio

Describe what to extend instead of simply cropping.

Choose the matching output aspect ratio in Settings.

Rebuild this image for [ASPECT RATIO] without cropping. Keep the subject and visual hierarchy, extend and rearrange the rest of the frame intelligently.

07Stage a product interface

Place an interface inside a device or promotional scene.

Keep the uploaded interface exactly as designed, no redesign. Place it inside a premium [DEVICE/SETTING] scene with realistic reflections, lighting and negative space.

08Apply artwork to merchandise

Specify print placement, folds, and material texture.

Take the artwork in image 1 exactly as is, same composition and colors. Apply it onto [PRODUCT] the way print actually behaves on that material, following its folds and texture.

09Continue a storyboard

Carry the subject and setting into a different shot.

Use this as the fixed reference for character, outfit and location. Generate the next shot: [SHOT DESCRIPTION]. Only camera angle, framing and pose should change.

10Prepare a modeling reference

Request a clear view and neutral reference pose.

An image reference does not establish accurate 3D geometry.

Use image 1 as the character reference. Generate a [VIEW] full body A pose version. Keep silhouette, proportions, materials and details matched, neutral studio light.

11Replace one garment

Specify the clothing change while preserving the portrait.

Upload the portrait first and the garment reference second.

Keep identity, pose, camera, lighting and background untouched. Replace only [GARMENT], using image 2 as reference for cut, material and color.

12Change one interior detail

Keep the architectural reference while varying one element.

Keep architecture, dimensions, perspective, doors and windows exactly as shown. Change only [ELEMENT] to [NEW VERSION].

13Localize the visible text

Replace campaign wording and proofread the new version.

Keep product, layout and lighting exactly as is. Replace only the visible text with '[NEW TEXT]', adjust background details minimally for [MARKET]. No extra text added.

14Turn a sketch into an infographic

Specify the nodes and relationships that must remain.

Turn this rough sketch into a clean [DIAGRAM/INFOGRAPHIC]. Keep every node, connection and label exactly as drawn. Only refine spacing, alignment and hierarchy.

15Clean up a product background

Isolate the product and inspect the edges.

Transparent output is not enabled. Use a plain background until verified.

Keep the product exactly as is. Remove the background, isolate it on [transparent/plain background]. Only clean up edge artifacts, no changes to color, shape or branding.

How to use GPT Image 2.5

  1. 01

    Prepare your input

    Separate your brief into what changes and what stays fixed. Name the subject, composition, lighting, and details to preserve. Then select an owned or authorized reference image. You can prepare this draft before signing in. Model selection, prompt, and references are kept in this browser tab for up to 30 minutes without a save.

  2. 02

    Review the settings and cost

    Choose Flare for generation or editing, or Sunburst for text generation. Open Settings for aspect ratio and resolution, and review the resolution-based quote before confirming. Sign-in returns to this page with the saved input when browser storage is available. Switching models or applying a prompt does not start a task.

  3. 03

    Inspect, refine, and download

    The workspace keeps a completed result beside the input. Check text, faces, edges, and product details before downloading. To continue editing, choose the result as a reference and write a new instruction; each edit requires another explicit submission. Completed images are saved to your private History, with download and task status.

GPT Image 2.5 use cases

Change the scene, preserve the product

OpenAI reports more natural lighting and textures, with targeted edits that preserve surrounding composition. Use a product photo as the fixed reference, then change the setting or light. This suits campaign backgrounds, merchandise mockups, and product cutouts. Check the label, color, shape, and material against the original before approving each variation.

Keep the subject recognizable

OpenAI highlights improved reference fidelity for recognizable people and pets. Name one local change, such as replacing a garment, and specify which features to preserve: face, markings, hair, or pose. The same discipline helps with character references and storyboard continuity. Use an authorized image and compare every output with the original.

Control text, layout, and visual style

OpenAI describes better handling of complex layouts and visual styles. For a poster, quote the title, date, and price and define their hierarchy. For localization, specify only the text to replace. For diagrams, preserve nodes and connections while refining spacing. Proofread the output; exact wording in a prompt is not proof of accurate rendering.

Build on each edit without starting over

OpenAI reports improved consistency over successive edits. Establish an approved base image, then change one variable per task: background, lighting, framing, or an interior element. This makes A/B variations easier to review. Compare each image with the base as well as the previous output, and return to that checkpoint if details drift.

Input requirements and limitations

Prepare your input

Supported inputs and output settings

Upload up to 16 JPEG, PNG, or WebP references, each up to 30 MB. Choose 1K, 2K, or 4K and one of the aspect ratios in Settings. Downloads preserve the returned image file. Transparent output and format selection are not currently offered.

A quote before a submission

The current VibeAha price is 3 credits at 1K, 10 at 2K, or 16 at 4K per generation. Flare supports reference edits; Sunburst currently supports text generation only. GPT Image 2 welcome credits and other model-specific benefits do not apply. Signing in, switching models, or applying a prompt does not spend credits. Review the quote before every submission.

Know the limits

Inspect the image before publishing

Text may need proofreading; fine textures may contain noise; faces, labels, and object geometry may drift. A complex request can change details you wanted to preserve. Simplify the instruction, keep an original checkpoint, and compare at full size instead of judging only the thumbnail.

Handle uncertain tasks carefully

Check the existing task in History after a network interruption before starting another. Retry that request to recover its status; repeated submissions of the same request do not create another charge. Failed tasks use the existing refund process. Track the task until its output or refund is confirmed.

GPT Image 2.5 vs alternatives

OptionBest forKey difference
GPT Image 2.5 FlareA speed-oriented starting pointOpenAI recommends Flare for most applications and reports 50% lower latency than GPT Image 2. Choose Flare here for text generation or reference editing from 3 credits per task. That comparison is an OpenAI claim, not a measured speed guarantee for this site.
GPT Image 2.5 SunburstMore demanding precision workOpenAI positions Sunburst for greater precision in polished product imagery and campaign work, with longer generation times. Sunburst text generation is available here from 3 credits per task. Reference editing is temporarily unavailable; use Flare for edits.
GPT Image 2Continuing an existing GPT Image 2 workflowUse the existing GPT Image 2 page for its available settings and credit quote. Its availability, settings, and generation history remain separate from GPT Image 2.5.

Frequently asked questions

Official sources

Model capabilities were verified on 2026-09-09. Available settings and credits are checked again when you submit.

Related AI tools

Generate with GPT Image 2

Continue an existing workflow with GPT Image 2 text generation and reference editing. Review its own settings and current quote.

Open tool →

Explore Grok Imagine Image 2

Choose this separate tool when mapping and editing image regions better matches the change you want to make.

Open tool →

Enlarge an image with Image Upscaler

Use an existing finished picture as input when your next task is increasing its resolution rather than generating a new composition.

Open tool →