01
Recreate a look you found
Read the prompt out of a reference, rewrite it your way, then take the render to print size.
Image workflows · Live
5 free runs for new accountsAI Image Agent for work that takes more than one prompt.
Describe the finished image. The agent proposes the steps, protects the details that matter, waits for your confirmation, and checks the result before handing it back.
No account to review a planGoogle or email sign-in
“Keep the glass shape. Move this into a cooler editorial scene.”
Proposed plan
Ready for your review
- 01Separate the product cleanly
- 02Build the new scene and match the light
- 03Keep the shape and fine edges unchanged
Change request
Confirm and run
Result checked against the brief
- 7
- specialist image tools
- 16
- prompt languages
- 2
- sign-in options: Google or email
- 0
- credits used before approval
A clear handoff
From one image prompt to a checked result.
You approve the intent and steps; the agent handles the process and returns a checked result.
- 01Describe
Start with the outcome
Say what should change and what must remain untouched. Use normal language, not tool names.
- 02Plan
Review the intended steps
The agent shows what it understood and what it intends to do. Revise it as many times as needed.
- 03Create
Confirm one controlled run
Only after confirmation does it execute. Internal routing and step-by-step processing stay out of the way.
- 04Verify
Check more than completion
It compares the result with your brief—including logos, text, identity, and details you asked to protect.
Why an agent
One model makes an image. An agent finishes the job.
Familiar image generators are excellent at one prompt and one result. The agent is for work that needs several decisions without losing the original brief.
Single-model tool
AI Image Agent
Input
A prompt for one operation
The finished result you want
Workflow
You choose and order the tools
The agent proposes the order
Control
Run first, judge afterward
Revise the plan before anything runs
Finish line
A model returned an image
The result is checked against the brief
Verification asks both “did the edit happen?” and “did anything else change?”
Working end to end
From one rough input to a finished asset.
The output of one step becomes the input of the next—without a re-upload, a new subscription, or a lost brief.
02
Put a product on the shelf
Cut the product out of a kitchen-table photo, place it on white, and clear the marketplace zoom threshold.
03
Find out what your prompt really says
Write a prompt, render it in your generator, then read the render back. What returns is what the model acted on.
The toolkit
One agent, backed by direct tools.
Go directly when you know the operation—or let the agent combine them into one controlled workflow.
Image to Prompt
Pick the format you need, then drop your picture into the workspace and get the prompt that recreates it, written in the syntax of the model you will paste it into.
Free · 1/day without an account
Remove Background from Image
Choose what the background should become, then drop your picture into the workspace and check the cut-out against a checkerboard before you download it.
10 credits · 1/day without an account
Upscale Image
Enlarge photos or line art up to 8x. Choose the smallest scale that reaches your target, then inspect the result at full size.
10 credits · 1/day without an account
Image Prompt Generator
Write one line. Get five prompts that each move a different axis — style, lighting, camera — so you can see which direction you actually wanted.
Free · 1/day without an account
Model ecosystem
A model is a specialist, not the product.
The agent sends each image operation to the model that fits it best. These are the models in use today, including GPT Image 2.5 for images with text and higher-quality work.
- Available
Precise editing
GPT Image 2.5 Sunburst
OpenAI’s quality-focused model, in use for edits that involve text or higher-quality outcomes, and as a backup for other edits.
- Available
Fast generation
GPT Image 2.5 Flare
OpenAI’s speed-focused model, in use for new images that need text, such as journal pages and cards, and as a backup for other generations.
- Available
Generate & edit
Seedream 4
Instruction-based generation and image editing in the live routing pool.
- Available
Generate
Qwen Image
A live option for text-led image generation.
- Available
Edit a region
FLUX Fill
Used for targeted changes after the requested area is identified.
- Available
Upscale
Clarity + ESRGAN
Specialist models for photographic and line-art enlargement.
Who it is for
Built around the job, not the jargon.
For people making AI images
You found an image with exactly the mood you want and you have spent an hour guessing adjectives. Read the prompt out of it instead, in the syntax of the model you use.
Image to Prompt · Prompt Generator · Anime Generator
For people selling things
A product shoot costs more than most first orders. Cut the product out of a phone photo, put it in a studio scene, and export at the size the marketplace actually requires.
Product Photo · Remove Background · Upscale
For people fixing photographs
Scans of prints come out soft, grainy and colour-shifted. Repair them with each step as its own switch, so face repair never quietly turns someone into a stranger.
Photo Enhancer · Upscale
Everything included
AI image creation and editing, without the clutter.
The practical details that turn a good demo into a tool you can use every day.
Reading images
- Six output formats
- General, Midjourney, Stable Diffusion, Flux, structured JSON and Gemini edit instructions.
- Sixteen prompt languages
- The image is analysed the same way; only the output language changes.
- Editable structure
- Change the lighting field and the prompt rebuilds — no second run, no extra wait.
- Paste, drag or URL
- Ctrl+V works, because you usually just copied the image from somewhere else.
Making images
- Five variations per idea
- Each moves a different axis — style, lighting, camera — so you can tell what changed.
- Negative prompts included
- Most generators give you the positive half and leave you to discover the rest.
- Tag syntax for anime
- Anime models are trained on tagged data and respond to tags far better than sentences.
- Twenty-four product scenes
- Marble, wood, studio sweep, outdoor, lifestyle — plus your own scene description.
Fixing images
- Two upscale models
- A photo upscaler smears line art; an anime upscaler plasticises skin. Pick the right one.
- Checkerboard cut-out preview
- Every background remover looks good on white. The checkerboard is where the truth is.
- Every repair step optional
- Denoise without sharpening, fix colour without touching faces.
- Old-photo restoration mode
- Tuned for film grain and fading, not for sensor noise and compression.
The account
- Free daily runs, no card
- Several tools work with no account at all. Nothing asks for payment details to try.
- One pool of credits
- Credits work across every tool. Failed and blocked runs are refunded automatically.
- Results kept until you delete them
- Signed-in results stay until you delete them; anonymous results are removed after seven days. Upload bytes are held only for the request.
- Batch for members
- Queue several images at once instead of babysitting them one at a time.
FAQ
Your questions, answered
Yes. GPT Image 2.5 has been live in the AI Image Agent since September 13, 2026. Flare creates new images and Sunburst handles edits. The agent picks them automatically when an image needs text in it or the outcome costs 30 credits or more, and uses them as a backup for other image work, so there is nothing to select and the credits stay the same. Other image steps, such as background removal and upscaling, keep using their specialist models.
Your next image
Describe the result. See the plan. Stay in control.
See what the AI image agent plans to do before you decide anything — no account, no card, and no credits spent before you approve the plan.