Skip to main content

Review before it runs

AI Image Editing Workspace

Describe what you want done, add your image, and approve the plan — nothing runs until you confirm.

Agent workspace

Ready

Drop it here, or press Ctrl+V

JPG · PNG · WebP · GIF · MP4 · WebM · MOV

What you attach is read, not kept — the file goes to the model and is never written to our storage. Up to 6 files at a time. Only what the agent produces gets saved.

Try:

Reviewing and revising the plan is free · no account

Plan
Revisable, as often as you like
Execution
Only after you approve
Failed run
Refunded automatically

This is where the work happens. Describe what you want done to an image and the agent shows the intended steps for confirmation. Once approved, it handles the work privately and returns the checked result.

Also here

Other ways to start

Everything below runs in this same workspace, on the same balance. You can also skip picking altogether and describe the result you want — the agent will work out which of these it needs.

Or describe something else entirely

Multi-step jobs are where the agent earns its place: “cut this out, put it on a marble counter, and give me a 2000px version for Amazon” is one run, not three. Video is not available yet — we would rather say so than show a tab that does nothing.

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.

  1. 01Describe

    Start with the outcome

    Say what should change and what must remain untouched. Use normal language, not tool names.

  2. 02Plan

    Review the intended steps

    The agent shows what it understood and what it intends to do. Revise it as many times as needed.

  3. 03Create

    Confirm one controlled run

    Only after confirmation does it execute. Internal routing and step-by-step processing stay out of the way.

  4. 04Verify

    Check more than completion

    It compares the result with your brief—including logos, text, identity, and details you asked to protect.

Capabilities

What the image agent can do

Text → Image
Describe it and get it, in the style you name.
Image → Image
Hand it a reference and a direction.
Image editing
Change one element and leave everything else alone.
Background replacement
Cut out, re-scene, and relight to match.
Product photography
A phone photo becomes a studio shot, at marketplace sizes.
Style transfer
Apply a look consistently across a set.
Image variations
Four takes on the same decision, so you can choose.
Upscaling
To print size, with the right model for photos or line art.
Multi-step workflows
Cut out, compose, relight, upscale — one instruction.

What an image agent is

An agent takes the goal, not the operation

A normal AI image tool does one fixed thing. You already know you need the background removed, so you go to the background remover. That works right up until the job needs three tools in the right order, and you have to know which three.

An AI image agent takes the outcome instead. You describe the picture you want to end up with; it works out the steps, picks a model for each one, and asks you to confirm the plan. After that it executes without exposing internal processing. When it finishes it checks its own work — not just “did it do the thing” but “did anything else change”, which is the failure people actually get burned by.

Why one model is not enough

Different models are good at different things

No single image model wins everywhere. One holds a product’s logo and label steady while the scene around it changes. Another relights a composite so the shadows agree. A third is the only one that can give you an exact output resolution, because editing models cannot.

Picking correctly per step is most of what the AI image agent is for. It also means the choice can change as models change, without you having to relearn anything.

FAQ

Questions about the agent

What is an AI image agent?
A normal AI image tool does one fixed thing: you pick "remove background", it removes the background. An agent takes the goal instead of the operation — you say what you want the picture to end up like, and it works out which steps get there, in what order, with which models. It shows that plan for confirmation, then handles the execution privately.
How is this different from just using ChatGPT or Gemini to edit an image?
Those give you one model. An agent picks between models per step, because no single model is best at everything — one is better at keeping a logo untouched, another at relighting a scene, another at getting an exact output resolution. It also checks the result against what you asked for, and redoes it if something you told it to preserve got changed.
Can I change the plan before it runs?
Yes. The agent shows its understanding and the planned steps before execution. You can revise the request repeatedly, and nothing runs until you confirm. Internal risk scoring, provider routing and execution details stay out of the interface.
What happens if it cannot do what I asked?
It says so and returns your credits. It does not quietly swap in a cheaper approach and hand you something worse — a run that fails, gets blocked by our content checks, or exceeds its budget is refunded automatically without you asking.
Can I still use the individual tools directly?
Yes. Every capability the agent uses is also its own page, free to try without an account. The agent is for when you would rather describe the outcome than pick the operations yourself.
Is there an AI video agent?
Not yet. Turning a finished image into a clip is the next thing we are building, since that is the natural step once you have an image you are happy with. We would rather say this plainly than put a video tab on the page that does not work.

Available now

Every capability also works on its own

The agent decides which of these to use and in what order. You can also just go straight to the one you want.