PhotoGPT Agent

PhotoGPT Agent helps you create through conversation. Describe the visual you want, and Agent can prepare prompts, settings, references, and connected steps in your dashboard. You review and approve generation before it starts.

Agent works alongside the page controls, so you can inspect the setup and make changes yourself.

Open Image Generation with Agent

Start a conversationCopy Anchor Link

1. Sign in to PhotoGPT and open Image Generation.
2. Select Agent beside the prompt controls. On a larger screen, the panel may already be open.
3. Describe the result you want. Include details that matter, such as the subject, setting, format, and anything you want to preserve.
4. If you have source media, select Add images or videos in the message box and explain how Agent should use it.

For example:

Use this product photo to prepare a studio image with a warm beige background. Keep the product shape and label unchanged.

Or:

Build a Flow with two image variations from this photo, then connect each image to a video step. Prepare the workflow for me to review.

What Agent can help withCopy Anchor Link

Agent can move between supported pages while continuing the conversation. Models, input roles, and available settings vary by page and model. If your request needs an unsupported combination, the setup must change before it can run.

For Retouch, select the spot you want to change in the editor. Agent can prepare the prompt, but that selection is required.

Attach media with a clear purposeCopy Anchor Link

An attachment tells Agent which image or video you mean. It does not automatically become an input to a generation or edit.

• Agent attachment: Media included in the conversation for discussion or inspection.
• Generation reference: Media added to the selected image or video model's inputs.
• Editing source: The image loaded in the editor that the next edit will change.
• Flow input: Media or text connected to a step on the canvas.

Say “use this as the first frame” when the image should be where a video starts. Say “edit this image” when it should be the editing source. Selecting or previewing a result alone does not load it as that source.

You can attach up to 16 images and 2 videos per message. Each video can be up to two minutes, and attachments must total no more than 100 MB. A video needs a ready video file for inspection. Generation models have their own input limits, separate from these chat limits.

Agent can help inspect an image or answer focused questions about a video. For example, ask which moments in a clip show the product clearly, then explain how you want to use those findings.

Review preparation and approve generationCopy Anchor Link

Agent can update the page while preparing your request. Prompt changes, model selection, references, and Flow nodes can be applied before an approval card appears. Preparation does not start media generation.

When Agent is ready to request image generation, video generation, an edit, an upscale, or a Flow run, it shows an approval card. Review the model, prompt, quantity, and credit estimate where available. Estimates may be provisional or unavailable. In Flow, text produced by an earlier step may not be available until the workflow runs.

• Select Approve to proceed with the prepared action.
• Select Deny to decline it.
• To change the request, send a correction while approval is pending, then review the revised setup.

If required inputs are missing or settings are incompatible, resolve those problems before approving. Declining generation does not undo preparation changes already applied to the page.

After approval, a message that work has started is separate from confirmation that it has finished. Check the result and its status before downloading it or using it in another step.

Understand creditsCopy Anchor Link

Agent responses consume credits based on usage. Images, videos, edits, upscaling, and Flow runs have their own generation costs. The generation estimate on an approval card does not include a fixed price for the entire conversation.

Review the current estimate before approving. If there aren't enough credits, add credits or change the request before trying again. See pricing for current plans and purchase conditions.

Stop a response or continue laterCopy Anchor Link

Select Stop response to interrupt Agent's current response and pending Agent actions. Stop does not undo changes already applied. Media generation already submitted can continue, and Agent usage incurred before Stop can still be charged.

Use New chat to start a separate conversation. Open the Sessions menu to return to a saved chat, rename it, or delete it. A saved chat records the conversation; reopening it does not restore an older version of your editing form or Flow canvas.

When returning to a task, tell Agent what you want to do next and check the current page setup. If a connection or generation fails, inspect what already succeeded before requesting another run.

Related guidesCopy Anchor Link

• PhotoGPT Flow
• Writing prompts
• Using reference images
• AI Photo Editor
• Audio inputs for video

For account or product help, contact PhotoGPT support.