For PhotoCraft 0.10.0. This is a feature and interface reference based on the pinned editor bridge, catalog and billing implementation. Example requests below are suggestions, not a recorded AI-generated result or a performance benchmark. Results vary by prompt, model and document.

PhotoCraft AI has two modes, and they solve different problems. The image tools create a picture from a prompt or a reference image. The Agent edits the document you already have open — layers, text, opacity, adjustments — by executing a fixed set of document actions in your browser. This guide explains the controls and billing rules for both modes; it does not present a tested production case study.

What you need before starting

The PhotoCraft editor open in a desktop browser. No account or installation is required to edit; AI requests use a guest wallet you can recover with your purchase email.
Credits for AI tasks — one-time packs are described on the pricing page. Manual editing without AI stays free.
For reference-guided generation: your own image, or one you are authorized to use.

Mode 1: generate an image with the tools

Open the AI panel and switch to the image tab.
Describe the image, and optionally attach a reference image for models that support image-to-image editing.
Pick the model, aspect ratio and resolution. The panel shows the exact credit quote for that combination before you confirm — the model name and settings are always displayed, with no silent substitution.
Confirm. The status line shows plain-language progress (preparing → processing → saving) plus how long you have waited.
When the image is ready, download it or insert it into the open document. A generation failure releases its credit reservation automatically.

Mode 2: ask the Agent to edit your document

Open a document (PSD, PNG, JPEG, WebP) in the editor and switch the AI panel to the Agent tab.
Describe the change in one sentence: “Rename these layers to something readable”, “Make this photo a little brighter”, “Add a black & white adjustment on its own layer”, “Change the headline text to …”.
The Agent plans one round of supported document actions and hands the commands to your browser, which applies them and confirms back. You will see what was applied.
Undo the applied batch from the conversation if it is not what you wanted — the canvas stays yours.
Keep editing by hand, or send a follow-up message. Each model step the Agent takes is billable; the task pauses before exceeding the spending cap it was given.

Supported Agent actions today include creating and renaming layers, changing opacity, editing text layers, adding brightness or black & white adjustment layers, and inserting generated images. It operates the PhotoCraft editor, not Adobe Photoshop, and it does not support every possible editing request.

What a task costs

Image tasks show a fixed credit quote per model and output setting before confirmation — for example, Seedream 5.0 Flash at 1K is 50 credits and Nano Banana 2.1 at 1K is 70 credits (see the current table on the pricing page).
Agent tasks charge by actual model usage: 2 service credits + 1 credit per billable step + 3,000 × confirmed model cost in USD, rounded up once per task. A task with no billable model steps costs nothing. Reservations are released when a task settles.
Failed image generations release their reservation. Confirmed model usage stays charged even if you undo the edit afterward.

When something goes wrong

The status line says what is happening in plain words; raw state codes stay out of your way.
If the browser cannot apply a command batch, check the status and document before retrying. Task limits, confirmed usage and reservations are handled according to the rules above.
If a task ends in an unusual state, your balance and task history remain available in the panel; the 14-day refund rules for unused orders are in the Terms of Service.

Where to go next

Edit text in a PSD — the classic text-layer workflow, with and without AI.
Pricing — packs, per-model image counts and billing rules.
The AI workspace — open it and try one small task first.