"Put this kurti on a model against a festive background, then three colourways and a catalog card." The Agent works out which tools that needs, in what order, and runs them — telling you the credit cost first.
The wizard is still there. This is the faster route when you already know what you want.
In your own words, in your own language, with up to four images attached. "Same saree but walking, on a beach" is a complete instruction.
Garment type, model, background, pose, camera angle and framing are chosen for you from what it can see in your photo and read in your words.
Up to five steps a turn, and steps can feed each other — generate a hero shot, then compose a catalog card around that exact result, in one reply.
The Agent does not re-implement anything. It calls the same endpoints the wizard does.
Garments and all fourteen accessory types, with pose, body angle, framing, hand position, expression, dupatta drape and carry style all selectable — and picked for you if you do not say.
"Three more colourways" or "make the background warmer" become colour-variant and edit runs against the image you just made.
Upscale, background removal, beautify, blur background and the simple drop shadow cost nothing, and the Agent says so when it uses one.
"Make a card for this" composes one for free — but it will never invent an article number, a price or a fabric you did not state. Unstated fields stay empty.
On Business and above, the Agent can animate a shot into a product clip. It quotes the cost (12, 18 or 24 credits) and the app asks you to approve it before a single credit moves. On other plans the tool is not offered at all, rather than proposed and then refused.
Past generations from any chat or from the wizard can be used as sources. "My saree photo from yesterday" is a valid instruction — no re-upload.
An agent that spends without asking is a support ticket, not a feature.
The total credit cost of a turn is quoted in the reply, and video — by far the most expensive step — gets an explicit approve-or-cancel prompt before it runs.
It is capped at five tool runs per turn, and when you ask for several shots it makes them visibly different rather than burning credits on near-identical crops.
Missing garment image, or an instruction where a wrong guess is expensive — the dupatta drape on a bridal lehenga — and it asks one short question instead of spending.
Because it runs the ordinary endpoints, credits, plan gates, concurrency limits and the fidelity checks all apply to Agent runs exactly as they do in the wizard.
No. The orchestration itself is free — you pay only for the tools it runs, at exactly the prices they cost in the wizard. A turn that only removes a background or makes a catalog card costs nothing at all.
No. The Agent is available on every plan. The only thing that changes with your plan is which tools it can reach: video is offered only on Business and above, and pose reference only on Pro and above.
It tells you the credit cost of a turn in its reply before running anything, and it is capped at five tool runs per turn. Video, which is by far the most expensive step, always goes through an explicit approve-or-cancel prompt showing the cost before a single credit moves.
Yes. Your recent generations from any chat or from the wizard are available to it as sources, so you can say "my saree photo from yesterday" instead of re-uploading. If more than one image plausibly fits what you described, it asks which one rather than guessing.
Yes, and the steps can feed each other. Ask for a festive hero shot and a catalog card and it will generate the shot first, then compose the card around that exact result, in one reply.
No. Article numbers, prices, fabric and size details come only from what you typed — anything you did not state is left blank. It will compose a catalog name or a one-line subtitle from the garment itself, which is presentation rather than fact.
It replies in the language you write in.
Attach a garment photo, say what you want, and let it set the shoot up for you.