100 product shots on GPT Image 2.5: $1.65 from text, $2.09 as edits

One hundred 1024x1024 product shots on ChatGPT Image 2.5 cost about $1.65 at medium when generated from a text prompt and $2.09 when each one is an edit of your own product photo. At high the same hundred cost $6.59 and $8.35.

A text prompt can only invent a product, so a shot of your actual item has to be an edit with the photo as a reference. The extra input image adds about $0.0044 per call at medium.

The grid

Estimates from the repository pricing code; the input image token count is estimated.

A workable pipeline

Draft at low to pick the scene and the angle, because 100 low edits cost only $0.94. Then rerun the approved scenes at medium for the catalog and high for the hero images only.

A split of 80 medium edits and 20 high edits comes to $3.34, against $8.35 for all high.

• Always set quality; omitting it defaults to high.
• Use aspect_ratio: "auto" on edits to keep the source framing.
• Use mask_url to keep the product pixels fixed and replace only the background.

Background swap with a mask

For catalogs where the product should stay as close to the photo as possible, a masked edit is the safest route. Provide the product photo as the reference and a mask_url that marks the background, then prompt for the new surface and light. The mask tells the model which region to change, but it guides the edit rather than guaranteeing every pixel outside it, so check labels and logos on the results.

You need to produce the masks yourself. A cut-out tool or a segmentation step creates a mask once per photo, and the same mask works for several background prompts, which makes a hundred variants of twenty products cheap to run.

• Make one mask per product photo.
• Reuse the mask across background prompts.
• Host masks at public HTTPS URLs.

What the number leaves out

The table is image generation only. It does not include retries from your own review, the time to check each shot against the real product, or any video step. If a shot becomes a clip, add the video model's per-second price from the Video API docs separately. The token rates behind the table are on the Fal page.

If you sell on a marketplace with image rules, such as a white background and a minimum size, put those rules in the prompt and in your size choice before the batch starts. A 1024x1024 image may be below a marketplace minimum, and rerunning 100 images at a bigger size changes the totals above. Check the rule first, pick the size, recompute from the grid in the 4K ladder or the size grid, and only then run the hundred.

Plan for rejects. If one shot in five fails review and you rerun it, a hundred approved images means about a hundred and twenty-five generations, and the totals above rise by a quarter. Build that margin into the quote instead of discovering it on the invoice, and tighten the prompt after each batch so the reject rate falls.