AI image use cases: six workflows, with the tool order and the real cost
“AI image generation” is not a use case. Shipping a landing page, normalising a supplier catalogue and never posting a grey link preview again are use cases. Each page below is one specific job: the images it needs, the order the tools go in, the mistakes that cost a second pass, and what the whole thing costs in credits.
No case studies, no logos, no invented percentages — and where a workflow runs into something this platform genuinely cannot do, the page says so rather than routing around it quietly.
Landing pages
Copy is done, layout is done, a dozen grey rectangles are why it has not shipped.
293 credits per page
generate_imagegenerate_transparent_imagecompress_imageEcommerce catalogues
Forty suppliers sent forty different backgrounds and the product grid looks cheap.
24 credits per SKU
upscale_imageremove_backgroundcompress_imageBlog and editorial
Every article in your industry uses the same stock photo of the same sunlit desk.
16–53 credits per post
generate_imagetext_to_svgcompress_imageSocial campaigns
One idea has to become four aspect ratios, and then again next week.
44–72 credits per campaign
generate_imagemulticallgenerate_transparent_imageIcons and favicons
Default globe in the tab, twenty icons from three libraries, a manifest pointing at nothing.
14 credits per icon set
text_to_svgconvert_formatOpen Graph previews
Someone pasted your link and it rendered as a grey box with a truncated title.
16 credits per section
generate_imagecompress_imageCompared side by side
What each workflow produces, the tools in the order they actually run, and the total published on that page. Costs assume a standard model where generation is involved — a fast model is cheaper, a premium one roughly doubles the generation half.
| Workflow | What it produces | Tools, in order | Cost |
|---|---|---|---|
| Landing pages | Hero, feature art, avatars, logo placeholders and an OG card — about a dozen assets in one style. | generate_image → generate_transparent_image → compress_image | 293 credits per page |
| Ecommerce catalogues | Supplier photos normalised into one grid: enlarged, cut out, composited and encoded. | upscale_image → remove_background → compress_image | 24 credits per SKU |
| Blog and editorial | A featured image, inline illustrations and vector section marks, on one house style. | generate_image → text_to_svg → compress_image | 16–53 credits per post |
| Social campaigns | One concept rendered natively into four platform shapes, plus a transparent sticker. | generate_image → multicall → generate_transparent_image | 44–72 credits per campaign |
| Icons and favicons | A vector mark, rendered out to every pixel size a site needs. | text_to_svg → convert_format | 14 credits per icon set |
| Open Graph previews | A text-free backdrop per section; your framework renders the title over it for nothing. | generate_image → compress_image | 16 credits per section |
Four habits every one of these workflows shares
The specifics differ. The technique underneath is the same in all six, and it is worth learning once.
Approve one, reference the rest
Get a single image right, then pass it in referenceImages everywhere else. It is a real image input to the model, not a description folded into the prompt — which is why it beats repeating adjectives.
Ask for the shape you need
Six ratios are available at generation time and cost nothing extra: 1:1, 16:9, 9:16, 4:3, 3:4 and 21:9. Omit aspectRatio and you get a square, so name it — cropping afterwards throws away composition you paid for.
Batch what is independent
multicall turns twenty sequential round trips into one, billed once and refunded per failed item. There is no cap on batch size and every item fires at once, so impose your own limit and keep batches in the tens.
Compress with the destination in mind
For assets your own pages serve, compression is the page-weight lever and one credit is well spent. For assets you hand to a social platform, it is only a way under the size ceiling — lossy passes compound, so upload the cleanest file that fits.
Before you plan around it
Four things these workflows cannot do
Each of these bites on a specific page, and each page says so where it matters. Collected here so you can check them against your own plan before writing any code.
No resizing, at all
No endpoint takes a width or a height, and nothing downscales. The only route to an exact pixel size is rasterising a vector whose own dimensions you set.
No background colour
A cutout encoded to JPEG lands on black, not white. Compositing onto a specific colour has to happen in your code.
No ICO output
PNG, JPEG, WebP, GIF, TIFF and AVIF. A favicon.ico has to come from somewhere else.
No vision or OCR
These tools produce and transform image files. Reading what is in a picture is a different category of API.
Questions about these workflows
Which workflow should I start with?
The one whose problem you recognise. If you are shipping a page this week, start with landing pages; if you publish regularly, start with blog imagery, because the house-style technique there is the one that pays off most over time. Every workflow uses the same handful of tools, so whichever you start with teaches you the rest.
Do the credit figures include retries?
Mostly yes — the totals assume a realistic number of attempts on the image that matters rather than a perfect first try, and each page shows its own breakdown. The exception is that a genuinely difficult prompt can take more, and generation is priced by the model you choose, so a premium model roughly doubles the generation half of any figure here.
What order should the tools go in?
It differs per workflow, but one rule cuts across all of them: if an image needs both enlarging and a transparent background, upscale it first and cut it out afterwards. The upscaling model does not support transparency and returns a white background, so cutting out first quietly destroys the alpha channel you just paid for.
Can one agent run a whole workflow end to end?
Yes, and that is the point of the tools being MCP tools rather than only REST endpoints. Most calls return a hosted URL that the next call accepts as input, so an agent can chain them without ever holding image bytes. Two exceptions to know: compression returns the file itself as base64, and vector generation returns SVG source with a vendor URL rather than one on our CDN.
Can I try these on the free tier?
Yes — 30 credits a month with no card. That is enough to run the cheap workflows end to end and to generate a couple of images on the expensive ones, which is the right way to judge output quality before committing. Checking your balance and listing models cost nothing.
What can these tools not do?
Four things worth knowing before you plan around them: nothing resizes a raster image, nothing sets a background colour behind a cutout, ICO is not a supported output format, and there is no vision, OCR or classification here at all. Each of those is called out on the workflow page where it bites.
Reference
Every credit figure on this page is the total published on the workflow it links to, and each of those was read from the request handlers rather than from the tool schemas.
Pick the workflow closest to your job
Then run it on the free tier before you decide anything. Thirty credits a month, no card, and checking your balance or listing models costs nothing.