This page describes the first version of the package. For version 0.3, with every studio, API keys and price approval, see the new setup guide.
The yesilcam command exposes the studios to scripts. Anything you click in the app, you can run from a shell, with JSON output for the next tool in the chain. It uses your normal account and the same credit wallet, so a script and a colleague in the app never see different balances or different prices. Here is how we use it, and the handful of rules that keep automated production from surprising you.
Install and sign in once
The package is yesilcam-studios, installed globally with npm. Run yesilcam login, then yesilcam balance to confirm the wallet is visible. The token is saved to ~/.config/yesilcam/config.json with owner-only permissions; yesilcam logout removes it. Keep the token private, because it grants full access to the account, and do not commit the config file anywhere.
The commands you will actually use
yesilcam image "a lighthouse at dusk, 35mm film" --tier standard --out lighthouse.pnggenerates an image on a tier and saves it.yesilcam video --from 1234 --duration 5 --waitanimates an existing asset and waits for the result.yesilcam tryon --garment 88 --person 77 --out look.pngruns a try-on for a garment asset on a model.yesilcam assets upload file.pngputs a file into your library and returns its id.yesilcam maket analyze plan.dxfreports what a floor plan contains;yesilcam maket create "Villa"andyesilcam maket save ground.dxf --project 12 --floor 0file plans into projects.
Add --json to any of them for machine-readable output. That flag is the whole point of the CLI: it turns each studio into a step in a pipeline you own.
A nightly catalog
The loop we run for a fashion client looks like this: for each garment photo in a folder, upload it with yesilcam --json assets upload, read the asset id from the JSON, then run yesilcam --json tryon with that garment id and the model id, with --wait and an output path named after the garment file. By morning a folder of garment photos has become a folder of looks, each with its provenance recorded in the ledger under the same job ids the JSON printed. New colourways land in the folder and are processed the next night without anyone opening the app.
Storyboards from a script
Split a screenplay into scene prompts, one per line, and loop yesilcam image over them on the draft tier with numbered output files. Review the folder as a strip. Then rerun only the keepers on standard. The draft pass costs a fraction of the standard pass, and the review happens in a file browser rather than in a queue, which is how we prefer to look at thirty frames at once.
Plan checks in CI
yesilcam maket analyze plan.dxf --json returns walls, doors, windows and rooms. Fail the build if a plan has no rooms or its units were guessed. An architecture practice that exports DXFs from a template can catch a broken layer scheme before anyone drops the file in front of a client, and the check takes a second.
Rules that keep scripts honest
- Run jobs sequentially when you need their ids in order. The queue is shared with the app; a script that fires twenty jobs in parallel gets them back in whatever order they finish.
- Check the balance first.
yesilcam balance --jsonbefore a batch, and stop if it is short. A script that runs the wallet dry at 3 a.m. is a script that fails half a batch. - Use draft for anything you will look at before keeping. Automation makes it easy to spend on standard by default. Do not.
- Keep the job ids. They are how you find the asset in the library, how the ledger explains a charge, and how a refund is traced if a job failed.
- Do not parse human output. Always
--json. The human format is for eyes and can change.
What the CLI does not do
It does not bypass pricing or the wallet; every operation costs what it costs in the app, with your plan multiplier applied. It does not run Maket's 3D conversion, which happens in the browser; it analyzes and files plans. And it does not retry failed jobs for you. Failures are refunded automatically, and the JSON tells you which job failed, but the decision to rerun is yours.
Frequently asked
Can two people share one CLI token? They should not. The token grants full access to the account it came from. Each person signs in with their own account, and the wallet they spend from is their own.
Does a script see the same prices as the app? Yes. The CLI, the app, the MCP tools and the pricing page read the same live table, so there is one price per operation everywhere.
Where do the outputs go? Into your library, as they would from the app, and to the path you give with --out. Nothing generated by a script is hidden from the app.
