Module 6 • Lesson 3🔁 Pipelines — Headless Batch, APIs & QA

👤 The Goal: Output Without a Human

Two ways to make ten thousand assets. On the right, a person opens, edits, checks, and exports each — a job measured in weeks and mistakes. On the left, an automated pipeline ingests a queue, processes via the API, runs a QA gate, and delivers only the passes — measured in hours and unattended. Drag the handle:

🧠 Mental Model: The Assembly Line

A pipeline is an assembly line with a quality inspector. Assets enter a queue (ingest); each is transformed by a script, action, or API call (process); a QA gate checks it against spec and routes passes forward and fails to a review pile; passes are packaged and shipped (deliver). The whole thing runs unattended, logs everything, and retries transient failures. Your craft knowledge lives in the process step; your professionalism lives in the QA gate.

The mindset shift from a batch (Intermediate Module 6) is autonomy plus verification. A batch runs your steps; a pipeline runs them at scale, checks its own work, and is honest about what failed. Silent success on a broken file is worse than a loud failure — the gate exists to make failure loud.

🌐 Headless & the Photoshop API

The desktop app needs a screen and a person; a pipeline needs neither. Adobe's Photoshop API (part of Firefly Services) runs Photoshop operations server-side over HTTP — you POST a job describing inputs, an action or edit, and an output target, and it returns the rendered result. It even runs your recorded Actions (.atn) and PSD template edits (smart-object and text replacement — the data-driven idea from Intermediate Module 6, at cloud scale).

// process step: run a recorded Action through the Photoshop API (Node 18+ sketch)
const res = await fetch("https://image.adobe.io/pie/psdService/photoshopActions", {
  method: "POST",
  headers: { Authorization: `Bearer ${token}`, "x-api-key": CLIENT_ID,
             "Content-Type": "application/json" },
  body: JSON.stringify({
    inputs:  [{ href: srcUrl, storage: "external" }],
    options: { actions: [{ href: actionUrl, storage: "external", actionName: "Web Delivery" }] },
    outputs: [{ href: dstUrl, storage: "external", type: "image/jpeg" }]
  })
});
const job = await res.json();
const statusUrl = job._links.self.href;   // poll this URL until status is "succeeded" or "failed"

Endpoint paths and payload fields change between API versions (Adobe has been introducing newer endpoints), so treat this as the shape of the call and check the current Photoshop API reference before you build. Access also takes setup: an Adobe Developer Console project with OAuth Server-to-Server credentials and a Firefly Services entitlement (typically an enterprise plan or a trial). Running the desktop app as an unattended "server" isn't a supported or licensed setup; the API is the sanctioned route.

✔️ The QA Gate

The gate is a function that returns pass or fail with a reason. It encodes the spec you'd otherwise check by eye, and each deliverable gets its own spec. A web delivery checks pixel dimensions, color mode and profile, file size, and no empty/blank output. A print delivery is a different gate: CMYK mode, the press profile, total ink under the limit (Module 5), and enough ppi at final size. Passes flow on; fails are logged with the reason for a human. Here's the shape of a web gate:

// qa-gate.js: WEB delivery spec, returns { ok, reasons[] }
function qaCheck(meta) {
  const reasons = [];
  if (Math.max(meta.width, meta.height) !== 2000) reasons.push("long edge != 2000 px");
  if (meta.colorMode !== "RGB")                   reasons.push("not RGB");
  if (meta.iccProfile !== "sRGB IEC61966-2.1")    reasons.push("profile is not sRGB");
  if (meta.bytes > 5 * 1024 * 1024)               reasons.push("file over 5 MB");
  if (meta.isBlank)                               reasons.push("output appears blank");
  return { ok: reasons.length === 0, reasons };
}
// A PRINT gate is separate: CMYK mode, press profile,
// total ink (TAC) under the limit, and ppi at final size.

for (const asset of processed) {
  const result = qaCheck(asset.meta);
  if (result.ok) deliver(asset);
  else           log.fail(asset.id, result.reasons);   // loud, not silent
}

🛠️ Guided Build: Design a Pipeline

You'll design the flow and implement the one piece that makes it trustworthy — the QA gate.

🧗 Now You: Solo Variation

🍳 Recipe Card: Pipeline

📓 Learning Journal

📝 Module 6 Summary

🎉 What You've Accomplished — Across All of Module 6

You've crossed from operating Photoshop to programming it: scripting decisions, shipping tools with a UI, and building self-checking pipelines that produce at scale. This is the highest-leverage skill in the whole course — it multiplies every other skill you have across thousands of assets, unattended.

❓ Common Questions at This Stage

🔭 Looking Ahead

You can automate the deterministic. Next you handle the probabilistic, professionally: Module 7 — AI Mastery & Provenance uses generative tools at a controlled, rights-clean, disclosed standard.

✅ Before the Next Module

• Ship one script and one panel button that do real work.
• Write and test a QA-gate function for a deliverable you make.
• Write your Learning Journal entry.

📚 Additional Resources

• Adobe: Photoshop UXP scripting
• Adobe: Photoshop API (Firefly Services)