PSD Reader — layers, extract & editable templates (API)

Read Adobe Photoshop .psd/.psb files without Photoshop, behind a clean HTTP API (and MCP tool). Get the full layer tree, pull text and colours, extract any layer as a PNG, flatten the composite — and turn a PSD into an editable template you can fill programmatically to auto-generate variations (think logo-vs-logo match banners, localized creatives, batch product cards). Powered by ag-psd. No Photoshop, no key. Pay per operation.

• 🗂️ Info — the whole layer tree: names, types (text / pixel / group / smart object / shape / adjustment), bounds, opacity, visibility, blend mode, and text content + font/size/colour.
• 🧩 Template — every layer becomes a field; you choose which are editable (text, image, colour). Mark them in Photoshop by naming a layer edit_title, or pass an editable list, or let the auto heuristic decide. The base for a dynamic-creative pipeline.
• 🖼️ Extract — render chosen layers (or all pixel layers) to PNG/JPEG/WebP, returned as URLs with their positions.
• 🏉 Render — flatten the PSD's composite to a PNG/JPEG/WebP (optionally resized).
• 💵 Pay per operation · no key.

The use case it was built for

Design teams keep the same PSD template and swap a few layers per output — two team crests and a score, a product photo and a price, a translated headline. Doing that at scale means reading the PSD structure, knowing which layers are the "slots", and rendering variations. This Actor is that engine as an API: Template turns a PSD into a typed list of fields (with the editable ones flagged), and Extract/Render produce the images. Wire it into an automation or an agent (over MCP) and generate hundreds of on-brand creatives from one PSD.

Marking editable layers — three ways

1. In the PSD (recommended): name the layers you want to expose edit_headline, edit_logo, edit_bg… Any layer starting with edit_ becomes an editable field, and its key is the name without the prefix. Zero config at call time.
2. At call time: pass editable: ["Headline", "Logo"] (layer names or ids).
3. Auto: leave both empty and text layers, smart objects and solid-colour layers are marked editable by default.

Sample output (template)

{
  "ok": true, "canvas": { "width": 1080, "height": 1080 },
  "editableBy": "naming convention \"edit_…\"", "fieldCount": 12, "editableCount": 3,
  "fields": [
    { "id": "2.0", "name": "edit_headline", "key": "headline", "kind": "text",
      "editable": true, "rect": { "x": 80, "y": 120, "width": 920, "height": 90 },
      "text": { "content": "GAME DAY", "font": "Poppins-Bold", "size": 72, "color": "#ffffff" } },
    { "id": "1", "name": "edit_home_logo", "key": "home_logo", "kind": "image", "editable": true,
      "rect": { "x": 120, "y": 300, "width": 300, "height": 300 } }
  ]
}

How to use

• HTTP API (Standby): POST /info, /template, /extract, /render. GET / returns help.
• Normal run: pass { operation, psd, ... }; results land in the dataset + key-value store.
• MCP tool: expose it to your agent via Apify's MCP server.

Input

• psd — the file as an http(s) URL or base64.
• operation — info · template · extract · render.
• editable / editablePrefix / includeGroups — (template) pick editable layers, the naming prefix (default edit_), and whether to emit group fields.
• layers — (extract) which layers to render (names/ids; empty = all pixel layers).
• format / width / quality — (extract/render) png/jpeg/webp, resize, compression.

Pricing

Per operation from $0.02 (down to $0.008 on higher tiers). One info, template, extract or render call = one operation.

Related Actors

• Image Converter & Optimizer — convert/resize/compress the images you extract or render.

FAQ

Does it need Photoshop? No — it parses the PSD binary directly.

Which layer data do I get? Names, types, exact positions/sizes, opacity, blend mode, visibility, text content + font/size/colour, and solid-fill colours. Extract/Render give you the actual pixels.

Can it flatten any PSD? Render uses the embedded composite preview. PSDs saved without "Maximize Compatibility" have none — use Extract per layer instead.

Notes & limits

Up to 80 MB per PSD; Extract renders up to 60 layers per call. Very exotic layer effects may not round-trip perfectly. This reads PSDs — it does not (yet) write edited PSDs back.