Skip to main content
Use manual stock-site lookup when a person is choosing a few images and can keep the source and credit with each selection. Consider an image-search API when your application repeatedly turns content briefs into candidate lists and needs to carry image metadata into a review or export step. The deciding question is what you need to repeat, not simply how many images you need. An API can automate retrieval and metadata handling. It does not eliminate the work of deciding whether an image fits the brief or whether its stated conditions fit the intended use.

​
Choose around the work you actually do

These are workflow tradeoffs, not a benchmark against named stock services. Manual sites differ in their search, download and metadata features. Evaluate the specific service you would otherwise use.

​
What Lightdrift adds, and what stays in your application

Lightdrift supports text, reference-image and image-plus-text search. The search API accepts filters such as orientation, minimum width and source-declared commercial or derivative permissions. Results include asset IDs, file URLs, dimensions, sources and a structured rights object. That gives a developer a way to retrieve candidates and preserve their metadata together. Your application still owns the selection interface, layout, review and publication steps. A reference image expresses visual direction; similarity does not verify the subject’s identity. A natural-language request for space beside a subject is a preference, not a guaranteed composition. Check actual returned dimensions and the response’s relaxed and degraded fields. A search can return fewer candidates than requested, none, or degraded ranking. File URLs redirect to time-limited signed downloads: keep the asset ID and source information, and follow download redirects without forwarding your API key.

​
Separate integration cost from search price

The public pricing endpoint reports 0.005persuccessfulsearch,or0.005 per successful search, or 5 per 1,000, checked September 26, 2026. Requesting five candidates in one call is one search. Three successful slide searches would cost $0.015 at that rate; revisions and repeated searches add calls. This is illustrative arithmetic, not a measured cost per usable image. Include implementation, review and credit-preservation work in your comparison. A low request price alone does not establish savings over manual lookup. Account credit and request limits are separate constraints. Check both before running real searches.

​
Try one brief before integrating

This request reuses the wind-turbine brief and five-candidate landscape setup from the published presentation example. It makes one real search. Use Python 3 only if you continue to the linked three-slide example; the trial below needs curl. Create an account and API key. Supply LIGHTDRIFT_API_KEY through your backend environment or secret manager. Keep it out of browser code, shared files and exported content. With available search credit, run:
The command has no automatic retry. If it fails or times out, inspect the response and account usage before repeating; a timeout can leave the search outcome uncertain. Do not treat an error response as a candidate list. For a no-search first step, open the pinned Python example and run its documented --dry-run. That prints three request bodies without an API key or network calls. Its requests retain the separate lig103_presentation_search_v1 tag; the one-call trial above uses lig118_api_vs_manual_v1.

​
REST, Images MCP and Docs MCP are different connections

For Images MCP, choose OAuth or an API key for the connection, not both. The client handles the MCP protocol: the REST JSON above is not a complete MCP request to send to /mcp.

​
Compare the result with a manual selection

Use the same real brief for both routes and record:
  1. Usable selection: Did a reviewer find a candidate that fits the subject, layout and actual dimensions? Keep query_id and the complete result, even when no candidate qualifies.
  2. Source and license review: Open rights.provenance_url; read the source declaration, license conditions and full rights object. Missing permission information stays unresolved. The commercial filter reflects source declarations, not approval of every use. See Rights answers.
  3. Credit survives export: Keep required attribution and source/license references with the selected image and in the final output where required. Hold a candidate if its required credit is missing. Render external titles and credits as text, not trusted HTML.
  4. Total work: Record searches attempted, successful calls, time to a reviewed selection, integration effort and any export repair. Separate retrieval time from review time. Do not infer time savings from one successful HTTP response.
Start with manual lookup if integration would add work without a recurring benefit. Continue with the API if the trial demonstrates a useful candidate queue and metadata that survive your workflow. If an agent already owns the drafting step, test the same review process through Images MCP. Verification: Request fields, authentication guidance and price checked against the live OpenAPI and linked product documentation on September 26, 2026. The pinned example’s dry-run was executed; this page’s request was checked against the schema. No authenticated search, retrieval-quality comparison, time saving or customer outcome is claimed.