> ## Documentation Index
> Fetch the complete documentation index at: https://docs.lightdrift.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Stock-image search API or manual lookup: which fits your workflow?

> Decide when recurring presentation and content workflows justify an image-search API, then try one brief with source and license review.

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

| Your workflow                                                                            | Sensible starting point                              | What to check                                                                                                                                             |
| ---------------------------------------------------------------------------------------- | ---------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- |
| One presentation, with a person choosing each image                                      | Manual lookup                                        | Can you record the source, license and required credit with the selected file?                                                                            |
| A recurring deck or article builder that creates a candidate queue from each brief       | API integration                                      | Can the returned candidates and metadata move through your existing review and export steps?                                                              |
| An agent preparing image options while drafting a page                                   | Images MCP                                           | Does your client support the connection and keep search calls within the brief you approved?                                                              |
| A project that depends on a particular stock provider's catalog or licensing arrangement | Start with that provider's offering                  | Verify that provider's actual catalog, access and terms. Lightdrift is not a promise to search every stock site or replace an existing license agreement. |
| A workflow where no returned image passes review                                         | Leave the image unset and revise the brief or source | More automation does not make an unsuitable candidate usable.                                                                                             |

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](/guides/search) 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](https://api.lightdrift.ai/v1/pricing) reports **$0.005 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](/guides/plans-and-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](/guides/presentation-image-search). 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](https://lightdrift.ai/sign-up?utm_source=lightdrift_docs\&utm_medium=owned_content\&utm_campaign=lig118_api_vs_manual_v1\&utm_content=decision_trial). 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:

```bash theme={null}
: "${LIGHTDRIFT_API_KEY:?Set LIGHTDRIFT_API_KEY through your secret environment first}"
curl --fail-with-body --silent --show-error --max-time 60 \
  https://api.lightdrift.ai/v1/search \
  -H "X-API-Key: $LIGHTDRIFT_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{"query":"Wind turbines in open countryside, wide composition","k":5,"filters":{"commercial":true,"orientation":"landscape"},"experiment":"lig118_api_vs_manual_v1"}' \
  --output image-candidates.json
```

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](https://github.com/JacksonHolland/lightdrift-claude-plugin/tree/8bb34d49537f460d77970337c90b1a13cff08188/examples/presentation-image-search) 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

| Route                                                          | Authentication                                                                                                | What it does                                                                                                         |
| -------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------- |
| REST: `https://api.lightdrift.ai/v1/search`                    | Lightdrift API key in `X-API-Key`                                                                             | Your backend sends a JSON search request and receives candidates.                                                    |
| [Images MCP](/guides/images-mcp): `https://lightdrift.ai/mcp`  | Lightdrift OAuth sign-in in a compatible client, or an API key through supported private header configuration | Your agent calls image search, similar-image search and image-details tools. Successful searches use account credit. |
| [Docs MCP](/guides/docs-mcp): `https://docs.lightdrift.ai/mcp` | No account or API key                                                                                         | Reads public usage documentation; it does not search for images or spend search credit.                              |

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](/guides/rights).
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](https://api.lightdrift.ai/openapi.json) 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.