| name | stock-sourcing |
| description | Use when sourcing stock images or video clips from Pexels or Pixabay APIs, including query construction, download behavior, licensing metadata, and asset-pipeline integration. |
Stock Sourcing
Use this skill for stock media tools and stock-source adapter workflows:
pexels_imagepixabay_imagepexels_videopixabay_videodirect_clip_searchcorpus_builder
direct_clip_search and corpus_builder route through the shared
tools/video/stock_sources/ adapter family. That family can include Pexels,
Pixabay, Unsplash, NASA, Archive.org, Wikimedia, NARA, Library of Congress,
Mixkit, Coverr, Videvo, NOAA, ESA, JAXA, Pond5 public domain, and Dareful,
depending on local configuration and provider availability.
Provider Setup
Pexels requires PEXELS_API_KEY. Send it as the raw Authorization header
value, not as a bearer token.
Pixabay requires PIXABAY_API_KEY. Send it as the key query parameter.
These are network/API or hybrid tools. Do not offer them as fully offline paths
unless the requested source is an already-populated local project corpus.
Query Rules
Use short concrete search phrases. Prefer nouns, setting, activity, and mood
over sentence-like prompts.
For stock video, include duration constraints when the scene length is known.
This avoids downloading clips that are much longer than the edit needs.
For visual consistency, choose provider filters from the project playbook:
- orientation or aspect ratio
- color or palette cues
- category
- media type, such as photo, illustration, vector, film, or animation
- safe search or editor selection flags
Provider Choice
Use Pexels when the project needs curated real-world photography, high-quality
B-roll, strong orientation filtering, or 4K-capable stock video.
Use Pixabay when the project needs category filtering, illustrations, vectors,
animation clips, or higher request-rate tolerance.
Use public-domain/government/archive adapters when provenance, historical
footage, scientific imagery, or license clarity matters more than commercial
stock polish.
Use direct_clip_search when a shot list already exists and the goal is fast
download plus thumbnail inspection. Use corpus_builder when the project needs
a reusable local corpus with thumbnails, CLIP embeddings, and query-time
retrieval through clip_search.
If exact composition, product identity, brand text, or novel visual structure is
required, use generated-image or generated-video providers instead of stock.
Download Rules
Always download selected assets immediately into the explicit output_path,
output_dir, or corpus_dir requested by the tool. Do not persist provider CDN
URLs as durable artifacts. Provider URLs can expire, and result ordering can
change.
Write project-scoped outputs under the path family required by the tool:
- generated or acquired media:
projects/<project-name>/assets/... or
projects/<project-name>/renders/...
- corpora:
projects/<project-name>/corpus/...
- thumbnails and sidecars: inside the selected project-scoped output directory
Metadata Rules
Preserve provider metadata in the tool result and downstream asset manifest:
- provider name
- source page URL
- author or photographer name when returned
- provider media id
- license label
- downloaded file path
Pexels and Pixabay allow broad free commercial use, but do not imply creator,
person, brand, or venue endorsement from stock footage alone.
