Find and download licensed media from 9 free sources

web-media queries many free image, video, and GIF sources in one fan-out call, returning a normalized result list and optionally downloading the top-K results with an attribution sidecar. It is a zero-dependency stdlib Python script (webmedia.py) for retrieving a real or archival photo/clip, or a reaction/animated GIF, rather than a generated one. Nine sources are wired in: openverse, wikimedia, internetarchive, and loc for historical/archival/factual images (internetarchive also covers films, no key needed for any of the four); nasa for space imagery and video (no key); pexels and pixabay for modern stock photos and short video clips (free key); and klipy and giphy for GIFs (free key, GIF sources only fire with --type gif). Each result record follows the schema {source, title, url, thumb, dl, page_url, author, license, w, h, type}, where dl is the directly-downloadable media URL and is None when only a page exists.

Use it when a task needs a real or archival photo/clip - hero image, texture, reference, historical footage - or a reaction/animated GIF, rather than a generated one. For a modern single video clip, use pexels or pixabay directly since they are born as short clips with a direct MP4. For a historical single shot, retrieve the archival film first, then extract the shot: either pass the Internet Archive public MP4 URL to Twelve Labs' Marengo search (free 600 minutes) for a timestamped semantic moment and clip it with ffmpeg, or use PySceneDetect (free, local) to cut the film into shots and rank keyframes with CLIP via the muser skill, fully offline. Results depend on third-party API availability, quotas, credentials, and license metadata quality, so license tags and attribution must still be reviewed before commercial or public use, and final aesthetic fit, brand safety, and audio suitability require human inspection.

Input is a text query plus flags: --source all|nokey|<comma list>, --type image|video|gif, --count N, --json, and --download --out DIR. --download fetches each result's direct media URL and writes an attribution.json (source, author, license, url, page_url) alongside the files, for example:

webmedia.py "1950s street scene" --type image --count 8 --json

Optional keys PEXELS_API_KEY, PIXABAY_API_KEY, KLIPY_API_KEY, and GIPHY_API_KEY live in central/.env; the five no-key sources work without any of them. The klipy adapter is unverified and assumes a Tenor-compatible request/response, so it should be checked against docs.klipy.com when keyed; the tenor adapter itself was removed after Google ended that API on 2026-06-30.

For sound effects, a sibling script freesound-fetch.py searches freesound.org and downloads short hq-mp3 previews with license/user attribution, keyed by FREESOUND_API_KEY (token-based read; full originals would need OAuth). For judging audio, since the assistant cannot hear, audio-judge.py sends a clip to OpenAI's audio-native gpt-audio and returns {heard, score, matches, suggestion}, auto-sourcing a real sk- OPENAI_API_KEY and padding sub-2s clips so the speech-tuned model doesn't refuse; it reliably describes audio and filters obvious mismatches but is not a trustworthy judge of subjective qualities, so it should cull candidates rather than make the final aesthetic call. web-media is the internet-retrieval capability, a peer to muser (local semantic search) and fal (generation); a future media router could fan out across all three and rank by CLIP relevance, but the skill notes not to build that router until the model demonstrably mis-routes without it.

Anyone assembling real or archival visual or audio assets - hero images, textures, reference photos, historical footage, reaction GIFs, or sound effects - who needs license-tagged results from free sources rather than a generative model, and who will review attribution and aesthetic fit by hand before shipping.