Tenor API closed to new clients: make short looping clips instead

If you wanted to add Tenor GIF search to a new app, you cannot start now: Google's Tenor quickstart states that as of Jan 2026 it is no longer accepting new API clients, and shows a notice about the API service being discontinued. Sume does not replace a GIF search library, but if the real need is a short reaction clip of your own, Sume can cut and condition an MP4 for it.

The Tenor facts come from Quickstart | Tenor | Google for Developers, read 2026-10-02. The upload facts come from Docs | GIPHY Developers, read 2026-10-02.

What does the Tenor page actually say?

The quickstart carries the line that as of Jan 2026 Google is no longer accepting new API clients, and the remaining instructions are for reference only. A separate notice points to an update about the API's discontinuation. The page still describes the media formats: it recommends loading a smaller optimized size such as tinygif for previews and the full gif when sharing.

The page does not give a shutdown date in the content I read, so do not build a plan around a specific day. Existing clients are a different case from new ones, and the page does not say what happens to them.

Where can the clips come from now?

Three practical sources: your own footage, clips you generate, and a library you are licensed to use. GIPHY documents an upload endpoint, upload.giphy.com/v1/gifs, which accepts animated GIFs or video files up to 100 MB, with either a file or a source_image_url. That means a short MP4 is acceptable there without converting to GIF first.

Sume's role is the middle step: take a longer clip you own and turn it into a short, silent, loop-friendly MP4.

What can Sume do for a reaction clip?

Video trim cuts [start, end) from a Sume-hosted clip into a new MP4. Use duration between 0.2 and 900 seconds, audio: "drop" for a silent loop, and an output conform of width, height or fps if you want a smaller file. The public rate in the docs is $0.02 per job.

What Sume does not do: it does not return a GIF file, it does not search GIF libraries, and it does not detect loop points. If a platform strictly requires .gif, convert the MP4 in a tool you run.

curl -X POST https://api.sume.com/v1/video-trim \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: reaction-001" \
  -d '{"video_url":"https://media.sume.com/artifacts/artf_demo/clip.mp4",
       "start": 12.4, "duration": 2.5, "audio": "drop",
       "output": {"width": 480, "fps": 24}}'

What does the reaction-clip workflow look like?

Say you run a community app and want a library of five hundred two-second reactions. Start with long source videos you have the rights to. Use video inspect with transcribe: true if you need to find lines by what is said: it returns word timings and, with sentence segmentation, caption-line shaped segments, billed at $0.01 per audio minute. Then trim each hit with the settings above. Each cut is a separate job, so the bill is simple to estimate: the number of clips times the per-job rate in the catalog.

Keep a manifest: source URL, start, duration, resulting video_url, and the tags you assign. Sume stores the files but has no tagging or search of its own, so your search index, whatever it is, lives on your side. That is the real gap versus Tenor: discovery.

For serving, a short MP4 with no audio is small. Set the output conform to 480 pixels wide at 24 fps, as in the example, and most clips will be a small fraction of a megabyte, though the exact size depends on the motion in the clip and Sume does not guarantee a target size. If a destination wants a .gif, convert at the edge, at serve time, and keep the MP4 as the master.

How do I decide between MP4 and GIF?

The summary: Tenor's door is closed to new clients, and the working answer for a new project is to own the clips.

• If the destination accepts video (GIPHY upload, Telegram animations, X video), send the MP4 and skip GIF's 256-color limits.
• If it only takes .gif, keep the Sume MP4 as the master and convert last.
• Keep it short: a 2 to 3 second cut loads fast anywhere.