YouTube Shorts 16:9 thumbnail swapped for 4:5: make a safe one

If you upload a 16:9 custom thumbnail to a vertical video, YouTube replaces it with an auto-generated 4:5 thumbnail on the home, explore and subscription pages. The fix is to design the Shorts thumbnail for 9:16 and keep everything that matters inside the middle of the frame, so a 4:5 crop cannot cut it off. With Sume you can preview that crop before you upload, using a video filter pass and a frame grab.

The thumbnail rules below come from YouTube's help page Add custom thumbnails on YouTube, read 2026-10-02. The Sume steps use Video filter and Video frames.

What does YouTube say about vertical thumbnails?

The help page lists 2160 x 3840 pixels (9:16) as the recommended Shorts thumbnail size, with a minimum height of 640 pixels for Shorts. Thumbnails must be image formats such as JPG or PNG. The size limits are 2 MB on mobile and 50 MB on desktop for video, Shorts and podcast thumbnails.

The sentence that matters here is that vertical videos with 16:9 custom thumbnails are replaced by an auto-generated 4:5 thumbnail on the home, explore and subscription pages, while the custom thumbnail stays visible on watch feeds and other platforms. The page does not say how the auto 4:5 is chosen, so do not assume it is a centered crop of your image.

How much of a 9:16 frame survives a 4:5 crop?

Take a 1080 x 1920 frame. A 4:5 window at full width is 1080 x 1350, so 570 pixels of height are lost. Split evenly, that is 285 pixels, about 14.8 percent, from the top and the same from the bottom. Anything you put in the top or bottom sixth of the frame is at risk in the feeds where YouTube applies its own 4:5.

This is a safety rule of thumb, not a statement about YouTube's algorithm. Keep the face, the product and any text between roughly 15 and 85 percent of the frame height, and the 4:5 version and the 9:16 version will both read.

How do I preview the 4:5 crop with Sume?

Sume's video filter crops with fractions of the frame, and video frames pulls a still at the source size. Crop the Short to the centered 4:5 window first, then grab a frame from the cropped clip. For a 1080 x 1920 source the window is x 0, y 0.1484, width 1, height 0.7031 (1350 / 1920 = 0.703125, 285 / 1920 = 0.1484). The crop costs one encode job at the public rate in the docs ($0.02 per job), and the check endpoint is free.

Video filter reads a Sume-hosted clip (import first with POST /v1/media-imports if it lives elsewhere) and returns a new MP4, so your original is untouched. Sume cannot crop a still image directly, which is why the crop happens on the clip.

# 1) crop the Short to the centered 4:5 window (free check first)
curl -X POST https://api.sume.com/v1/video-filter/check \
  -H "Authorization: Bearer $SUME_API_KEY" -H "Content-Type: application/json" \
  -d '{"video_url":"https://media.sume.com/artifacts/artf_demo/short.mp4",
       "ops":[{"op":"crop","x":0,"y":0.1484,"width":1,"height":0.7031}]}'

# 2) after video-filter returns the new clip, grab a PNG frame at 1.5 s
curl -X POST https://api.sume.com/v1/video-frames \
  -H "Authorization: Bearer $SUME_API_KEY" -H "Content-Type: application/json" \
  -H "Idempotency-Key: thumb-4x5-001" \
  -d '{"video_url":"https://media.sume.com/artifacts/artf_cropped/short.mp4","at":[1.5],"format":"png"}'

What should I upload as the actual thumbnail?

Upload the full 9:16 still, taken from the original clip with video frames at the source size, as the custom thumbnail. It is the version shown on watch feeds and other platforms. The cropped preview is only a check that nothing important sits in the cut zones.

If a frame fails the preview, pick another time. A single call takes up to 24 explicit times, and the extract is unbilled. Remember that the extracted frame is a video frame, not a designed thumbnail: Sume does not add text or graphics in this step.

What does Sume not do here?

Keeping those limits in mind, the workflow is short: filter check, filter encode, one frame grab, and a visual look at both stills.

• It does not upload to YouTube or set the thumbnail for you; you do that in Studio or through the YouTube API.
• It cannot predict YouTube's own auto 4:5 choice; the preview is a conservative stand-in.
• It does not crop still images; crop runs on a clip with video filter, and a source must be 300 seconds or shorter for both video filter and video frames.