YouTube fhd, qhd, uhd thumbnails in the API: a 3840 AI image

On 2026-09-11 YouTube's Data API notes said the snippet.thumbnails.(key) property now covers higher-resolution sizes for some videos: fhd (1080p), qhd (1440p) and uhd (4K). On Sume's side, the image docs for ChatGPT Image 2.5 say custom image_size allows a maximum edge of 3840, so you can generate an image at the 4K tier; this post covers the keys, not the upload cap.

What are the new thumbnail keys?

The note says the larger sizes exist "for some videos", so code that reads thumbnails must tolerate a missing key. The table below is arithmetic on Sume's image_size rules for the three named resolutions, not YouTube pixel specs. The separate 2 MB to 50 MB upload change of 2026-09-14 is in the 50 MB limit post.

What does Sume accept for a large image?

The normalized resolution tiers are 512, 1K, 2K and 4K, with ratios such as 16:9 and 21:9 through aspect_ratio. For ChatGPT Image 2.5 a custom image_size needs both edges as multiples of 16, a maximum edge of 3840, an aspect ratio of at most 3:1, and 655,360 to 8,294,400 pixels, per Image generation. 3840 by 2160 is 8,294,400 pixels, exactly at the top of that range. For a 1080p-class image, use edges such as 1920 by 1088 that satisfy the multiple-of-16 rule; the docs do not map the 2K tier to 1080p.

The tier-to-model mapping is not asserted here; confirm which model supports which tier in the docs.

Should I generate more than one candidate?

The image schema lets n range from 1 to 4, so one request can return up to four candidates to compare before you upload one with thumbnails.set. Keep image_size and aspect_ratio consistent across candidates so they crop the same way.