YouTube vertical thumbnail size: 9:16 at 2160x3840
YouTube's thumbnail help lists 9:16 at 2160x3840 for vertical videos, 10 MB on mobile podcasts, and 16:9 at 3840x2160. Make one with GPT Image 2.5 on Sume.
YouTube's thumbnail help lists a Shorts thumbnail as 9:16 at 2160x3840 and a video thumbnail as 16:9 at 3840x2160, with a 640-pixel minimum. Files are JPG or PNG. openai/gpt-image-2.5 on Sume can generate exactly 2160x3840. Read 2026-09-29.
What sizes does YouTube list?
The page also says a vertical video with a 16:9 custom thumbnail gets an auto-generated 4:5 thumbnail on home, explore and subscription pages.
| Type | Ratio and size | File size |
|---|---|---|
| Standard | 16:9, 3840x2160 (min 640 px) | 2 MB mobile, 50 MB desktop |
| Shorts / vertical | 9:16, 2160x3840 (min 640 px) | 50 MB desktop (podcasts: 10 MB on mobile) |
| File types | JPG, PNG |
How do I generate 2160x3840 with Sume?
openai/gpt-image-2.5 accepts custom image_size with both edges multiples of 16, the long edge at most 3840, ratio at most 3:1 and 655,360 to 8,294,400 pixels. 2160x3840 is 8,294,400 pixels, exactly at the ceiling, and 2160 = 16 x 135, so it is valid. Expect a slow request; large sizes may return a 202 job to poll.
Ask for space: keep text large and centered, since thumbnails shrink.
curl -X POST https://api.sume.com/v1/images \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-image-2.5",
"prompt": "Bold vertical thumbnail: surprised chef holding a giant croissant, big empty space at top for text",
"image_size": "2160x3840"
}'Can I use a frame from the video instead?
Yes: POST /v1/video-frames returns stills as JPEG at the source size unless you set max_edge (16 to 2160). Note the 2160 cap: a still from a 1080x1920 clip stays 1080x1920, above the 640 minimum but below YouTube's largest listed size.
What are the limits of this answer?
Sume does not upload thumbnails to YouTube. The listed file-size limits are YouTube's and are not checked by Sume.
Sources
Related posts
More in Use cases
- Skincare video ads with AI: three Formats, one packshot
Skincare video ads can be made with AI from a packshot: a model-and-product studio film, an application or texture demo, or a before-and-after with proof.
- Snapchat Spotlight monetization: 30-second videos and follower rules
Snapchat's Monetization Program page says Spotlight videos must be 30 s or longer to earn revenue, with 50,000 followers and view-time limits. Sume builds 30 s+
- AI model for product photos on Sume: Auto or a pinned id
Sume does not rank image models for product photos. Use sume/auto when any good result will do; pin a model id for repeatable, reference-driven shots.
- Edit part of a song with an AI prompt: what to do on Sume
Sume has no edit-a-section route for songs. Instead regenerate with a section-marked brief, change one axis at a time, and keep the takes you like.
Written by Sume