AI thumbnail generator for YouTube: 3840×2160 from a prompt

To make a YouTube thumbnail with AI, generate a 16:9 image at the size YouTube recommends, 3840 × 2160 pixels (2160 × 3840 at 9:16 for Shorts), then check the faces and any words before you upload it. On Sume, ChatGPT Image 2.5 takes both sizes exactly as a custom image_size, from a prompt alone or with your own photo as a reference.

YouTube's numbers are quoted from its Help page Add custom thumbnails on YouTube, which can change; Sume facts come from the Image API docs and the Sume API reference. All were read on 2026-09-27. To use a frame of your own video instead, see YouTube thumbnail from a video frame.

What size is a YouTube thumbnail?

YouTube says a custom thumbnail should be as large as possible and names image formats such as JPG or PNG. Its recommended sizes pass every ChatGPT Image 2.5 custom-size rule: both edges are multiples of 16, the long edge is 3840, the shape is inside 3:1, and 3840 × 2160 is exactly the 8,294,400-pixel ceiling.

How do I generate a thumbnail from a prompt?

Send POST /v1/images to openai/gpt-image-2.5 with the size and a prompt that says what goes where: the subject large on one side, a plain area for the title on the other, strong contrast. Ask for jpeg or png, the formats YouTube names.

• The call waits up to 30 seconds. High quality, the default on this model, and a large n are slow configurations most likely to return 202 with a job to poll instead of the image.
• Add n for up to 4 versions to choose from, the model's ceiling today; each image is billed. The price is estimated from size and quality, and usage.cost is the USD amount billed.
• YouTube's size limit is 2 MB for a video thumbnail uploaded from mobile and 50 MB from desktop, so check the file size.
• Results are Sume-hosted, signed URLs in data[].url. Download the image, then upload it in YouTube Studio.

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": "YouTube thumbnail: a surprised chef holding a giant pancake on the right, bright kitchen, bold colors, empty space on the left for a title, no text",
    "image_size": { "width": 3840, "height": 2160 },
    "output_format": "jpeg"
  }'

Can I put my own photo in the thumbnail?

Yes. Send your photo, a product shot, or a still in input_references as public HTTPS URLs; ChatGPT Image 2.5 takes up to 16. Say how to use it, and keep image_size in the request so the result is 16:9 rather than the photo's shape.

The model draws a new image from the reference, so faces, logos, and labels can come out different. Compare the result with your photo before you publish it.

{
  "model": "openai/gpt-image-2.5",
  "prompt": "The person in the reference photo, pointing at a laptop with a shocked face, studio light, empty space on the left for a title, no text",
  "input_references": [
    { "type": "image_url", "image_url": { "url": "https://example.com/me.jpg" } }
  ],
  "image_size": { "width": 3840, "height": 2160 }
}

Should the AI write the title text?

It can, but check every letter: generated lettering can be misspelled. For a title that must be exact, ask for empty space and set the words in an editor. AI image with text covers both routes.

What are the limits?

• Sume's part ends at the image file. You upload it in YouTube Studio, where your own thumbnails need a verified account and a channel can upload only a limited number each day.
• The exact sizes above are ChatGPT Image 2.5 custom sizes. Nano Banana 2 and Nano Banana Pro list 16:9 and a 4K tier, but the docs don't give that tier's pixel size, so check the file.
• YouTube replaces a vertical video's 16:9 custom thumbnail with an auto-generated 4:5 one on the home, explore, and subscription pages, so make Shorts thumbnails at 9:16.