Video thumbnails and covers

A video can go unwatched for one reason — its thumbnail. In search and in the feed a person sees a small rectangle and a title, and decides in a second whether to click. A good thumbnail isn't a frame from the middle of the video but a separate picture: one face or object up close, two or three words in big letters, and contrast that reads even at the size of a matchbox.

A neural network makes pictures like that quickly, but it has a weak spot — letters. Ask it to write "5 MISTAKES" straight into the generated image and the letters can easily warp. So it's more reliable to split the work: the network draws the background and the main image, and the caption goes on top as real text — no typos, and editable one word at a time.

Thumbnail size: YouTube and vertical video

YouTube's requirements are quoted from its help as of 24 September 2026. For other platforms, take the size from the platform's own help — or, if it doesn't give one, use the proportions of the video itself.

The cover exports at double pixel density, so a 1920 × 1080 canvas gives a file of exactly 3840 × 2160 — the size YouTube recommends. More on YouTube thumbnails, Shorts included, in the YouTube thumbnail chapter. Uploading to Rutube instead? Its limit is 1920 × 1080 — see the Rutube cover chapter. For VK Video, which publishes no thumbnail size, see the VK Video cover chapter.

What to write to get a thumbnail, not just a picture

1. The video's topic in one phrase — "reviewing a budget smartphone", "a 30-minute soup recipe".
2. The caption on the cover — two to four words, not the whole video title. The title already sits next to the thumbnail; repeating it in the picture wastes space.
3. The main subject — your face with a clear emotion, a product, a dish. If it's you, upload a photo rather than asking for a look-alike to be generated.
4. The platform — it decides the canvas (table above).
5. The channel's style, if there'll be a series of covers: colour, font, where the caption goes. A consistent layout makes the channel recognizable in search.

Describe the video — the designer will put the cover together right here.

How to check a thumbnail before publishing

• Shrink it to the size of a card in search results. If the caption doesn't read on a phone — fewer words, bigger letters.
• Don't put anything important in the bottom-right corner — that's where platforms usually show the video's length.
• The cover must tell the truth about the video. A thumbnail promising something the video doesn't have earns a click and a exit ten seconds later — and platforms take watch time into account.
• Other people's logos, film stills and celebrity faces on a cover risk a rights-holder complaint, even if a neural network redrew them.

A series of covers without redrawing

Every cover is rendered in a browser and checked before you see it: is the caption clipped, is there enough contrast, has the main text got lost. After that, edits are edits to text and layout: "bigger caption", "make it yellow", "face further right", "change it to ‘PART 2’". The generated background isn't redrawn, so the next episode gets the same layout with a new caption and a new shot.

The video for a cover like this can be made with a neural network too — from video from a text description to vertical clips for social media.

In the Twelver chat a cover is built in layers: the image from the network, the caption as real text, a browser check before you see it and edits without redrawing. Back to the guide to video with a neural network.