Thumbnail Background
Generate a 16:9 background for a video thumbnail — bold, high contrast, and with room left for your face and title.
It will not write your title. The prompt forbids lettering, because image models produce text that looks convincing until you read it. Generate the background here and set the words afterwards, where they will be sharp and spelled correctly.
What should it look like?
These choices decide the result. You never have to write a prompt.
A short note — “with faint circuit patterns”, “with a subtle grid”.
0/200Your image
Generation runs on our server, because it needs a model your browser cannot. Your photo is sent for this step only and is not kept afterwards.
Ready.
Design for 210 pixels, not for the screen you are on
A thumbnail is chosen at a size where almost nothing survives.
Large shapes, a strong colour split, hard contrast between foreground and background. If you squint at the result and can still tell what it is, it will work in a feed.
Fine texture, subtle gradients, small elements near the edges, and anything mid-tone against mid-tone. It all turns to grey mush at the size that decides whether anyone clicks.
Why it asks what to leave room for
- Your face
Most thumbnails that perform put a face on one side. Asking the model to keep that half quiet means you can composite without fighting a busy background.
- Your title
Large type needs a calm area behind it. A clean lower third or centre lets the words sit without an overlay dimming the whole image.
- The duration badge
The platform draws its own timestamp over the lower right corner. Avoid putting anything you care about there.
Finishing it
- Add the title textAdd Text to Image
- It needs to be exactly 1280×720Resize Image
- It is over the 2 MB upload limitCompress Image
- You want a product in it, not a backdropProduct Background Creator
Frequently asked questions
16:9, typically 1024×576. YouTube asks for 1280×720, so scale it up slightly with Resize Image, or leave it — the platform will scale it for you and at these proportions the difference is not visible.
Image models are unreliable at lettering: they generate shapes that read as text until you actually try to read them, and a misspelled word on a thumbnail is worse than no thumbnail. Add the title afterwards with a tool that renders real type.
No. Generation is not deterministic, so the same choices give a different image each time. Download the ones you like — regenerating will not bring one back.
That is deliberate. Thumbnails compete at a size where detail becomes noise, so the prompt asks for large simple shapes and hard contrast. A busier image usually performs worse even though it looks better while you are making it.