Promptkoi

AI Prompts for YouTube Thumbnails

A thumbnail has one job: win the click at 200 pixels wide. That flips normal prompt priorities — subtlety loses, and the things that make gallery images beautiful (soft light, muted palettes, busy detail) actively hurt. Thumbnail prompts push exaggerated facial expressions, one dominant subject filling 60% of the frame, saturated complementary colors, and hard rim lighting that separates the subject from the background. Generate at 16:9, and leave deliberate empty space on one side for your title text — the prompts here ask for it explicitly. We also note which models handle text: Flux can render bold headline words in-image, while Midjourney thumbnails work best as backgrounds with text added in an editor afterwards.

Prompts in this collection

Frequently asked questions

How do I write a thumbnail prompt that wins the click?
Design for 200 pixels wide, which flips normal priorities: subtlety loses. Push exaggerated facial expressions, one dominant subject filling about 60% of the frame, saturated complementary colors, and hard rim lighting that separates the subject from the background. Generate at 16:9 and leave deliberate empty space on one side for your title text. The prompts here ask for that empty zone explicitly.
Should I add the title text in the prompt or afterward?
Add it afterward in most cases. Reserve empty space on one side in the prompt, then set the headline in an editor where you control font, size, and legibility. Flux can render bold headline words in-image if you want text baked in, while Midjourney thumbnails work best as backgrounds with text added later. Rendered lettering from other models is still unreliable.
Which model is best for YouTube thumbnails?
For text baked into the image, Flux can render bold headline words directly. For a striking background you add text over later, Midjourney's punchy grade and rim lighting are hard to beat. Both benefit from a 16:9 frame and a subject filling most of it. We note per entry which model we tested and whether it handles in-image text.