Nano Banana 2 YouTube Thumbnails: Prompts That Actually Click

Thumbnail prompts fail when they look like posters. Here’s how I use Nano Banana 2 for faces that read at phone size.

Rules that matter

• One face, big in frame
• High contrast background
• Emotion readable at 120px wide
• No tiny text in the image (add text in editor)

Copy-paste prompts

1. Shock-reaction face

Close-up portrait, exaggerated surprised expression, bright solid color backdrop, high contrast, YouTube thumbnail composition, face fills frame, photoreal, no text overlay

2. Before/after split energy

Same person as reference, split lighting left dull right punchy, bold thumbnail framing, sharp eyes, no text

3. Product + face

Creator holding product near face, bright backdrop, high clarity, thumbnail-safe composition, stable product edges, no text

Workflow

1. Generate face lock in /generate
2. Fan 4 emotion variants
3. Composite text in CapCut/Canva
4. A/B the top two

Related

/blog/nano-banana-2-prompt-guide · /banana-prompts

FAQs

Should AI write the title text on image?

Usually no—text tools are cleaner.

Vertical or horizontal?

Generate for the placement.

Consistency across a series?

Use refs every time.

Where to start?

/generate.