Build an AI thumbnail for the click—not just a pretty image.
A thumbnail has to communicate at small size. Start with one visual idea, strong subject separation and intentional space for any final text.
Where this job fails
Verify recognizability at small size, crop safety, face/product fidelity and whether text is better added after generation.
YouTube thumbnail prompt
Create a 16:9 YouTube thumbnail for [TOPIC]. Main subject: [SUBJECT]. Communicate one clear idea: [IDEA]. Use strong subject/background separation, readable lighting and simple composition that remains understandable at small size. Reserve [LEFT/RIGHT] negative space for optional headline text. Keep faces/products faithful to references. Do not add random text, logos, extra people or clutter. Optimize composition for thumbnail readability rather than cinematic complexity.
Use a decision loop, not prompt roulette
1 · DefineFreeze the requirement
Write down what may change and what must remain exact.
2 · ChooseMatch the tool to the job
Use Prompt Lab to compare editorial fit and tradeoffs.
3 · GenerateRun a controlled prompt
Change one important variable at a time so revisions remain interpretable.
4 · VerifyInspect the misses
Check fidelity, text, anatomy, materials and constraints before publishing.
Evidence boundaryTool fit is guidance; benchmark results are not published yet.
Our controlled image benchmark protocol is public before the runs. We will not present a benchmark winner until actual outputs and failure evidence are available.