Thumbnails decide whether anyone watches your video. Most top creators use the same formula: their face showing a clear emotion, a few bold words, and high contrast. Making one by hand means a photo shoot, background removal, layout and text. That's an hour per video if you're quick.
An AI agent can do it in a couple of minutes, with your real face on it. Here's how.
An AI agent is an assistant like Claude or ChatGPT that can use outside tools. When you connect a thumbnail tool to it, you don't open a design app. You describe the thumbnail in chat, attach your photo, and the agent makes it.
The advantage over a standalone AI image tool is context. The agent can read your script or your video link and work out the hook before it designs anything.
An AI agent: Claude (custom connectors work on the free plan), ChatGPT (paid plan, Developer mode), or Claude Code and Cursor.
A thumbnail tool connected to it. I'll use Unsora, which includes a thumbnail generator along with video and clipping tools.
A clear photo of yourself.
A reference thumbnail whose style you like (optional, but it makes a big difference).
In Claude: Customize → Connectors → + Add → Add custom connector, paste https://mcp.tryunsora.com/mcp, and sign in. In ChatGPT, turn on Developer mode under Settings → Connectors and add the same URL. Setup for every agent is on the connector page.
Your photo matters more than any prompt. Aim for:
Face clearly lit from the front, no harsh shadows
Eyes open, looking at or near the camera
A strong expression: surprise, curiosity, a confident grin
Decent resolution (any modern phone works)
Take several expressions in one go so you can reuse them across videos.
The best results come from three things together:
Your face: the photo.
The style: a thumbnail you like, from your channel or one you admire.
The context: the title, script or YouTube link of the new video.
Example prompt:
Make a YouTube thumbnail for my video "I Tried a 21-Day Energy Reset as a Founder". Use my photo (attached) and match the style of the reference thumbnail. Max 4 words of text, high contrast, my face on the right.
The agent reads the context, picks a hook, and makes a thumbnail in that style with your face on it.
Rarely is the first version the one. Ask for changes the way you'd brief a designer:
"Make the text bigger and yellow."
"Swap my expression for the surprised photo."
"Give me three variations with different text."
"Darker background so my face pops."
Making three or four versions and picking the strongest is the fastest route to a high click-through rate.
AI makes production fast, but the fundamentals are the same:
Three words or fewer when possible. Thumbnails are viewed tiny on phones.
One focal point. Your face or the object, never five things.
Don't repeat the title. The thumbnail should add curiosity, not duplicate it.
Test at small size. Shrink it to the size it shows on a phone. If you can't read it, simplify.
Stay honest. Clickbait that doesn't match the video hurts retention, and retention matters to the algorithm.
The agent approach really pays off with batches. If you're cutting a podcast into shorts, ask for a thumbnail for each clip in the same message:
Find the 5 best moments in this episode, cut them 9:16 with captions, and make a thumbnail for each using my photo.
I covered the clipping half of that workflow in how to turn one long video into captioned shorts.
Once you have a thumbnail style that works, reuse it. Save your best-performing thumbnail as the reference for future videos, keep the same font colour and face position, and tell the agent "match my last thumbnail" every time. Viewers start recognising your videos in their feed before they even read the title. That brand recognition builds up over dozens of uploads.
If you rotate expressions, keep a small library of approved photos (happy, shocked, thinking, serious) and pick the one that fits the video's emotion. Because the agent already has your references, switching is a one-line request rather than a new photo shoot.
Once the video and thumbnail are ready, the same agent can schedule the upload to YouTube, and to TikTok, Instagram, Facebook and LinkedIn as well. You approve the queue and nothing posts until you say so.
With Unsora, everything you generate comes with commercial rights at full resolution, so you can use it on a monetised channel. Plans start at $19 a month, with a 3-day free trial.
A good thumbnail needs your face, a clear emotion, a bold hook and a consistent style. An AI agent handles the production part: give it your photo, a reference and the video's context, then refine it in a few messages. You keep the creative call, and the hour of design work becomes two minutes.