A skill that makes Claude design YouTube thumbnails and video covers the way a senior thumbnail designer does. It starts with the one promise the thumbnail makes and the timestamp where the video keeps it. Then it writes three concepts that bet on different ideas: a result, a reaction, a contrast or a question. Each gets a real face with an emotion the video earns, three words or fewer that never repeat the title, and a pixel layout on the 1280 x 720 grid that stays clear of YouTube's own badges. It finishes with prompts for your image tool, a designer brief and a plan for YouTube's A/B testing.
Inside are five reference files. One holds sixteen concept patterns in four families, an emotion table at three intensities, the word rules and five tests every concept must pass. Another covers layout: placement templates, face sizes, type sizes measured from real font files, the WCAG contrast formula and a Python script that tests drafts at 168 and 320 pixels wide. A third has the face shoot list, plate prompts, a comp script and the designer brief. YouTube's file limits, testing rules and policies are cited with sources and dates. The rules are strict. Claude never invents a number, and every face stays a real photograph.
Claude cannot make the images or watch your video. You send the transcript, the title, photos and screenshots, make the plates in your own image tool, and build the final file in Photoshop, Figma or Canva. With code execution on, Claude measures contrast from your actual pixels, checks each draft at the sizes YouTube really shows, and can assemble a rough comp from your files. YouTube changes its tools without notice, so Studio wins when it disagrees with the skill. A/B testing needs advanced features and does not cover Shorts. No thumbnail guarantees a click-through rate, and the skill never promises one.
Claude skill · version 1.0
Thumbnail concepts
Checking your access.
In the download · 9 files
- SKILL.mdThe method Claude follows, in six stages: brief, concepts, layout, make, check and test.
- README.mdHow to install it in claude.ai or Claude Code, how to start, and the licence.
- references/concepts.mdThe click promise and title pairings, sixteen concept patterns in four families, face and emotion at three intensities, word rules and the five concept tests.
- references/layout-and-legibility.mdThe 1280 x 720 grid, the keep-clear map, six placement templates, face and type sizes, the contrast formula, a Python small-size test, and Shorts, podcast and Reel cover layouts.
- references/prompts-and-briefs.mdThe face shoot list, plate prompts, draft comp prompts, a Python comp script, the designer brief template and export settings.
- references/platform-specs.mdYouTube thumbnail files, A/B testing rules, thumbnail policies, AI and endorsement disclosure, WCAG contrast and image tool notes, with sources, checked 2026-10-17.
Install it and begin
01Install it in Claude CodeTerminal mkdir -p ~/.claude/skillsunzip ~/Downloads/thumbnail-concept.zip -d ~/.claude/skills/# Start it with /thumbnail-concept. If this is your first skill, type /reload-skills once.On claude.ai there is nothing to unzip: upload the zip as it is. Every step is on Install and use a Claude skill.
02Your first message to ClaudeText Use the thumbnail concepts skill. My video is [VIDEO TITLE] for [WHO WATCHES IT]. The payoff happens at [THE MINUTE AND SECOND]. The face in it will be [YOURS, A CAST PERSON OR NONE]. Start with the brief and give me three concepts to test.Attach the transcript, screenshots of your last few thumbnails with their click-through rates, and photos of the person who will appear. Turn on code execution so Claude can measure contrast and test each draft at 168 pixels wide.
Checklist
Before a thumbnail goes up
- One idea you can say in a sentence without "and", understood in one second.If it needs "and", it is two thumbnails fighting over a space the size of a stamp.
- The video pays off the promise, and you can name the timestamp.YouTube's A/B testing picks the winner on watch time, so a promise the video breaks loses even when it wins the click.
- Three words or fewer, none of them repeated from the title.The title already carries its words. The thumbnail carries what words cannot: a face, a result, a number at size.
- The face is a real photo, the emotion is one the video earns, and the eyes are sharp.At 168 pixels wide the eyes are four or five pixels tall. Soft eyes and a borrowed emotion both read as wrong.
- It still reads at 168 pixels wide and in greyscale.That is roughly the size of the list beside a playing video. If the face and the focal point merge in grey, no colour will save them.
- Every word measures 4.5:1 or more against the pixels directly behind it.Contrast that passes on the swatch can fail over a bright patch of background. Measure the image, not the hex values.
- Nothing that matters sits in the bottom-right corner or along the bottom edge.The duration badge and the progress bar cover them, and the badge takes a bigger share of the frame the smaller the thumbnail.
- The three test options bet on different ideas and are all exported at 1280 x 720 or larger.Options that look alike make a test run longer, and one small file makes YouTube downscale every option to 854 x 480.
What to do it with
- ClaudeFree tier, paid above it
A language model that writes, edits, reads long documents and runs tools, including agents that work inside a codebase.
Bad at It is agreeable by default, so it will improve a bad idea instead of telling you it is bad. Ask it to argue against you and the quality of the answer changes.
- Claude CodeSubscription
An agent that reads a whole repository, edits files, runs the tests and explains what it changed.
Bad at It is fast enough to make a large mistake quickly. Work on a branch, read the diff, and never let it push. Watch the starting mode: on Pro, Max and Team plans it opens in auto mode, which Anthropic's own list says may read a .env file and send the key in it to the API that key belongs to. Switch it off.
- PhotoshopSubscription
Still the place a picture is finished. Generative fill, expand and remove now sit on the same layer stack as everything else.
Bad at Generative fill matches its surroundings rather than obeying a brief, so it is a patching tool and not a directing one.
- FigmaFree tier, paid above it
The shared file where interface design happens: components, variables, prototypes and the handoff to code.
Bad at Its generative features make plausible screens rather than considered ones, and a plausible screen is the hardest kind to argue with in a review.
- CanvaFree tier, paid above it
Templates, brand kits and generation in one place, built so a non-designer can produce something acceptable.
Bad at Acceptable is the ceiling as well as the floor. Its templates are so widely used that a trained eye dates a deck in a second.
- MidjourneySubscription
Generates images from a written description, with the strongest default aesthetic of anything in the category.
Bad at That default aesthetic is the problem as often as it is the draw. It has a house look, it pulls everything towards it, and getting a plain honest photograph of a real product out of it is harder than it should be. Text inside an image is unreliable.
- GPT ImagePay per use
Generates and edits images with unusually good instruction-following, and can hold text inside a picture better than most.
Bad at Faces drift between generations unless you feed a reference every time. Skin can come out plasticky at default settings, and a heavy fidelity setting costs many times more and takes many times longer.
- GeminiFree tier, paid above it
Google's assistant, with a very long context window and tight links to their other products.
Bad at Behaviour changes between the app, the API and the versions inside other Google products.
- FLUXOpen weights or open source
A family of image models, several with open weights, known for prompt adherence and clean type.
Bad at The open versions need a machine and a person who is comfortable with one. Licence terms differ between the variants, so check which one you are running before it goes anywhere near a client.
- Google FontsFree
Open-licensed typefaces, hosted and self-hostable, covering many scripts.
Bad at Quality is uneven across the catalogue, and the defaults are the most-used type on earth.
Why it matters
Ask a chat for thumbnail ideas and you get ten loud phrases, a shocked face and an arrow, with no check that the video keeps the promise or that anything reads at the size people actually see. This skill makes Claude start from the promise and the payoff, design three concepts that test different ideas, measure the words against the real pixels, and set up a test YouTube judges on watch time, so the winner is a thumbnail the video deserves.
- Type
- technique
- Period
- ongoing