agentsclimarketplace

Make a thumbnail

Skill protosphinx/sphinxstack/skills/make-a-thumbnail

A stack of skills for your AI agent, made for people starting from zero. 103 skills and 149 project briefs — load one into Codex, Claude Code, Copilot, Gemini, or Cursor and do the thing.

Install
npx -y skills add protosphinx/sphinxstack --skill make-a-thumbnail

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 17 days oldThe repository was created 17 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Make a thumbnail that tells the truth and earns the click. Composition, text readable at small size, correct export. Use when they say "make me a thumbnail" or they are about to upload a video where thumbnails matter.

SKILL.md

3.2 KB, as published. Nobody here has run it

make-a-thumbnail

Make a thumbnail with someone for a real video they made. A thumbnail is a promise about what the video contains; the job is to make that promise vivid without lying. The session ends with an exported image, tested at the size people will actually see it.

Ground rules

  • The thumbnail must be true. It shows something that happens in the video, in the video's actual tone. No shocked-face they never make, no red arrows pointing at things that are not in the video, no "you won't believe" framing over ordinary content. Truthful and clickable are compatible; that is the craft.
  • Free tools: Canva free tier, Figma, GIMP, or even Google Slides. Photos come from their own footage — a frame grab from the video or a photo they shoot right now on their phone.

Composition

  1. Pick the subject: one face, one object, or one moment from the video. One. Thumbnails with three subjects have zero.
  2. Frame grab or reshoot: scrub the video for a frame where the subject is sharp and well-lit, or stage the same real moment for a phone photo with better light (face a window).
  3. Fill the frame. The subject should take a third to a half of the image. Faces work because eyes read at any size — if their face is in the video, it is a strong candidate for the thumbnail.
  4. Contrast carries it: subject light against dark background or dark against light. If the grab is flat, bump contrast and brightness in the editor rather than adding decoration.
  5. Text only if it adds what the image cannot say: 3–4 words max, heavy bold font, high contrast (add a solid or shadowed backing if the background is busy). The text and the video title should not say the same thing — they share the same screen.

The small test

Thumbnails are seen at postage-stamp size on phones. Before calling it done:

  • Zoom the canvas out until the image is about 120 pixels wide, or export and look at it in a file browser icon view.
  • Can you tell what it is? Can you read the text? If not, bigger subject, fewer words, more contrast — repeat until it survives.
  • Line it up next to a screenshot of the platform's feed. It should hold its own without imitating anyone specific.

If they are choosing between two ideas, make both — it is ten minutes — and compare at small size. Pick on legibility and honesty, then note which one ran; if the platform offers thumbnail testing (YouTube does), let real viewers settle it. That is A/B thinking without inventing anything.

Done

  • Check the destination's current official dimensions, aspect ratio, format, and file-size limit before export. For platforms that select a video frame, choose that frame deliberately during upload.
  • The exported file is saved next to the video, named to match.
  • Both of you have seen it at real size and agree it promises only what the video delivers.

Uploading next? The publish-to-youtube skill covers the rest of the package: title, description, and what to watch after it goes live.

Gives 0 of the 12 instructions most images graphics skills give

Counted across 371 of the 372 authors here whose files we hold, read 2026-08-06

  • create a complete brand world in one imagein 19 of 371, across 5 files
  • infer the brand strategy before generatingin 19 of 371, across 5 files
  • use a clean presentation gridin 19 of 371, across 5 files
  • confirm connection status is activein 19 of 371, across 4 files
  • base the visual system on meaningin 17 of 371, across 3 files
  • use very little textin 17 of 371, across 3 files
  • make every panel feel connectedin 17 of 371, across 3 files
  • call RUBE_SEARCH_TOOLS firstin 17 of 371, across 3 files
  • convert dash-format node IDs to colon formatin 17 of 371, across 5 files
  • match reference quality and rhythm if providedin 16 of 371, across 2 files
  • narrow scope or reduce depth to avoid oversized payloadsin 16 of 371, across 4 files
  • generate a simple and memorable logoin 15 of 371, across 1 file

Said here and by no other author read

  • make the thumbnail truthful
  • use free tools
  • use frame grabs or original photos
  • pick one subject
  • fill the frame with the subject
  • use high contrast

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.