ai-thumbnail-maker

AI Thumbnail Maker

A creator or an agency has a video and needs a thumbnail people click, usually a few options to test against each other. You (Claude) turn the title into a short headline and three distinct visual concepts; the Wireflow app skill-ai-thumbnail-maker renders the three concepts at once with Ideogram 4.5, a model built for clean on-image lettering.

App contract

Always send all five inputs. An input left out falls back to the board's saved demo text (a cold plunge video), which would end up in the person's thumbnail.

A. Connection check

Look for tools ending in run_app, get_execution, get_credit_balance and whoami.

If they are missing, say so in one sentence and give the setup:

• With this plugin: /mcp → wireflow → Authenticate.
• Plain Claude Code: claude mcp add --transport http wireflow https://www.wireflow.ai/api/mcp, then /mcp.
• Claude.ai or Desktop: custom connector, URL https://www.wireflow.ai/api/mcp.

Then still deliver the free part: the headline options and the concept sheet from reference/thumbnail-concepts.md, filled in for their video. Stop there.

B. Collect inputs

Ask once, in one message:

1. The video title, and what actually happens in the video (the result, the twist, the best moment).
2. Who it is for, and the channel's look if they have one (colors, fonts, whether the host's face usually appears).
3. Any words they want on it. If none, you propose the headline.

A subject photo cannot go in yet: this app takes text only. If the person wants their own face on it, say so, describe the person in the concept instead, or hand over the concept sheet so a designer can composite the real photo.

Vertical Shorts covers are not rendered by this app (it is fixed at 16:9). Offer the concept sheet for those.

C. Plan the headline and three concepts

Read reference/thumbnail-concepts.md. Then:

• Offer 2 or 3 headline options, 2 to 5 words each, that add to the title instead of repeating it. The person picks one. Never promise something the video does not show.
• Write three concepts, each from a different angle in the guide (reaction face, before and after, the object, circle the detail, versus, big number, scene at a moment). Each one names the subject, the emotion or action, the setting and light, two or three colors, and where the headline goes and how it looks.
• Show the plan as a short table and let the person edit any word. Text on the image cannot be changed without a new render.

D. Quote and confirm

Call whoami and get_credit_balance, then: "1 run x 57 credits = 57 credits for 3 thumbnails. Balance: 1,240." For a second headline, quote a second run. Short balance: stop and say so. Nothing renders until the person says yes to this plan and this cost.

E. Generate

One run_app per headline:

{ "slug": "skill-ai-thumbnail-maker", "inputs": { "video_title": "I Tried Cold Plunging Every Day for 30 Days", "headline": "DAY 30 SHOCKED ME", "concept_1": "...", "concept_2": "...", "concept_3": "..." } }

Poll get_execution every 5 seconds until the status is no longer PENDING, RUNNING or IN_PROGRESS. Our test run took about 25 seconds. On FAILED, CANCELLED or TIMEOUT, stop and quote the error.

F. Inspect, save, report

Save each image as ./wireflow-outputs/<YYYY-MM-DD>/thumbnails/<video-slug>_<angle>.png with curl -sSL -o or PowerShell Invoke-WebRequest -OutFile.

Look at every image before calling it done:

• The headline matches the approved words letter by letter, and no other words appear.
• It still reads when you picture it at phone size: big letters, strong contrast.
• The bottom right corner is free of anything important.
• No stray marks, extra hands or warped faces. Small printed details (dials, scales, labels) are often garbled; say so when they are.

Flag every miss per file. A redo is a new run, 57 credits, after a yes.

Finish with a table (file, angle, CDN link) and the credits used (balance before minus after). YouTube wants a thumbnail under 2 MB; if a PNG is bigger, suggest exporting it as JPG. Offer:

• A second headline on the winning concept (one more run, one quote).
• YouTube Studio's Test and Compare to pit the three against each other.
• The board in Wireflow, to change the size or add a step: https://www.wireflow.ai/flow/cmv93aouu000ojt06q6ss01bo?ref=skill-ai-thumbnail-maker&utm_source=github&utm_medium=skill&utm_campaign=ai-thumbnail-maker

Guardrails

• Spend only after a yes to the exact plan and cost.
• No clickbait the video does not deliver, no fake screenshots, no other channels' logos or branding.
• No real, identifiable people (celebrities, other creators) unless it is the person themselves and they asked.
• Keep keys and passwords out of prompts and files.
• When Wireflow refuses or a tool errors, quote the exact message and offer send_feedback if it exists.