Create reviewable thumbnails with ak:youtube-thumbnail-design

Use ak:youtube-thumbnail-design to turn a video topic and short text treatment into one or more PNG thumbnail variants. The Skill gathers an art-direction brief, can use a consented reference face, generates the complete image through Gemini, and presents the files for review. It designs thumbnails only; it does not edit the video, create channel art, or upload anything to YouTube.

Choose this Skill for a thumbnail deliverable

Use ak:youtube-thumbnail-design when

• You need a new YouTube thumbnail or variants for an existing concept.
• You have a video topic, a short text treatment, and a target niche or style.
• You want brand colors, a channel name, a font direction, arrows, or a consented face reference included in the brief.
• You can inspect small-size readability, text accuracy, visual truthfulness, and platform compliance before upload or testing.

Choose another workflow when

• You need a script, storyboard, generated video, or platform export. Use ak:video.
• You need channel art, a campaign banner, or a website hero. Use ak:banner-design.
• You need to repurpose an existing video's content. Use ak:youtube.
• You need measured experiment analysis. This Skill creates variants; it does not run an A/B test or prove a CTR outcome.

Prepare the brief and image provider

Before you start:

• Complete Onboarding, and confirm Marketing Kit is installed for the current runtime and scope.
• Provide the video topic and thumbnail text. The packaged brief recommends one to three words and warns above 25 characters.
• Choose a niche and style, or ask the Skill to recommend directions from its packaged niche and style data.
• Provide brand name, colors, and font direction when they are approved.
• Use only a reference face or source image you may send to the provider and transform. Obtain the depicted person's informed consent.
• Make GEMINI_API_KEY available, install google-genai, and allow network access when generation is requested.
• Confirm the output project and expected number of variants. The brief's default quantity is three.

See Runtime adapters for component-level differences.

Run the Skill

Include the topic, exact text, style or niche, quantity, brand constraints, and the boundary that files are drafts until approved.

/ak:youtube-thumbnail-design Design three thumbnails for a privacy-first analytics tutorial. Text: "OWN YOUR DATA". Niche: tech. Styles: diagram, bold-text, dark-dramatic. Use the approved blue palette, keep the duration-badge area clear, and do not upload

/ak:youtube-thumbnail-design Design three thumbnails for a privacy-first analytics tutorial. Text: "OWN YOUR DATA". Niche: tech. Styles: diagram, bold-text, dark-dramatic. Use the approved blue palette, keep the duration-badge area clear, and do not upload

$ak:youtube-thumbnail-design Design three thumbnails for a privacy-first analytics tutorial. Text: "OWN YOUR DATA". Niche: tech. Styles: diagram, bold-text, dark-dramatic. Use the approved blue palette, keep the duration-badge area clear, and do not upload

Understand the packaged generator controls

When the workflow uses its bundled Python generator, these are the canonical controls most relevant to the result:

The packaged preset mapping uses a Flash image model for fast, a Pro image model at 2K for normal, and the same Pro family at 4K for ultra. Treat those as packaged requests, not guarantees that the provider will keep the model or resolution available.

Follow the design stages

1. Confirm requirements. The Skill captures topic, short text, niche, style, brand, reference face, quantity, and upload boundary.
2. Choose art direction. It searches local niche and style guidance and can prepare a design brief before generation.
3. Generate variants. The generator sends the prompt and any supplied reference image to Gemini, renders text into the image, and writes PNG files.
4. Present the options. Each result should include its style, preview, rationale, path, and any failed generation.
5. Review and iterate. Check the exact text, likeness, composition, contrast, mobile readability, brand use, and absence of fake UI.
6. Stop before distribution. Selection, experiment setup, upload, metadata changes, and publication remain separate approvals.

Keep consent and publication separate

Use only brand assets, fonts, photography, faces, and symbols you have the right to use. Do not imply an outcome the video does not deliver, fabricate a person's endorsement, or add fake duration badges and play controls.

Review outputs and evidence

The normal batch convention is:

assets/thumbnails/{video-slug}/
├── thumb_bold-text_01.png
├── thumb_facecam_02.png
└── thumb_dark-dramatic_03.png

Before approval, record:

• The exact prompt fields, style, niche, font, brand inputs, quality preset, and whether a reference image was sent.
• Every generated path and any provider failure.
• A visual check at full size and at small feed sizes, including the bottom-right duration-badge area.
• Text spelling, likeness and consent, source-asset rights, and truthfulness of the implied claim.
• The selected candidate and any upload or experiment action still pending.

Troubleshoot safely

Continue with the Marketing Kit overview or Projects, artifacts, and checkpoints.

Know the current limits

• The style reference describes 17 directions, while the bundled generator's --style choices are facecam, before-after, listicle, tutorial, reaction, cinematic, bold-text, mystery, product-showcase, dark-dramatic, diagram, and whiteboard. A reference entry is not automatically a generator flag.
• The generated file may not be exactly 1280 × 720 or under a platform upload limit merely because the prompt asks for a 16:9 thumbnail. Inspect and resize or compress the approved asset as needed.
• Text rendering, face consistency, fonts, and safety filtering depend on the provider and must be reviewed visually.
• No style, arrow, facial expression, or variant count guarantees CTR, engagement, revenue, or experiment significance.
• Platform specifications and UI safe zones can change; verify current upload requirements before publication.