Orshot YouTube Thumbnail: one template per channel, one render per video

A channel's thumbnails share a look: the same font, the same color box, the face in the same place. Only the words and the picture change. That is a template render. Design the channel's thumbnail template once in Orshot, then render it per video with the hook words, the subject image and the accent color. Video 40 looks like video 1, and the text is real text, so it is sharp and spelled right.

What people ask for

• "Make a thumbnail for my new video: a frame at 1:23, my face cutout, and the words 'I QUIT MY JOB'. Same style as my last ones."
• "Here are my last 6 thumbnails. Turn the style into a template so I only swap the face and the words each week." (Match the current style first.)
• "Give me 3 variants for Test & Compare: same picture, different words." (A/B variants.)
• "Every time I upload, make the thumbnail from the title and set it on the video automatically. I use n8n." (A thumbnail for every new video.)
• "Redo the thumbnails for my 40 old tutorials from this sheet of titles and timestamps." (Redo thumbnails on old videos.)
• "No AI art. A gameplay screenshot of my Glasswing Rogue build and the build name, big and readable." (Screenshot + name layout.)

The Orshot docs are the reference for every parameter, value and limit. Each docs page has a markdown version at https://orshot.com/docs/<path>.md. This skill gives the order of steps, the thumbnail craft and worked examples.

What you get per video

Setup

1. Connect the Orshot MCP server: https://orshot.com/docs/integrations/mcp-server.md has the steps for each client. Headless or scheduled jobs use REST with an API key: https://orshot.com/docs/quick-start/get-api-key.md
2. Get the brand: orshot_get_brand_kit (colors, fonts, logo, and any photos of the host). Empty kit and a website? orshot_scan_brand_website with the URL, and ask before importing with orshot_import_brand_website. No brand at all? Load the orshot-design skill to pick a palette and a font pair.
3. orshot_plan_design with the brief, job: "youtube thumbnail for <the video>", 1280x720 and on_brand: true when the channel has a brand kit. It returns the layouts that fit (geometry and a full pages_data example each) and the palette and type. Read orshot_get_design_spec topic template-json once per session; the same content is at https://orshot.com/docs/definitions/template-json.md

Thumbnail rules

• Real over generated. Use the video's own frames and the host's real photos. Many viewers say they skip or block channels whose thumbnails look AI-made, while a plain frame with big words still gets clicks. Never generate the host's face or a pose. No layout guarantees a click-through rate: let Test & Compare decide between variants.
• 3 to 5 words, and not the title. The title says what the video is. The thumbnail says why to click: the result, the stakes or the surprise. Title "I tried 5 budget mics for a month", thumbnail "$20 BEATS $400?".
• One subject, big. One face with a clear expression, or one object. It fills 40 to 60% of the frame. No group shots, no busy scenes.
• High contrast. Light words on a dark ground or dark words on a bright box. Add a dark gradient behind words that sit on a photo.
• Readable at 168x94. That is about the size in search results and the sidebar. Shrink the render to 168 wide and look. If a word is hard to read, it is too small or too long.
• No small text. Headline 120 to 200px on the 1280x720 canvas. Nothing under 60px. No dates, URLs, subtitles or paragraphs.
• Brand color as the accent. One color box or one highlighted word in the channel color, the same on every video.
• Same series style. Same layout, font, color and face position on every video in a series, so a viewer knows it is your channel before reading.
• Keep the bottom-right corner clear. YouTube puts the video length there.

Match the current style first

If the channel already has thumbnails, the template copies that look. Ask for the last 5 to 10. Note the font, the text box, where the face sits, the background treatment and the accent color, then build the template to match. Show the first render next to two of the originals, and change only what the user asks for. A new channel starts from a layout below.

Pick a layout per series

Geometry for each layout and a full pages_data example come back from orshot_plan_design; all five at once: orshot_get_design_spec topic youtube-thumbnail-layouts.

Build the template once per channel style

1. Check first: orshot_list_studio_templates with search Thumbnail. Reuse what exists.
2. Create: orshot_create_template_design with canvas_width: 1280, canvas_height: 720, a name like Thumbnail: <channel> (Face + big words), tags ["youtube", "thumbnail"], and includeThumbnails: true.
3. Look at the returned preview. Fix the designWarnings that show a real problem in one orshot_update_template_design call, and tell the user which ones you left and why.
4. Shorts cover: add it as page 2 (1080x1920) with the same parameterIds, laid out for the tall frame: words in the middle band, the subject large below. Resizing the 16:9 page with extraSizes leaves the words too small to use. A two-page template takes page1@ and page2@ keys.

Make these layers dynamic:

Give each text layer the Fit text mode with a minimum font size, so a long word shrinks instead of spilling out (https://orshot.com/docs/orshot-studio/text-modes.md). Text color can also change per render with a style key such as "headline.color" (https://orshot.com/docs/orshot-studio/style-parameters.md).

Get the subject image

• The host's face. Use a photo the user gives you. Upload it with orshot_upload_brand_image (attachment for a file in the chat, or file for a URL) and tags ["face"], then keep its direct_url. Ask for 3 to 5 photos with different expressions once and reuse them.
• Cut it out. Check the fields with orshot_get_library_template_modifications for image-background-removal-api, then call orshot_render_library_template with templateId: "image-background-removal-api", modifications: { "inputImage": "<photo URL>" } and format: "png". Trim the transparent edges before using it (for example magick cutout.png -trim +repage subject.png), check for leftover background, and upload the trimmed file: an untrimmed cutout keeps the photo's full canvas and sits small in the subject box. It uses an AI credit, not a render credit. Over REST it is POST https://api.orshot.com/v1/generate/images with the same templateId and modifications (https://orshot.com/docs/api-reference/render-from-template.md).
• A frame from the video. Grab one locally: ffmpeg -ss 00:01:23 -i video.mp4 -frames:v 1 frame.jpg, then upload it the same way. Use it as background, or cut out the person in it for subject.
• Stock. orshot_search_stock_media with the topic in 1 to 3 words and type: "photo". Put the result's stock:// ref in the template, or call orshot_use_stock_media with the ref to get a URL for a render.

Render per video

Over MCP, orshot_render_template:

{
  "templateId": 123,
  "modifications": {
    "headline": "$20 BEATS",
    "headline_accent": "$400?",
    "subject": "https://.../face-shocked-cutout.png",
    "background": "https://.../frame-0123.jpg",
    "accent": "#FFD400"
  },
  "format": "jpg",
  "quality": 90,
  "fileName": "mic-test-thumbnail"
}

The thumbnail URL is in content. With a Shorts page, prefix every key (page1@headline, page2@headline) and both pages come back in one render. Costs: https://orshot.com/docs/definitions/credits-usage.md

A/B variants in one call

YouTube's Test & Compare tests up to 3 thumbnails per video. Render them all at once with orshot_batch_render_template:

{
  "templateId": 123,
  "format": "jpg",
  "quality": 90,
  "renders": [
    { "headline": "$20 BEATS", "headline_accent": "$400?", "subject": "https://.../shocked.png" },
    { "headline": "CHEAP MIC", "headline_accent": "WINS", "subject": "https://.../shocked.png" },
    { "headline": "$20 BEATS", "headline_accent": "$400?", "subject": "https://.../holding-mic.png" }
  ]
}

Change one thing per variant (the words or the image, not both), so the test tells you which one won. Show the user all three before they upload.

Redo thumbnails on old videos

1. A sheet with one row per video: video ID, title, hook words and a frame (an image URL, or a timestamp to grab with ffmpeg as above).
2. Write 3 to 5 click words per row, not the title (rules above). Show the list before rendering.
3. Say the cost first: 1 credit per thumbnail, plus 1 per Shorts page (Credits). Then render in chunks with orshot_batch_render_template (the tool schema gives the per-call cap) and write each URL back to its row.
4. The user sets them in YouTube Studio, or an automation calls thumbnails.set (next section).
5. Change a few at a time and watch those videos in YouTube Studio before changing the rest.

Automate it: a thumbnail for every new video

The template has dynamic fields, so any job can render it. A common setup: a sheet with one row per upcoming video (hook, hook_accent, face_url, frame_url). Each new row renders a thumbnail:

curl -s -X POST https://api.orshot.com/v1/studio/render \
  -H "Authorization: Bearer $ORSHOT_API_KEY" -H "Content-Type: application/json" \
  -d '{
    "templateId": 123,
    "modifications": {
      "headline": "$20 BEATS",
      "headline_accent": "$400?",
      "subject": "https://.../shocked.png",
      "background": "https://.../frame.jpg"
    },
    "response": { "type": "url", "format": "jpg", "quality": 90 }
  }'

data.content is the thumbnail URL; write it back to the row.

1. Get the exact field keys with orshot_get_studio_template_modifications before wiring anything.
2. Add the render step to the tool the user already runs:
   • n8n, Make or Zapier: the Orshot node, module or app, or an HTTP request with the body above. Docs: integrations/n8n, integrations/make, integrations/zapier (each opens as https://orshot.com/docs/<path>.md).
   • No orchestrator yet: an Orshot Workflow can watch a Google Sheet and render. Use the exact node keys from orshot_list_workflow_nodes (for example google_sheets_new_row and render), then orshot_validate_workflow, then orshot_create_workflow with status: "draft". Docs: integrations/google-sheets.

Per upload in n8n, set on YouTube

Orshot Workflows have no YouTube or RSS trigger, so this one runs in n8n:

1. RSS Feed Trigger on the channel feed https://www.youtube.com/feeds/videos.xml?channel_id=<channel id>. It fires once per new video.
2. LLM node: 3 to 5 click words from the title (rules above), not the title itself.
3. HTTP Request to Orshot: the render above, with a face photo from the brand kit and a frame or channel background. For variants, render 2 to 3 and let the user pick one through an n8n form, Slack or email before the next step.
4. HTTP Request to the YouTube Data API thumbnails.set for that video ID, with the image file as the body and the user's own Google OAuth2 credential in n8n. Orshot never holds the YouTube login.

Gotchas

1. Orshot does not set the thumbnail on YouTube. Social posting does not send images to YouTube (https://orshot.com/docs/publish/social-posting-guidelines.md). Upload the file in YouTube Studio, or through the YouTube Data API thumbnails.set in the same automation.
2. YouTube takes custom thumbnails up to 2 MB. Render jpg with quality 85 to 90; a photo-heavy PNG can be larger.
3. Plain text breaks lines with \n. Inside rich text (a <span> or <b>), \n is not converted, so use <br> there. See the Text section of https://orshot.com/docs/definitions/template-json.md
4. An accent word in another color stays in the headline as an inline <span>. An accent line in a colored box is the exception: a <span> with a background color stops the text from wrapping, so put a separate headline_accent layer over the accent shape. Make the accent line the biggest text, at least 1.4 times the other line.
5. Template images must be public URLs or stock:// refs. Never use a stock search previewUrl; it is a temporary link.
6. Reuse one template per series. A new template per video spends design effort twice and breaks the series look.

Safety

• No misleading thumbnails. The words and image must match what the video shows. No fake quotes, fake screenshots or invented numbers. A before and after shows the same person or thing in both halves, from the video.
• Use only photos the user owns or has the rights to: their own face, their own frames, licensed stock. Never cut out another person's face or a creator's frame without permission.
• No generated faces or poses of the host. Cut out a real photo instead.
• Do not promise a click-through rate. Suggest Test & Compare.
• Show the thumbnail to the user before it goes anywhere. Say what a batch costs before rendering more than 10.
• Ask before importing a website into the brand kit or overwriting existing files.

Related skills: orshot (API, render options, automation), orshot-design (design craft), orshot-video-generator (the video itself), orshot-blog-images (blog covers), orshot-social-media-content (social posts that promote the video).