YouTube thumbnail maker from a host photo

What it is

This n8n workflow turns a video’s title and a photo of its host into a finished YouTube thumbnail, 3840 by 2160 pixels. A form takes the title, a description, the host’s photo, up to five tool or logo images and, if you like, an existing thumbnail to imitate. Cloudinary stores every upload. With a reference thumbnail, OpenAI reads it into a JSON design spec, and two agents rewrite that spec for your video. Without one, an agent writes five concepts, you tick one in a second form, and two agents turn it into a prompt. Either way, Seedream v4 edit on WaveSpeedAI renders the picture from the prompt and your images, and the last node holds its link.

Who it suits

YouTubers who film themselves and want a thumbnail for each video that keeps their own face and the real logos of the tools they cover. Small channel teams and editors who make several a week can use it too, with the form as their brief.

You need some n8n: importing a file, adding an OpenAI credential and filling in four values in one node. If that is new and you would rather not learn it, skip this one. The page it came from estimates 11 minutes, which assumes the three accounts already exist.

Where to use it

Any n8n you run will do, as long as it can reach Cloudinary, OpenAI and WaveSpeedAI.

• n8n Cloud: import the file into a new workflow there
• Your own n8n server: the same file, the same steps
• YouTube Studio: the workflow ends with a link, so download the picture and upload it as the video’s custom thumbnail yourself
• An image editor: open the picture there to fix any text before you post it

Setup

Get it saves a zip with workflow.json and Mindket’s setup notes in your downloads folder. Have these ready before you import it.

1. In n8n, create a workflow, open its menu, choose Import from File and pick workflow.json.
2. Open the node API Values. Replace YOUR_CLOUD_NAME, YOUR_PRESET_NAME, YOUR_FOLDER_NAME and YOUR_WAVESPEED_API_KEY with your own values. Every upload and render node reads them from here.
3. Add an OpenAI credential and select it in OpenAI, OpenAI1 and Analyze image.
4. Save, open the form’s test address and send a title, a description and a host photo, then press Publish.

How to use it

Open the form from On form submission. The title and the host photo are required. The description is where the agents learn which tools, products or people to show, and they are told not to invent names, so list them. Add logos if you have them, and a reference thumbnail if you want its layout. With a reference, the run goes straight on to the render. Without one, the form moves to a page of five concepts as checkboxes: tick one and submit. The workflow then waits one minute, fetches the result once and writes the picture’s address in Here is the download link. Open the run under Executions to copy it.

Example

A form entry for a list video:

Enter your Video Title?: Five AI Tools Every Small Shop Should Try
Describe what your video is about?: A walkthrough of Zapier, Canva, Notion AI, Otter and Perplexity for shop owners
Host Image: host.jpg
Tools and Logo images (max 5): five logo PNGs

With no reference thumbnail, Thumbnail Idea Generator returns five concepts, each with a Visual, a Background, Assets, Text, a Palette and Notes. For a list title it adds numbered badges and spreads the logos evenly. Once you tick one, Image Prompt Generator writes a paragraph about it and Prompt Patterns Converter sorts that paragraph under nine headers:

Portrait (KV series)
Change Clothes / Jewellery / Makeup
Background Replacement
Interior / Outdoor Replacement
Pose / Prop Adjustment
UI / Object Overlay
Typography / Title Overlay
Composition Constraints
Additional Details

Seedream receives that text with your photo and logos, and the last node ends with the picture’s address:

{
  "Here is the download link": "<the picture’s address>"
}

For newcomers and experts

If you are new. n8n draws a workflow as boxes joined by lines, and each box, a node, does one job. A credential is a key n8n keeps for you and hands to the node you pick it in. After the first submit, the form shows its second page in the same browser tab, so leave the tab open until the concepts appear. A run that has a reference thumbnail has no second page.

For experts. Two chains do the same job, and the nodes that end in 1 belong to the reference path.

Set Input and Set Input1 build the render request: the host photo first, then every logo that uploaded, a size of 3840*2160 and enable_sync_mode off, so n8n waits and then fetches the result. Logo uploads for files you did not send fail quietly and are skipped.

Good to know

Each run costs money, and the picture is only as good as the photo, the logos and the description you give it.

• Billing: OpenAI bills the vision and chat calls and WaveSpeedAI bills the render
• One fetch: the workflow asks for the result once, a minute after sending the prompt. A longer render leaves the last node without a link, so raise the wait
• The link: it appears only in the run’s data, so add a node after Download Image or Download Image1 to send it where you need it
• Faces and logos: the prompts ask the model to keep the host’s face and to use the logos you upload. A generated picture can drift, so check faces, spelling and logos before you post
• Reference thumbnails: that path asks the agents to match the reference closely, so use one you made or have the right to copy
• Photos: every upload goes to Cloudinary under your preset and to OpenAI and WaveSpeedAI
• Open form: once it is published, anyone with the form’s address can start a run on your keys, so turn on the form trigger’s basic authentication or keep the address private
• Keys: the WaveSpeedAI key sits in a field of API Values, not in a credential, so it travels with any export of the workflow
• A repair: the page’s copy of the file had lost the backslashes in the patterns of the Code node, which stopped it from splitting the five concepts. Mindket put them back
• Credentials: the JSON holds none, and nothing from its author’s account