News recap video with a HeyGen avatar

News recap video with a HeyGen avatar

An n8n workflow. It reads the latest Morning Brew issue with Apify, has an agent write a short spoken recap, asks HeyGen
to render an avatar video and checks until the video is ready. The original's setup note names Nate Herk as its author.

What is inside

• workflow.json: the workflow as n8n exports it, 18 nodes: 13 that work and five sticky notes. It holds no credential.
  The line HeyGen speaks is the placeholder YOUR_SCRIPT_TEXT, and avatar_id and voice_id are placeholders as they
  were in the original.
• Two quotation marks inside the crawler's JSON body were not escaped, which made the body invalid JSON. They are
  escaped here and nothing else in it changed.
• README.md: these notes.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A HeyGen account with API access, its API key, and an avatar and a voice you can use.
• An Apify account and its API token.
• An OpenRouter API key.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an HTTP Header Auth credential for HeyGen, with the header name X-Api-Key and your key as the value. Select it in
   Generate Video, Generate Video1, Get Video, Get Video1, Get Avatars and Get Voices.
3. Add an OpenRouter credential and select it in GPT 4.1 Mini.
4. Open News. It has no authentication yet. Set Authentication to Generic Credential Type and add an HTTP Header Auth
   credential with the header Authorization and the value Bearer followed by your Apify token.
5. Run Get Avatars and Get Voices once, each on its own with Execute step, and copy an avatar's id and a voice's id
   from what they return.
6. In Generate Video1, and in Generate Video if you want the one-off test, put those ids where the body holds
   avatar_id and voice_id.
7. In Generate Video1, replace "YOUR_SCRIPT_TEXT", quotation marks included, with this expression, so that the recap
   is what the avatar says:

{{ JSON.stringify($('Script Writer').item.json.output) }}

1. If you want HeyGen asked less often, open Wait and lengthen its pause (below).
2. Save, then press Test workflow.

The Wait nodes

30 Seconds stores an amount of 10 and no unit. This Wait node's default unit is seconds, so it waits 10
seconds before the first status check. Wait stores nothing and takes the default amount too: 5 seconds between checks.
A render takes a while, so 30 seconds there sends HeyGen far fewer requests.

The two paths in the file

• The loose nodes at the top, Get Avatars, Get Voices and Generate Video followed by Get Video, are not
  connected to the trigger. Run each by hand with Execute step. The first two list your avatars and voices, and the last
  pair sends one render request and asks for its status once.
• The main path starts at When clicking ‘Test workflow’, and goes through News, Script Writer, Generate Video1,
  30 Seconds and Get Video1. If then ends the run when the status reads completed. Otherwise Wait sends it back
  to Get Video1 for another look.

What it reads and sends

| Node | What it does |
|---|---|
| News | Posts to Apify's website-content-crawler, with https://www.morningbrew.com/issues/latest as its start address and markdown output |
| Script Writer | An agent on openai/gpt-4.1-mini through OpenRouter. It gets the page's text and answers with one spoken paragraph |
| Generate Video1 | Posts to HeyGen's v2/video/generate with an avatar, a voice, a speed of 1.1 and a size of 1280 by 720 |
| Get Video1 | Asks HeyGen's v1/video_status.get for the status of the video Generate Video1 started |

Change it

• Another source: the address in the body of News. The system message of Script Writer talks about Morning Brew
  issues, so edit it too.
• Length and tone: the system message of Script Writer. As exported it asks for one paragraph of 40 to 60 words
  covering the top two to four stories, with no intro, outro or call to action.
• Only the one page: add "maxCrawlPages": 1 to the body of News. The body sets no page or depth limit, so the
  crawler's own defaults apply.
• The look: avatar_id, avatar_style, voice_id, speed and dimension in Generate Video1.
• A daily run: put a Schedule Trigger in front of News. Nothing schedules the workflow as exported.

Good to know

• Each run spends HeyGen credits, Apify usage and a few OpenRouter tokens.
• As exported, Generate Video1 sends a fixed line, not the recap. Step 7 above is what connects them.
• Only completed ends the loop. A render that HeyGen reports as failed is checked again and again, so watch the first
  runs, and add a branch for failed if you automate it.
• The issue's text belongs to its publisher. Check Morning Brew's terms before you post a recap.
• The finished video is a link in HeyGen's status reply. Add a node on the true branch of If to download it, save it or post it.