We do not make videos. Especially not those ones.
HeyGen generates video from a script you give it: synthetic presenters, voice, translation into other languages. It is a video tool and this is not one. The more useful difference is what each of them starts from.
The short answer
Different medium, different job
A video generator needs a script. Someone has to decide what the video says, which is the same missing step a scheduler assumes you have already taken. Merge & Tell starts one step earlier than both: it reads the pull request you merged and works out what changed for the people who use your product, and the post comes out of that.
So there is no feature overlap to weigh up. Text posts to your own social accounts and a changelog feed, out of your git history. No video, no avatars, no voice.
Side by side
The input is the whole difference
| An AI video tool | Merge & Tell | |
|---|---|---|
| Input | A script you write or generate | A merged pull request and its diff |
| Output | Video, with a synthetic presenter and voice | Text posts, sized per network, plus an Atom feed |
| Where the facts come from | Whatever the script says | Code that shipped, with the PR linked from the entry |
| Scaling it up means | More renders | Nothing. Volume is capped by what you actually merged |
That last row is the design, not a limitation we are apologising for. The number of things worth announcing is set by how much you shipped, and a tool that can exceed it is a tool that will.
The position
The genre we are not building for
You have seen these. A face that does not belong to anyone, reading a script nobody wrote, over stock footage of a city at dusk, with burned-in captions bouncing a word at a time, posted eleven times a day. Two years ago it was novel. Now it is wallpaper, and the comments under it are people identifying which tool made it.
The tools that render it are not the problem, and this page is not about them. The practice is. Short-form at that volume is a bet that attention is a numbers game and that nobody checks, and for a developer audience both halves of that bet are wrong. Developers pattern-match generated video in about a second and a half, and what they conclude is not this product is popular, it is nobody is home.
There is a version of that failure that applies to us too, and it is worth naming since we also generate copy: a machine writing forty posts about nothing is the same failure in a cheaper medium. Which is why the filter matters more than the generator here.
The alternative
Every post traces back to a merge
The claim we want a reader of your feed to be able to make is that something behind the post really happened. So we built the machinery to keep that true rather than to increase output:
- The source is a diff. Drafting reads the pull request and the code that changed, not a topic you picked.
- Most merges never become posts. Bots, dependency bumps, noise-only diffs and changes under ten meaningful lines are dropped before a draft exists. The filter ladder.
- A human approves. Nothing goes out until you say so, unless you switch a repo to autopost yourself. Copy written in a person’s voice never auto-publishes at all, whatever the repo says.
- You can see how often it was worth keeping. The kept-rate is the share of drafts you published rather than discarded, and it is on a screen so a bad run is visible. How that is counted.
None of that is compatible with posting eleven times a day, which is the point.
Not us
When a video tool is the right answer
Video is not the problem and generated video is not automatically slop. Reach for a tool like HeyGen when:
- A human wrote the script, means it, and would happily say it on camera if a camera were available.
- You need the same real demo in six languages and cannot re-record it six times.
- It is onboarding, training or documentation, where a presenter is a format convention rather than a claim about a person.
- Your audience is on a network where video is the only unit of content that travels.
What separates those from the genre above is that a person is accountable for what was said. That is the same test we apply to our own output, which is why a human has to press publish.
Roadmap, in the negative
Things we are not going to ship
- Video generation, avatars, or synthetic voice.
- Posting to TikTok, Reels or Shorts. Not a capacity problem, a deliberate absence.
- Autoposting under a person’s name. The person-voice hold is in the code, not in a setting.
- A posting target, a streak counter, or anything else that rewards volume over having shipped something.
- Copy about a merge that did not happen. There is no generate-a-post button that works without a pull request behind it. The one hand-written path, the composer, is you typing.
Post the things you actually built.
You were going to merge the PR anyway.