CL2VID

• 

  The face consistency is protected by Higgsfield Character

• 

  Upload a link to changelog, or Changelog.md file to generate a scenario & video

• 

  Create a Higgsfield Character by uploading ~20 pictures

Inspiration

We wanted to create professional-looking product update videos, but it always required too much work. We don’t have a proper studio setup, and speaking clearly in English was tough for us. Normally, we’d only make one update video every month or so, but I wanted to create one for every changelog, to share and promote on social media easily.

What it does

CL2VID converts a CHANGELOG.md or release notes into a cinematic 30-second product update video. It uses a talking-head style with your own face, English narration, and auto captions. The output is ready to attach to a GitHub PR or Release.

How we built it

By creating a personalized Character in Higgsfield from the client app (powered by Rork) , we achieved a high-quality, realistic video where it looks like the same person (you) is presenting every time.

Backend is build with Hono.

Challenges we ran into

We initially tried to generate the entire video using Higgsfield, but its duration limit wasn’t enough for a 30-second update. We ended up using Higgsfield for character and image generation, and then composing the final video in Veo 3. The results were FAR BETTER than using Veo 3 alone — character consistency and realism were essential for this use case.

Rendering time was another challenge, but by making the client a mobile app, we solved it through push notifications and background job tracking.

Accomplishments that we're proud of

First of all, we're super happy that we build a tool that we'd love to use weekly, only in 4 hours!

• End-to-end flow from raw changelog to a polished, on-brand 30-second video.
• Reliable talking-head output that actually looks like the same person across runs.

What we learned

• Quality depends on character consistency more than just animation quality
• Time constraints (30s limit) force clarity — short videos are more watchable
• Combining tools pragmatically (Higgsfield + Veo) is better than using one stack

What's next for CL2VID

• Use TTS with voice cloning so the narration sounds like the user while speaking fluent English.

• Ship a GitHub App. Comment @CL2VID on a changelog pull request to generate multiple video variants and post them back as a PR. Allow reviewers to request changes from comments and regenerate on demand. (We didn't go with this interface because we wanted to use Rork this time :))

Built With

• higgsfield
• hono
• rork
• veo3