Skip navigation
I've been using Claude to produce videos walkthroughs for pull requests. Here's how it works:
0:00 / 4:51

I've been using Claude to produce videos walkthroughs for pull requests. Here's how it works:

Steve Ruiz

@steveruizok61K followers

Visit on X
261 likesDownloadOriginal post

27,884 views

I've been using Claude to produce videos walkthroughs for pull requests. Here's how it works:

33 Comments

Comments from the original post on X (Twitter)

  • Thariq's profile picture

    @trq212

    Now add subway surfers

    34 likes

  • Steve Ruiz's profile picture

    @steveruizok

    asmr release notes

    2 likes

  • Steve Ruiz's profile picture

    @steveruizok

    Starting from a GitHub pull request, the system first gathers context: it reads the PR title and description, commits, and final diff, and snoops around the repo as needed.

    11 likes

  • Steve Ruiz's profile picture

    @steveruizok

    Once it has enough information, it moves onto the second step: writing a script with an intro, context, implementation, and summary.

    10 likes

  • Steve Ruiz's profile picture

    @steveruizok

    Next, it sends a cleaned-up version of the script to Google for a TTS run. I use Chirp 3 here because it can handle longer generations. We generate all the text as a single audio file, with pauses indicating slide transitions.

    11 likes

  • Steve Ruiz's profile picture

    @steveruizok

    When we have the audio file, we then chop it up based on the detection of pauses, measure the length of each section, and generate slides. I have a few templates for intro, text, or code / diffs. These are just HTML files. We record animations or take screenshots.

    7 likes

  • Steve Ruiz's profile picture

    @steveruizok

    At the end, we combine the slides and audio, add back in pauses between slides, and put them all together. The resulting video takes about 10 minutes to generate. Operational, non-load-bearing slop, anchored in code, perfect for dropping into PR descriptions.

    6 likes

  • Steve Ruiz's profile picture

    @steveruizok

    What I'm still working on: - the design of the slides - using the tldraw canvas itself to display changes - taking screenshots or recordings of before and after using release/preview branches - reliable subtitles

    10 likes

  • Steve Ruiz's profile picture

    @steveruizok

    One more for bonus content:

    10 likes

  • Steve Ruiz's profile picture

    @steveruizok

    Any ideas to make these better?

    7 likes

  • Justin Bennett's profile picture

    @just_be_dev

    Man, I feel like you need to productionize and sell some of the operational shit you’re doing. It’s better than most the startups who are focused on these problems in the space.

    6 likes

  • Steve Ruiz's profile picture

    @steveruizok

    You don't know the half of it! I need an obfuscation feature on the bizops stuff we have so I can share it

    3 likes

  • Grace Yun's profile picture

    @jueungraceyun

    9:16 brain rot version pls

    4 likes

  • Steve Ruiz's profile picture

    @steveruizok

    Adam voice inbound

  • Varadh Jain's profile picture

    @varadh

    I love tha idea Could it do with half the words?

    4 likes

  • Greg Konush (。◕‿◕。)'s profile picture

    @GregKonush

    you can use react remotion for better presentation

    3 likes

  • brandon toups's profile picture

    @abrandoncdefg

    Can you add Subway Surfer so I can focus plz🙏

    1 like

  • Mustafa's profile picture

    @mustafa_2vec

    Brainrot for nerds

    1 like

  • dorian's profile picture

    @dorianmariecom

    not a great example, a video of a demo of the feature would be useful

    1 like

  • Steve Ruiz's profile picture

    @steveruizok

    Yeah, I'd like to see some demos too, though tbh the PR description usually has that. This would be a "take me through the code" resource

    2 likes