Diffvideo/Docs

Video generation from your repo

Diffvideo reads your code changes and repository context, then produces a branded video that matches the video type you configured. This page covers what context matters, how the output is shaped, and how to improve quality without changing your workflow.

What Diffvideo reads

When a trigger fires (PR merge, release, push, or schedule), Diffvideo uses the following context:

  • Code diffs — The full diff of changed files, including additions, deletions, and modifications.
  • Commit messages — All commit messages associated with the change, providing intent and context.
  • PR description — The pull request title and body text, which often contains the most useful context about why a change was made.
  • File paths — The structure of changed files, which helps interpret which parts of the codebase are affected.
  • Repository metadata — The repository name, language, and description for additional context.

How output is structured

Diffvideo produces structured content whose shape depends on your video type. A typical changelog or update includes:

  • Opening — A brief introduction that sets the context (what product, what changed).
  • Key changes — The most significant changes, explained in order of user impact. Each change gets a title and description.
  • Summary — A closing statement that ties the changes together and suggests what viewers should do next.

Getting better results

Output quality tracks closely with the quality of repository context. The most effective improvements are usually the simplest:

Write descriptive PR descriptions

The PR description is the highest-signal input. A PR with the description "fix bug" will produce a vague result. A PR that explains what was broken, why it happened, and how the fix works will produce a detailed, accurate video.

Use meaningful commit messages

When PR descriptions are sparse, Diffvideo leans more on commit messages. Conventional commits (feat:, fix:, docs:) are particularly helpful because they signal the type and scope of each change.

Choose the right video type

The video type sets tone and focus. A changelog video highlights user-facing changes; a feature demo goes deep on a single capability. Choosing the right type keeps the story aligned with the audience.

In suggest mode, you can review and edit drafts before rendering. Use that pass to correct tone, tighten wording, or add context Diffvideo could not infer from the repo alone.

When output is weaker

Diffvideo works best with well-documented changes. You may see thinner output when:

  • PRs have no description and generic commit messages like "updates" or "wip"
  • Changes are purely infrastructure-level with no user-facing impact to describe
  • The diff is very large (1000+ lines) — the summary will favor the headline changes over every line
  • Changes are in languages or frameworks with sparse public examples

For these cases, suggest mode lets you catch issues before a video is rendered.

Next steps