VideoVulcan, built by a Snippbot Swarm

The flagship movie-maker was scoped into a tight FTUE and a traceable AI pipeline.

3-8 photos per movie

Users upload a handful of their own photos to cast the movie.

1 line to direct it

One sentence is enough to set the story.

3 style presets

Epic, Noir and Cartoon looks at launch.

• Locked the simple user promise: photos plus one line become a short movie, not a general video editor.
• Chose stylize-first portraits so the making-of screen reveals meaningful cast art before render completion.
• Kept Google sign-in, Mints, watermarking, deletion, and Play docs in scope for an internal-testing release.

The plan divided the app into auth, ledger, render, audio, stitch, store, and release lanes.

90 tasks planned

Sign-in, credits, rendering, audio, store and release, split into owned tasks.

5 specialist agents

App, backend, pricing, art and launch each had their own agent.

• Tasks covered Google Sign-In, Mint ledger, RevenueCat mapping, Claude moderation/script, Gemini stylize-first, Seedance render, Hume audio, stitch, and Play submission.
• Dependencies kept entitlement, render, and release work tied to the right upstream contracts.
• Our developers owned product, pricing, store, legal, and spend decisions while the swarm executed the plan.

Focused agents built the client, backend, AI workflow, economy, and release assets in parallel.

751 automated tests passing

Sign-in, Mint credits, moderation, rendering and deletion are checked on every build.

100% movies watermarked

Every exported movie carries a visible AI-generated label.

• Landed onboarding, create flow, Google Sign-In hooks, Mint ledger, premiere screen, share/save affordances, and account deletion.
• Backend work connected Claude moderation and script planning, Gemini image stylization, Seedance video rendering, Hume audio texture, and the stitch stage.
• The stitch stage burns in the VideoVulcan AI-generated mark, adds MP4/XMP provenance, appends the end card, and strips source metadata.

The app moved from isolated work streams into one integration branch for review.

86/86 tasks merged

Every completed task landed in one shared app build.

10 branches merged

Each agent and reviewer worked on its own branch, then merged into the app.

188 agent work sessions

Short, focused runs: one agent, one task at a time.

• Client, backend, monetization, store-art, legal, and release evidence landed together for whole-app checks.
• Integration preserved the finished work while surfacing dependency gaps for owners instead of hiding them in chat.
• Follow-up commits from QA and E2E loops were reintegrated as the Android release build matured.

Independent review checked the release promises against the implemented app and backend.

81 issues caught before release

Reviewers filed each with evidence and routed it to the owning agent.

• Review covered auth, Mint grants, render contracts, watermark/provenance, deletion, Play docs, and release-build evidence.
• The stylize-first work passed 79 focused tests, then the backend suite passed 717 tests in the integrated codebase.
• Findings were routed back to owners as concrete work instead of becoming a single overloaded agent’s unchecked TODO list.

A release-configured Android build was installed and walked through the first-movie flow.

2 end-to-end test rounds

The signed release build was installed and walked through the first-movie flow.

0 crashes in the final test

Launch, sign-in and the create flow ran cleanly on version 3.

• The emulator walkthrough covered onboarding, create, first Mint, making-of, premiere, and Android share sheet.
• The run confirmed the Hermes Intl.PluralRules crash stayed fixed in a release-configured build.
• The share sheet received a real MP4 named for the VideoVulcan movie, proving the client produced a file rather than only in-memory state.

The swarm prepared the Android package, store surfaces, and public proof for internal testing.

v3 live on Google Play

Testers install it from Google Play internal testing.

5 days kickoff to version 3

The swarm started Sept 23; version 3 reached testers Sept 28.

• VideoVulcan was prepared for Google Play internal testing with launcher icon, feature graphic, Play docs, data safety, and deletion flow.
• The project page uses the real store feature graphic rather than an internal placeholder reel.
• Our developers made the product, pricing, store, legal, and spend decisions; the swarm did the rest.