Shorts / Reels Generator — Local MVP Phase 1 Complete

June 7, 2026

ShortsExperimentLocalMVPAICollaborationClaudeCodeVibeCodingFFmpeg

Summary

Working with Claude Code, I built a local MVP that converts photos and short clips into 9:16 vertical MP4s. Features expanded incrementally from v0.1.0 through v0.1.4, and I judged the project complete at Phase 1. This is not a public service release — it runs locally — and the core of this record is not the app itself but the process of building an MVP with AI while the operator owns direction, validation, and completion decisions.

Why It Started

Working on shorts and reels content, I needed a tool to quickly format photos or short clips into 9:16 vertical video. Using a public service raised data exposure concerns, and for repeated experiments a locally-run tool was a better fit.

Two reasons drove starting as a local MVP rather than a public service. First, I needed to validate whether the core features were actually useful by using them myself. Second, before thinking about deployment, I needed a clear judgment on how far to go in Phase 1.

Role Separation

Roles were clearly divided throughout this project.

Operator: Decided which features to prioritize, how to validate them, and what “done” means — confirmed the output behaved as intended
Claude Code: Designed the technical structure and wrote the implementation

This doesn’t mean the operator never touched the code. It means the operator owned direction and judgment; Claude Code owned implementation.

Technical Structure

Frontend: Next.js (UI controls and input)
Backend: FastAPI (Python, video processing API)
Video processing: FFmpeg (compositing, crop, captions, transitions)
Image processing: Pillow (intro/outro card generation)
Execution: Local dev servers (Next.js + FastAPI running in parallel)

Phase-by-Phase Expansion: v0.1.0 to v0.1.4

v0.1.0 — Core Structure

Photo upload, 9:16 MP4 generation, basic UI wiring. The goal was to confirm that the single most important feature actually worked before adding anything else.

v0.1.1 — BGM and Ken Burns

Added BGM file upload and volume control. Added Ken Burns motion (slow zoom and pan) to give static photos a sense of movement.

v0.1.2 — Transitions and Multi-Scene

Added scene transition effects (fade, slide, cut, etc.). Implemented multi-scene composition — combining multiple scenes into one MP4.

v0.1.3 — Intro/Outro and Captions

Used Pillow to generate intro and outro cards. Added per-scene caption text with controls for color, position, and style.

v0.1.4 — Workout Logging and Final Features

Added workout record input (distance, time, pace, etc.). Added card preview and custom download filename support, completing Phase 1.

Current Features

Photo / video upload
9:16 vertical MP4 generation
BGM upload, selection, and volume control
Ken Burns motion
Transition effects
Intro / outro cards
Per-scene captions
Workout record input
Text color, position, and style controls
Card preview
Custom download filename

Documentation Completed

Document	Contents
README.md	Project overview and how to run
CHANGELOG.md	Change history from v0.1.0 to v0.1.4
docs/LOCAL_RUN.md	Local execution procedure
docs/FEATURES.md	Feature list and usage
docs/KNOWN_ISSUES.md	Known issues and priority
docs/ROADMAP.md	Phase 2 candidate features

Phase 1 Completion Criteria

Phase 1 was judged complete when three criteria were met.

The core goal — 9:16 vertical MP4 generation — actually works.
The minimum editing features needed for shorts production are in place.
Documentation is complete and the project can be reproduced locally.

This is not “finished product.” It is “a defined point where the operator can make a judgment call and stop.” Knowing when to stop is harder than adding features, and more important.

Deferred to Phase 2

The following were classified as Phase 2 holdovers. They will be reviewed separately once Phase 1 features have been used enough to confirm real need.

Drag-based scene editing
AI-assisted automatic scene analysis
OCR (automatic text recognition)
Cloud deployment and user-saved sessions
Multi-user support

What This Log Means

This is not a “we built a shorts generator” announcement. It is an operating record: how AI collaboration was used to build a local MVP, and how the operator made direction, validation, and completion decisions throughout.

That record is what makes the next MVP cheaper. Knowing where to start and where to stop becomes a repeatable skill.