Building MVFlow: Turning Music into an Editable AI Video Workflow

Description and use

MVFlow builds a reliable music video workflow with validated plans, checkpoints, timing integrity, and durable recovery.

Problem

Designing a robust, editable AI video workflow that handles music, creative direction, and media generation while ensuring validation, checkpointing, timing integrity, and recoverability.

Outcome

MVFlow now reliably transforms music and creative input into a finished video, with validated plans, checkpointed generations, continuous shot timelines, durable recovery, and optimized composition.

Workflow

• 1. Provide audio and visual description with optional reference images.
• 2. Generate sections and shot timing.
• 3. Create a creative plan and review it for semantic validation.
• 4. Generate reference images and shot videos.
• 5. Select shot versions.
• 6. Compose final video with FFmpeg, ensuring continuity and compatibility.
• 7. Persist results and handle retries via checkpoints and leases.

Evidence

The workflow starts with audio and a description of the visual world you want to create. You can add reference images for characters, locations, or style, review a timed storyboard, generate different shot versions, and choose which versions appear in the final composition.

Source

MVFlow establishes section and shot timing before generating the complete creative plan. A separate review stage receives the source context and the generated plan, then returns a complete corrected plan.

Source

The generation result is saved before review begins. If review fails, the application can reuse that draft when its checkpoint still matches the current inputs.

Source

The project stores timeline boundaries on a hundredth-of-a-second grid and checks continuity before composition.

Source

Video generation can outlive a browser session or an individual HTTP request. The recovery design uses persisted project state, queued execution, and database leases.

Source

FFmpeg composition can sometimes reuse encoded video streams through stream copy, avoiding another encoding pass.

Source