End-to-End AI Video Production Pipeline

The challenge

Turning a written script into a finished, narrated, subtitled video normally means hours of manual storyboarding, image sourcing, voiceover recording, and editing for every single video.

The process also has to cope with very different lengths — from a 1-minute short to a 40-minute long-form video — without a person re-planning the whole pipeline each time.

The solution

A two-workflow n8n pipeline takes a script straight through AI storyboarding, image generation, voiceover, and FFmpeg assembly with no manual editing in between.

Gemini handles storyboarding, Kie.ai generates scene imagery, ElevenLabs produces the voiceover, and FFmpeg assembles the final 1080p cut with subtitles and motion.

Implementation

• Built the pipeline to scale dynamically so it handles both 1-minute and 40-minute videos from the same workflow.
• Added automatic image policy correction and missing-scene retries so a single bad generation does not break the run.
• Assembled output at the scene level — voiceover, subtitles, Ken Burns motion, and a final concat into one delivered file via Google Drive.

Results

• A written script becomes a narrated, subtitled, edited 1080p video with no manual editing step.
• Variable-length videos are produced from one pipeline rather than per-video manual planning.
• The system runs live and produces videos end to end.

Impact

• Handles 1-minute videos and 40-minute videos dynamically
• Automatic image policy correction and missing-scene retries
• Scene-level voiceover, subtitles, Ken Burns motion, and final concat