Automatic Watermark Removal for AI-Generated Video, at Scale
Built a self-hosted watermark removal service (FastAPI, Florence-2 + LaMa/IOPaint) integrated into an n8n content pipeline. Two detection modes: general zero-shot detection for any AI video source, and a purpose-built fast path for a specific generator's known watermark position, removing the detection step entirely for a 10-15x speed gain. CPU-only, async job queue, webhook callbacks, automatic fallback if the primary method fails. Fully integrated with Google Drive for hands-off file handling, no manual uploads or downloads at any step.

Snapshot: Self-hosted API + n8n integration. Stage 2 of a full AI video production pipeline (generate → remove watermark → combine → caption → post).
Challenge: AI-generated video comes with the generator's watermark baked in. At any real volume, removing it by hand, or one video at a time through a web tool, becomes the bottleneck sitting right after generation.
Solution: A dedicated FastAPI service with two removal paths: general-purpose zero-shot detection (Florence-2 + LaMa) that works on watermarks from any AI video tool without tuning, and a purpose-built fast path for a known, static watermark position that skips detection entirely, cutting processing to roughly 7% of the frame and 10-15x the speed. CPU-only. If the primary method fails for any reason, a fallback path still produces usable output, so a job never dies mid-queue. n8n owns the full file lifecycle: pulls from a Google Drive queue, sends the job, waits for a webhook callback, uploads the clean result, deletes the processed original.
Result: A manual cleanup step that used to require opening a website per video, uploading, waiting, and downloading, now runs unattended as one stage in a larger pipeline.
Want this built for your production pipeline? [Get in touch → [email protected]]
Want something like this for your business?
Tell us what you'd automate — a message is enough to start. We'll tell you straight whether it's worth building.
Start a conversation