n8n Video Workflow Debugging

We have a working n8n workflow that generates 24-second vertical marketing videos using FAL.ai, Google Gemini agents, ElevenLabs, and Cloudinary. A full technical audit has already been completed — the problems are identified and documented. We need an experienced n8n specialist with hands-on knowledge of AI video generation APIs to implement the fixes. The audit identified the following issues: Bridge frame captured at wrong timestamp causing repeated frames at splice points between segments Sanitisation function corrupting the negative prompt constraints No character anchor — character identity only carried as free text, causing drift between segments Image-to-video model treating reference frame as starting frame instead of identity reference 3 independent audio tracks per segment causing voice inconsistency QA gates never inspecting real prompts or real video output No background music or CTA step in the workflow Broken retry paths and dead logging nodes What a good fix looks like: Character appearance stays consistent across all 3 video segments No freeze or repeated frames at the 8s and 16s transition points Single consistent voice across the full 24-second video Background music present from start to finish CTA visible at the end of every video Workflow runs end-to-end without manual intervention Tech stack: n8n Cloud, FAL.ai (VEO 3.1 / WAN 2.6 / Seedance 2.0), Google Gemini, ElevenLabs, Cloudinary, Google Drive (knowledge bases)