AI Video Automation Platform – Script-to-Video Pipeline (Avatar + Stock Footage + Motion Graphics)

Overview I run several faceless YouTube channels (long-form, narrated documentary/story-style content, 2,000–4,500 word scripts / 15–25 min runtime). I already produce scripts and currently work with human editors who assemble everything into the finished video. I'm looking for a developer to build a platform that takes a finished script and automatically assembles it into a publish-ready edited video — replacing that manual editing step. What the platform needs to do Script input — paste or upload a plain-text script. Automatic scene breakdown — split the script into scenes/beats with visual cues via an LLM (timestamps + b-roll keywords), and flag which segments are "avatar on-screen" vs. "cutaway/b-roll" per my standard structure (avatar open → b-roll/AI-image body → optional split-screen → avatar close). Avatar narration via HeyGen — the tool calls the HeyGen API with the script and gets back a finished MP4 with the avatar already speaking (voice + visuals baked in). No separate TTS/voice pipeline needed — HeyGen already handles that part. Supporting visuals for the cutaway/b-roll portions — a mix of: AI-generated images/animated stills (I already have a working script-to-image pipeline using Replicate/ComfyUI — the new platform should integrate with or replace this, not duplicate it from scratch) Matching stock b-roll pulled automatically via API (Pexels/Pixabay/Storyblocks) Auto-assembly (the core of this job) — take the HeyGen avatar MP4 as the base track and cut in the b-roll/AI images, captions, and light motion graphics (pans/zooms, text overlays, transitions) on top of it, following the scene breakdown from step 2, via FFmpeg or Remotion. Channel presets — I run each channel from a JSON style profile (visual style, pacing, color palette). The tool should read these presets so output automatically matches the right channel's look. Simple dashboard — upload script → pick channel preset → render → download. A render queue with status tracking is enough. This is an internal tool for my own use, not a customer-facing product — no need for multi-user accounts or billing. Non-negotiables This is primarily a video-editing/assembly automation job, not an avatar-generation job — HeyGen already produces the narrated avatar footage. Keep the added cost/complexity of the editing layer itself as low as possible. The pipeline should run end-to-end without manual editing for the large majority of a typical video. - https://www.youtube.com/@EliasYoderAmish - https://www.youtube.com/@DiscoverywithAmbrose - https://www.youtube.com/@SilasMercerYT What to include in your application -Similar automated video-editing/assembly projects you've built (links/portfolio), ideally involving HeyGen or a similar avatar API -Your proposed tech stack for the assembly layer (b-roll/image sourcing, FFmpeg/Remotion assembly) and why -Estimated additional cost per finished video for the editing layer (on top of HeyGen's own per-minute cost) -Timeline and price estimate for a working MVP Whether you'd recommend a small paid test task before the full build