AI Character Video Reel System

How We Built a System That Generates AI Character Video Reels From a Single Idea

AI Video Production | Character Creation | n8n + Airtable + fal.ai + ElevenLabs

Theme Idea Generator interface showing 10 generated theme concepts for BigMoves including ‘Secret Society of AI Elite’, status AI Complete, with the Idea Generator panel showing brand context, idea angle, and generated theme links

The Situation

The brief was to build a repeatable content engine for short-form video reels built around AI-generated fictional characters. The idea: instead of a talking head or screen recording, each reel would feature a cast of distinct AI characters with unique voices, faces, and personalities, acting out scripted scenes on AI topics. Shareable, entertaining, brand-consistent, and scalable without a production crew.

The target audience is small business owners and creators who are learning to use AI in their workflows. The content format needed to feel produced without requiring a production team to touch it between idea and published clip.

Brands view showing BigMoves as the brand configured in the system, with navigation showing Video Reel Generator, Theme Idea Generator, Themes List, and Prompt Library

The Problem

AI character video content of this type had no off-the-shelf production path. Creating it manually meant: generating character concepts by hand, prompting image models separately for each character, cloning or designing voices one at a time in a separate tool, writing scripts with character awareness, generating video assets per line of dialogue, sourcing or generating B-roll images at timestamp precision, composing custom music per theme, and stitching everything into a final video.

Doing that once for one video was a multi-day effort. The goal was to do it repeatedly, at volume, with a consistent output quality, and with the human in the loop only at decision points: approve a theme, approve characters, approve a script. Everything else automated.

There was no existing system for this. It had to be built from scratch.

The Solution

We built a 12-workflow n8n pipeline connected to an Airtable base (appMz4ODIt6Bm0JHz) that takes a one-line idea angle and produces every asset needed to assemble a finished AI character video reel: character profiles, faces, voices, multi-character scripts, per-line video clips, B-roll images with timestamped prompts, action audio clips, and a custom music track.

The Airtable base is the control center. It stores brand assets, theme configurations, character profiles, prompt libraries, AI model settings, generated assets, B-roll records, music, and a full usage tracker. The Airtable interface gives the human operator a clean step-by-step flow through the production pipeline without needing to touch n8n directly.

The 12 n8n workflows, in production order:

• Generate Ideas: takes an idea angle and brand context, generates up to 10 theme concepts via GPT-5.1, and creates theme records in Airtable
• Create Characters: generates full character profiles (name, personality, voice description, image prompt) for each character in the theme, then creates character images via fal.ai Gemini-3-Pro-Image-Preview and clones or designs voices via ElevenLabs API — all in a single workflow

• Create Character Image: standalone image regeneration for individual character portraits using fal.ai image-to-image with the approved character image prompt
• Create AI Voice: standalone voice creation and replacement — designs the voice spec, submits to ElevenLabs, stores the voice ID and preview MP3 back in Airtable
• Generate New Script: fetches the theme angle and all character profiles, builds a character-aware prompt, generates multi-character dialogue scripts via dynamic LLM, parses and creates individual script records

• Generate Video Assets: the heaviest workflow — maps each script line to its character, generates speech via ElevenLabs, submits the character image + audio to a video generation model, polls for completion, uploads to Cloudinary, and creates a generated asset record per chapter

• Generate New Chapter: regenerates a single chapter within an existing script — replaces just that asset without rerunning the full pipeline
• Generate Music: creates a BPM-matched, theme-specific music track by composing a plan, submitting to a music AI model, and uploading the result to Cloudinary
• Generate Action Assets: generates short character action clips (laughing, reacting) from prompt library entries — used for cutaway moments in the final video
• Generate B-Roll Image Style Prompt: analyzes the approved character image to extract a consistent visual style, then generates a B-roll style guide that keeps AI-generated scene images visually coherent with the character aesthetic
• Generate Timestamped B-Roll Image Prompts: takes the script and chunk size (e.g. every 5 seconds), generates a timestamped image prompt for each B-roll slot in the video
• Generate B-Roll Image: submits each timestamped prompt to fal.ai, polls for completion, and stores the B-roll image back against its timestamp record

Characters are built with full creative depth. Each profile includes a character angle (personality and speech style), a voice prompt used to design the ElevenLabs voice, and a detailed character image prompt specifying appearance, lighting, lens style, and aesthetic. Multiple image versions can be generated and the operator approves one.

Character Details view for ‘Dom | Secret Society of AI Elite’ showing character_angle (The Strategist), voice_prompt (flat emotionless American accent), and character_image_prompt (sharp-featured man in his 40s, bald, cinematic noir style)

Character Image tab for Dom showing three generated image versions (v1, v2, v3) with v2 marked Approved, and the approved portrait displayed — ultra-realistic elderly man in dark suit, Rembrandt lighting, leather chair background

Character Voice tab showing approved ElevenLabs voice ID, voice preview audio player (54-second preview), and Character Assets section listing generated video lines linked to script chapters

Scripts are written with full character awareness. The script generator fetches every approved character profile for the theme, builds a structured prompt with their names, angles, and voice styles, and generates multi-character dialogue where each line is tagged with the speaking character. Scripts can be regenerated with a new angle at any time.

Generate Scripts view for ‘Secret Society of AI Elite’ showing script_angle input, 3 scripts configured, 15-second max length, Claude Opus 4.5 model, and list of generated scripts including ‘Prompts Are A Distraction’ and ‘Why The Elite Laugh At Your AI Tool Collection’ with Published & Scheduled status

Script detail for ‘Prompts Are A Distraction’ showing full multi-character dialogue: Dom, Perci, Alfie, and Dr Elias exchanging lines about prompt obsession, ending with all four laughing together — status Published & Scheduled

B-roll generation is timestamp-precise. The operator sets a chunk size in seconds, the workflow splits the script into time blocks, generates an AI image prompt for each block that matches the visual style extracted from the character images, then submits each prompt to fal.ai. The result is a full set of scene images timed to the script.

Script generation detail view showing Timestamped B-Roll Image Prompts section with chunk_seconds set to 5, Generate Timestamped B-Roll Image Prompts button, and Generated Script Video Assets section showing v1 Chapter 1 | Dom with AI Writing status and character portrait

Theme Actions & Assets view showing prompt library (Laughing Evil Hysterically), character selection (Alfie), Create Action Asset button, and theme_assets panel listing generated action clips for Dom, Alfie, Dr Elias, and Perci — all awaiting approval

Music is composed per script. The system generates a composition plan based on the theme’s music prompt (BPM, instrumentation, tone), submits to a music AI model, extracts the BPM and duration from the result, and uploads to Cloudinary. The ‘Shadow Council’ track for Secret Society of AI Elite runs 3 minutes 30 seconds at 117 BPM.

Results

A one-line idea angle now produces a complete AI character video production package: cast of characters with faces and voices, multi-character script, per-line video clips, timestamped B-roll images, action cutaway clips, and a custom music track.

Multiple character universes were built and tested: Secret Society of AI Elite (4 characters, neo-noir boardroom aesthetic), Escape Pod Ops (Pixar-style rocket scientists), and others in the idea bank. Each theme is self-contained in Airtable with its own characters, assets, and production status.

Scripts for Secret Society of AI Elite reached Published and Scheduled status, with titles like ‘Prompts Are A Distraction’, ‘Why The Elite Laugh At Your AI Tool Collection’, and ‘Why We Love AI Deniers’. The system had generated 9 distinct scripts for that theme alone at the time of this writing.

The operator’s role in the process is approval only: approve a theme, approve character images, approve voices, approve a script angle, review generated assets. The system handles every generation step between those checkpoints.

Theme Angle view for ‘Secret Society of AI Elite’ showing Approved status, 4 characters configured, tone style (Neo-noir elite conspiracy with Guy Ritchie energy), setting description (private gentlemen’s club), and character appearance style (ultra-realistic cinematic portraits, Rembrandt lighting, 8k detail)

Character Profiles tab showing all 4 approved characters — Dom, Alfie, Dr Elias, and Perci — each with Approved status and character portrait thumbnails

Prompt Library view showing action audio prompts: Laughing Evil Hysterically, Hot Girl Laugh, Chuckles In Rememberence, A Soft Chuckle — each with the exact ElevenLabs voice direction text

How It Worked

Before: video content with original characters required a writer, a visual artist, a voice actor, a video editor, and a music composer. One episode could take days or weeks. Scaling meant hiring.

After: one operator, one Airtable interface, a series of button clicks at approval checkpoints. The pipeline handles ideation through asset generation without creative production staff.

1. Idea in. Operator enters a theme angle into the Theme Idea Generator. GPT-5.1 returns up to 10 theme concepts. Operator approves one and it becomes a Theme record.
2. Characters built. Operator clicks Create Characters on the Theme. The workflow generates character profiles, submits each image prompt to fal.ai, designs the voice spec, and clones or creates the voice in ElevenLabs. All assets stored in Airtable. Operator reviews and approves images and voices.
3. Script generated. Operator sets a script angle, picks an AI model (Claude Opus 4.5 was used for Secret Society scripts), and clicks Generate New Script. The workflow builds a character-aware prompt from all approved profiles and returns multi-character dialogue. Operator approves or iterates.
4. Video assets produced. Operator clicks Generate Video Assets. The workflow maps each script line to the right character, generates speech via ElevenLabs, submits to fal.ai video generation, polls for completion, uploads to Cloudinary, and creates a record per clip. Chapters can be regenerated individually if one needs a redo.
5. B-roll and music. Operator generates the B-roll style prompt (extracts visual DNA from character images), then generates timestamped image prompts at 5-second intervals, then submits each to fal.ai. Separately, Generate Music produces a BPM-matched track from the theme’s music brief.
6. Action assets. Operator selects a prompt from the library (e.g. Laughing Evil Hysterically), picks character targets, and clicks Create Action Asset. The workflow generates a short audio-video clip per character for use as cutaway reactions in editing.

B-Roll Style tab showing Generate B-Roll Style Prompt button and Theme Actions & Assets section for the Secret Society theme

Theme Music tab for ‘Secret Society of AI Elite’ showing music_title (Shadow Council) and the detailed music_prompt: dark cinematic score, 70-80 BPM, plucky strings, upright bass, muted piano, harpsichord flourishes, Guy Ritchie crime film meets neo-noir chamber music

Want a System Like This?

If you’re producing video content and want to move from manual production to a system that handles the heavy lifting, this is the kind of infrastructure we build.

Character-driven AI video, social content automation, workflow architecture for creators and agencies.