A local-first short-video production studio for experimenting with multi-LLM script pipelines and local media generation.

The app can generate a short-video project from a topic, route scenario creation through different LLMs, create styled image prompts, generate images locally with AUTOMATIC1111 Stable Diffusion WebUI, generate voiceover with Edge TTS, create subtitles, and render a vertical video with FFmpeg.

This project is designed as a practical local tool: run it on your own Windows machine, keep API keys in .env, and keep generated media out of GitHub.

Topic + category + language + style
  ↓
Stage 1 LLM — first draft
  ↓
Stage 2 LLM — script doctor
  ↓
Stage 3 LLM — final JSON script
  ↓
Image prompt generator
  ↓
AUTOMATIC1111 Stable Diffusion API
  ↓
Edge TTS voiceover
  ↓
Subtitles
  ↓
FFmpeg
  ↓
Final vertical video

• Web UI for generating and editing video packs.
• Four genre prompt packs:
  • Horror
  • Facts
  • Kids
  • Documentary
• Flexible LLM routing:
  • Stage 1 can use Ollama, Groq, or Mistral.
  • Stage 2 can use Ollama, Groq, or Mistral.
  • Stage 3 can use Ollama, Groq, or Mistral.
• Ollama models are detected automatically from local Ollama.
• Groq and Mistral model lists can be extended in .env.
• Image style presets:
  • pixar
  • cartoon
  • anime
  • realism
  • comic
  • watercolor
  • clay
  • dark_cinematic
• Mass input field for adding recurring character/world details to all image prompts.
• Prompt Pack Editor in the browser.
• Stable Diffusion settings in the browser:
  • width
  • height
  • steps
  • CFG scale
  • sampler
• Local image generation through AUTOMATIC1111 API.
• Local-ish voiceover through Edge TTS.
• Subtitle rendering.
• FFmpeg final video rendering.
• Backup/restore per project.
• Safe GitHub setup: .env, generated projects, music, and media files are ignored.

Use this folder structure:

C:\ai\
  multi-ai-video-factory\
  stable-diffusion-webui\

This README assumes the app will be installed here:

C:\ai\multi-ai-video-factory

You need:

• Windows 10/11
• NVIDIA GPU recommended
• Python 3.10.x recommended
• Git
• FFmpeg
• Ollama
• AUTOMATIC1111 Stable Diffusion WebUI
• At least one local Ollama model, recommended: qwen2.5:14b
• Optional cloud API keys:
  • Groq
  • Mistral

Recommended: Python 3.10.x.

Download Python from:

https://www.python.org/downloads/windows/

During installation, enable:

Add Python to PATH

Verify in PowerShell:

python --version
pip --version

Download Git for Windows:

https://git-scm.com/download/win

Verify:

git --version

Option A — winget:

winget install -e --id Gyan.FFmpeg

Close and reopen PowerShell, then verify:

ffmpeg -version
ffprobe -version

Option B — manual:

Download a Windows build from:

https://ffmpeg.org/download.html

Then add the bin folder to your Windows PATH.

Install Ollama from:

https://ollama.com/download/windows

After installation, open PowerShell and pull Qwen:

ollama pull qwen2.5:14b

You can also use a lighter model:

ollama pull qwen2.5:7b

Verify:

ollama list
ollama ps

Test the API:

$body = @{
    model = "qwen2.5:14b"
    prompt = "Hello"
    stream = $false
} | ConvertTo-Json

Invoke-RestMethod `
    -Uri "http://localhost:11434/api/generate" `
    -Method Post `
    -ContentType "application/json" `
    -Body $body

You should see a response from the model.

Go to:

C:\ai

Clone the repository:

cd C:\ai
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
cd C:\ai\stable-diffusion-webui

Edit:

webui-user.bat

Find:

set COMMANDLINE_ARGS=

Change it to:

set COMMANDLINE_ARGS=--api

Optional for NVIDIA GPUs, if your setup supports it:

set COMMANDLINE_ARGS=--api --xformers

Now run:

.\webui-user.bat

First launch can take a long time because it downloads dependencies.

When it starts, open:

http://127.0.0.1:7860

Verify the API docs:

http://127.0.0.1:7860/docs

The app uses:

A1111_BASE_URL=http://127.0.0.1:7860

Clone or download this repository.

Recommended location:

cd C:\ai
git clone https://github.com/YOUR_USERNAME/multi-ai-video-factory.git
cd C:\ai\multi-ai-video-factory

Or if you downloaded ZIP:

Extract it to:
C:\ai\multi-ai-video-factory

Allow PowerShell scripts for this window:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass

Optional prerequisite check:

.\scripts\check-prereqs.ps1

Run setup:

.\scripts\setup.ps1

This will:

• create .venv
• install Python requirements
• create .env from .env.example if missing

Open:

notepad .env

Minimal local-first config:

GROQ_API_KEY=
MISTRAL_API_KEY=

OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=qwen2.5:14b

A1111_BASE_URL=http://127.0.0.1:7860

GROQ_MODEL=llama-3.3-70b-versatile
MISTRAL_MODEL=mistral-small-latest

GROQ_AVAILABLE_MODELS=llama-3.3-70b-versatile
MISTRAL_AVAILABLE_MODELS=mistral-small-latest,open-mistral-nemo

SD_WIDTH=512
SD_HEIGHT=768
SD_STEPS=28
SD_CFG=7
SD_SAMPLER=Euler a

MUSIC_DIR=C:/ai/multi-ai-video-factory/music
PROJECTS_DIR=C:/ai/multi-ai-video-factory/projects

If you want Groq or Mistral in the dropdowns, add real API keys:

GROQ_API_KEY=your_groq_key_here
MISTRAL_API_KEY=your_mistral_key_here

Never commit .env.

Start the app:

cd C:\ai\multi-ai-video-factory
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
.\scripts\run-web.ps1

Open:

http://127.0.0.1:8001

Start with a simple test:

Topic:
A mysterious signal from an abandoned satellite

Category:
Documentary

Language:
English

Image style:
realism

Stage 1:
OLLAMA — qwen2.5:14b

Stage 2:
OLLAMA — qwen2.5:14b

Stage 3:
OLLAMA — qwen2.5:14b

Disable image generation for the very first test if you only want to test the LLM pipeline.

Then test with images enabled once AUTOMATIC1111 is running.

Stage 1: ollama:qwen2.5:14b
Stage 2: ollama:qwen2.5:14b
Stage 3: ollama:qwen2.5:14b

Stage 1: ollama:qwen2.5:14b
Stage 2: groq:llama-3.3-70b-versatile
Stage 3: mistral:mistral-small-latest

Stage 1: groq:llama-3.3-70b-versatile
Stage 2: groq:llama-3.3-70b-versatile
Stage 3: mistral:mistral-small-latest

Install another model:

ollama pull qwen2.5:7b
ollama pull llama3.1:8b
ollama pull mistral:7b

Restart the web app. They should appear automatically.

Edit .env:

GROQ_AVAILABLE_MODELS=llama-3.3-70b-versatile,another-groq-model
MISTRAL_AVAILABLE_MODELS=mistral-small-latest,open-mistral-nemo,another-mistral-model

Restart the app.

Prompt files live here:

prompts/horror/ollama.txt
prompts/facts/ollama.txt
prompts/kids/ollama.txt
prompts/documentary/ollama.txt

Each category also has:

prompts/<category>/groq.txt
prompts/<category>/mistral.txt
prompts/<category>/image_prompts.txt

You can edit these files directly or use the Prompt Pack Editor inside the web UI.

Put music files into:

C:\ai\multi-ai-video-factory\music

Supported formats:

.mp3
.wav
.m4a

The app will show them in the Music dropdown.

Generated projects are saved to:

C:\ai\multi-ai-video-factory\projects

Each project contains:

pack.json
backup.json
script.txt
images/
audio/
clips/
final_video.mp4

The projects folder is ignored by Git.

On GitHub:

New repository → multi-ai-video-factory

Recommended settings:

Public or Private: your choice
Initialize with README: No
Add .gitignore: No
Add license: No

This project already contains those files.

In PowerShell:

cd C:\ai\multi-ai-video-factory
git status

Make sure .env is not listed.

If .env appears, stop and check .gitignore.

git init
git add .
git status
git commit -m "Initial release"
git branch -M main

Replace YOUR_USERNAME:

git remote add origin https://github.com/YOUR_USERNAME/multi-ai-video-factory.git
git push -u origin main

You can also use:

.\scripts\git-first-upload.ps1 -RepoUrl "https://github.com/YOUR_USERNAME/multi-ai-video-factory.git"

Before doing that, still run:

git status

and verify .env is not included.

Run:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass

Then retry.

Your Mistral key is missing or invalid.

Check:

notepad .env

Check:

ollama list

and:

Invoke-RestMethod http://localhost:11434/api/tags

Check:

http://127.0.0.1:7860
http://127.0.0.1:7860/docs

Make sure webui-user.bat contains:

set COMMANDLINE_ARGS=--api

Check:

ffmpeg -version
ffprobe -version

If either command is not found, reinstall FFmpeg or fix PATH.

• Do not commit .env.
• Do not commit generated videos, images, audio, or project folders.
• Do not commit private API keys.
• Generated content should be reviewed before publishing.
• Documentary/facts outputs should be fact-checked before use.

MIT License.