Skip to main content
GPU 42%
ComfyUI
Online

Notifications

1

No Events

Start the automation service to see live notifications

Development Journey

The Mole World

An AI-powered animated short film built from scratch by one engineer. From screenplay to screen — ML pipeline, voice acting, and production suite.

RTX 4090 WanVideo 2.1 14B 89 AI-Generated Shots
1

The Vision

One person. Zero budget. Full animated film.

What if a solo engineer could build an entire animated short film using AI? The Mole World is a dystopian narrative set in underground tunnels where humans are controlled via neural implants. Protagonists Anaya and Deepak discover each other and spark a rebellion. Chapter 1 covers the opening 2–5 minutes.

14

Scenes

25

Shots

6

Characters

2–5 min

Runtime

Original screenplay and storyboard
25 sequential shots forming continuous narrative
7 unique underground locations
2

Prompt Engineering

Crafting the cinematic language for AI

Each of the 89 shots required carefully engineered prompts that balance visual fidelity, narrative coherence, and character consistency. A style prefix system ensures uniform cinematic quality across every frame, while emotion-driven descriptions guide the AI toward the intended mood.

89

Unique Prompts

12 tokens

Prefix Length

3–5x / shot

Iterations

Style Lock

Consistency

Cinematic prefix: grain, lighting, composition
Character-specific visual descriptors maintained across shots
Emotion tags drive camera angle and color grading
3

GPU / ML Engineering

Enterprise-grade inference infrastructure

Running WanVideo 2.1 (14 billion parameters) on a local RTX 4090 with 16GB VRAM. Built a custom ComfyUI pipeline for automated batch generation, queue management, and fault recovery. Each clip renders in ~34 minutes at full VRAM utilization.

RTX 4090

GPU

14B params

Model

~34 min

Per Clip

19.8 hrs

Total GPU

WanVideo 2.1 14B — state-of-the-art video generation
ComfyUI orchestration with automated queue and restarts
Fault-tolerant: auto-retry on OOM, checkpoint recovery
4

Dual Model Architecture

V1 for speed. V2 for quality. Both for comparison.

A two-pass rendering strategy: V1 Standard generates all 25 shots quickly for editorial review, while V2 Enhanced re-renders each shot with higher quality settings. This enables side-by-side comparison and iterative refinement without blocking the production pipeline.

25 / 25

V1 Clips

9 / 25

V2 Clips

36.8 min

V1 Avg

34.4 min

V2 Avg

V1 complete — full editorial cut available
V2 in progress — 2.7x higher fidelity output
Diff overlay tool for frame-by-frame comparison
5

Voice & Narration

AI-generated voice acting pipeline

All 89 narrations generated through a custom TTS pipeline with multiple voice profiles. Each character has a dedicated voice actor profile with reference audio for consistent tone. The system supports emotional range mapping and scene-appropriate delivery.

89

Narrations

6

Voice Profiles

76.6 MB

Audio Size

WAV/MP3

Format

Character-specific voice actor profiles
Emotional tone mapped to scene mood descriptors
Automated audio compositing with video clips
6

Production Dashboard

Real-time production intelligence

Built a comprehensive production dashboard through 43+ iterative feature batches. Started as a single HTML file (15,000+ lines) with glass-morphism UI, then migrated to Next.js + TypeScript for portfolio deployment. Live pipeline monitoring, clip management, render analytics, and storyboard visualization.

43+

Feature Batches

15K+

Lines of Code

8

Pages

Next.js

Stack

Glass-morphism dark UI with live data refresh
Real-time render tracking, GPU monitoring, activity feed
Migrated to Next.js + TypeScript + Zustand + Recharts
7

The Outcome

From concept to production in one sprint

A complete AI film production pipeline — from screenplay to rendered clips to narrated audio to editing suite — built by one person. The system demonstrates end-to-end ML engineering, prompt engineering, infrastructure management, and full-stack development. Ready for chapters 2–7.

End-to-End

Pipeline

1 person

Team Size

$0

Budget

Ch. 2–7

Next

Full stack: AI generation → Voice → Editing → Dashboard
Production-grade monitoring and fault tolerance
Portfolio-ready with live Vercel deployment

Technical Skills Demonstrated

PythonTypeScriptNext.jsReactFlaskPyTorchComfyUIWanVideo 2.1TTS/Voice AIZustandTailwind CSSRechartsGPU ComputingML InferencePrompt EngineeringCI/CDSystem Design

Built by Deep Chand

Solo founder & engineer. Turning bleeding-edge AI into production-grade products.