EMNDR — AI Story Studio

about emndr

Storytelling engineered for the age of AI

EMNDR is a solo-founded AI video production studio built around a fully custom platform — Combined Studio — developed to automate every stage of narrative video production.

The platform integrates story discovery, AI scriptwriting, ElevenLabs voice synthesis, image and video generation, assembly, and YouTube marketing into a single unified workflow.

EMNDR is an early-stage operation actively building and scaling its production capacity on the Polish-language channel NARRATIO PL.

100%

AI-Native Pipeline

12+

Integrated Modules

~12 min

Avg. Video Length

Revenue Streams

capabilities

Built to produce at scale

Every module was built to solve a specific production bottleneck.

🔍

Story Discovery

Searches Reddit, Quora, Medium and YouTube simultaneously. Scores results by engagement potential and AI quality rating.

Groq · OpenAI · YT API

😂

Humor & Absurd Stories

AI searches 16 curated literary sites across EN, ES and PL. Cards include Polish summaries and comedic premise translations.

Pratchett / Adams style · PL · EN · ES

🔬

Topic Research

Describe any topic — AI runs 3 parallel web searches across Reddit, Medium, YouTube, academic sources and blogs in EN, PL and ES.

3× Web Search · EN · PL · ES

✏️

AI Scriptwriting

DeepSeek-V3 and Claude Sonnet 4.5 rewrite raw material into a ~15-minute script. 5 narrative style modes with side-by-side AI comparison.

DeepSeek-V3 · Claude Sonnet 4.5

🎬

AI Production Studio

Analyzes the script and generates voice recommendations, image prompts, animation prompts, and assembly guide.

Claude · DALL-E 3 · Runway

🎙️

ElevenLabs Voice

Casts every role from a proprietary 20-voice palette — matching gender, then age, then measured acoustics. Balances loudness across characters and rotates narrators between productions.

20 voices · dBFS balancing

🎞️

Video Assembly Pipeline

Whisper aligns scenes to the real narration, then moviepy renders Ken Burns motion, crossfades, background video, logo and mixed audio — with an outro scene when the recording outlasts the scenes.

Whisper · moviepy

📺

YouTube Marketing

Generates titles, full description, hashtags, YouTube tags and viewer question — directly from Studio content.

GPT-4o mini

🧮

Cost Calculator

Models 22 cost line items across the whole pipeline — voice, SFX, images, video seconds, transcription and every API key, including free-tier ones. Auto-picks the cheapest plan for your volume.

22 items · auto-pick

📡

YT Channel Explorer

Discovers storytelling channels in chosen niches and fetches top-performing videos sorted by views, likes or comments.

YouTube Data API

💾

Project Export

One-click full project save — exports source, both AI scripts, Studio prompts and YouTube materials into a dated folder with README and cost summary.

Auto-organized · Dated folders

📌

Saved Stories

Bookmark discovered stories for later. Quick-send to Editing or Studio. Scores, summaries and source URL preserved across sessions.

Persistent · JSON storage

🌐

Bilingual Interface

Full EN/PL interface switch. All tabs, labels, status messages and AI-generated content translate simultaneously.

Python · customtkinter

🎚️

Sound Design

AI reads the script and proposes ambient beds and spot effects — each with a ready ElevenLabs SFX prompt and a volume level tuned to sit under the narrator.

ElevenLabs SFX · Freesound

📊

Production Planner

Forecasts every run before it starts: missing images by model, video clips by length, totals in USD and PLN at the live NBP rate.

Pre-flight forecast

api architecture

Multi-Engine Architecture

Every module uses the best AI for its task — with automatic failover to a backup engine.

🔍

Search & Research

Inspirations · Humor · Research

🔎 Perplexity Sonar PRIMARY

🤖 GPT-4o-mini + web_search FALLBACK

✏️

Script Rewriting

Writing · Colorize

🧠 Claude Sonnet 4.5 PRIMARY

🐋 DeepSeek-V3 FALLBACK

🎬

Studio JSON & Scene Prompts

DALL-E · Midjourney · Runway

🐋 DeepSeek-V3 PRIMARY

🤖 GPT-4o-mini FALLBACK

⚡

Scoring & Ranking

Results · Virality

🌪️ Groq Llama-3.3-70B PRIMARY

🤖 GPT-4o-mini FALLBACK

🎬

Video Generation

Key scenes · image-to-video

⚡ Seedance (fal.ai) $0.045/s

🎞️ Kling · Runway FALLBACK

~60% cheaper per clip

🖼️

Image Generation

Cost-aware auto routing

✨ GPT Image — key scenes PREMIUM

⚡ GPT Image mini — the rest 4–7× CHEAPER

~0.07 PLN per image

production pipeline

From story to screen

Every video passes through seven tightly integrated AI-powered stages inside Combined Studio.

🔍

Story Discovery

Perplexity Sonar sweeps Reddit, Quora, Medium and YouTube; Groq scores every candidate.

Perplexity Sonar · Groq · YT API

✏️

Script & Narrative

Claude Sonnet 4.5 and DeepSeek-V3 rewrite in parallel — two drafts, side by side.

Claude Sonnet 4.5 · DeepSeek-V3

📐

Pre-production

A Style Bible locks the visual canon; the script becomes a numbered scene manifest.

Style Bible · manifest.json

🎤

Voice Casting

Roles cast from a 20-voice palette by gender, age and measured acoustics.

ElevenLabs · 20 custom voices

🎚️

Sound Design

Ambient beds and spot effects with ready SFX prompts; music matched to mood.

ElevenLabs SFX · Jamendo · FMA

🖼️

Image Generation

Prompts inherit the Style Bible; generation routes key scenes to the premium model.

GPT Image · auto routing

🎞️

Animation

Key scenes get image-to-video prompts with explicit camera motion, animated by Seedance.

Seedance · Runway · Pika

🎬

Assembly

Whisper aligns scenes to the narration; moviepy renders the final MP4.

Whisper · moviepy

📈

Publish & Shorts

Search-intent mapping, SEO-weighted tags, and Shorts cut from the finished film.

Claude · YouTube SEO

voice at the core

Voice is not a feature. It is the product.

In narrative video, the voice is what makes a viewer stay or leave. ElevenLabs is the only platform that delivers the emotional range storytelling requires.

At our production scale, voice synthesis is the highest-impact layer. ElevenLabs quality is non-negotiable.

Natural prosody — emotional peaks, pauses, and rhythm matching the narrative arc
11,000+ characters per video — high-volume synthesis for 10–15 minute productions
Consistent voice identity — the same recognisable voice across the entire channel
A proprietary 20-voice palette — designed, generated and acoustically profiled in-house
Casting by measurement, not vibes — gender, age, then measured timbre, with an enforced contrast rule
Loudness balancing — every voice measured in dBFS and gain-corrected to one perceived level
Narrator rotation — remembers the last six narrators and avoids repeats
v3 audio tags & pronunciation dictionary — per-fragment emotion, consistent proper names

"Audio quality is more critical to viewer retention than video quality. Audiences tolerate a blurry frame. They will not tolerate a lifeless voice."

— TechSmith, Video Engagement Research 2023

production craft

The details that survive contact with production

Anyone can chain a few APIs together. What makes a pipeline usable at 30 videos a month is everything that happens when something goes wrong — or gets expensive.

💸

Nothing is generated twice

Every image and clip is fingerprinted by prompt, model and source image — re-runs skip finished assets instead of re-billing them.

Fingerprint cache

📊

Forecast before you spend

Runs open with a forecast — what's missing, on which model, and the cost in USD and PLN. Video clips are approved scene by scene.

Live NBP rate

🎚️

Loudness that matches

Voices measured in dBFS and gain-corrected; ambience and effects ducked under the narrator.

dBFS normalisation

🎯

Visual canon, not vibes

A Style Bible locks each character's look; every scene carries an explicit MUST SHOW list and an avoid list.

Consistency guard

⏱️

The tail is never cut

Scene timings come from Whisper alignment; when audio outlasts the scenes, a spare image becomes an outro.

Audio-driven timeline

🔐

Safe to show on stream

Console paths are shortened to the project root — screenshots never expose the machine behind them.

Path sanitising

🌍

Bilingual to the core

UI, console and the AI's own analysis follow one language switch — while the source story is never translated.

PL · EN

🛡️

Fails soft, not loud

Every engine has a fallback, rate limits are honoured, and exhausted quota is explained in plain language.

Graceful degradation

🛡️

A watchdog for model retirements

Weekly checks against every provider's model list — warning before the deadline, with the exact file to change.

Weekly check

🔁

Survives a retirement mid-run

Every request passes a resolver that swaps a retired model ID for its successor — production keeps running.

Zero-downtime migration

💱

Prices that keep themselves current

Token prices refresh weekly into the cost calculator — no spreadsheet edits.

Auto price sync

🧮

Every cost accounted for

A calculator models 22 line items and auto-picks the cheapest plan for your volume, with break-even against subscriptions.

22 line items

From Story to Screen —
Fully Automated

Storytelling engineered for the age of AI

One platform. Every stage.

Built to produce at scale

Multi-Engine Architecture

From story to screen

Voice is not a feature. It is the product.

The details that survive contact with production

Built for long-term scale

Let's build something together

From Story to Screen —Fully Automated

Storytelling engineered for the age of AI

One platform. Every stage.

Built to produce at scale

Multi-Engine Architecture

From story to screen

Voice is not a feature. It is the product.

The details that survive contact with production

Built for long-term scale

Let's build something together

From Story to Screen —
Fully Automated