From first file to first real edit

What changed in the first month

ShadyCut started as a relatively straightforward pipeline around transcripts, gameplay signals and story selection.

That did not last long.

Once we started testing on real multi-hour streams, it became obvious that individual “good moments” were not enough. The system needed to understand the streamer, the game, the changing goal of the session, what the viewer needed to see, and how one moment connected to the next.

So the pipeline expanded and was repeatedly rebuilt around those problems.

Game Profiles

ShadyCut became game-aware.

Instead of treating every stream as generic video, Game Profiles provide game-specific context that the rest of the pipeline can use when interpreting mechanics, characters, terminology and events.

That gave the system a much stronger foundation for understanding what was actually happening on screen.

A deeper understanding layer

The pipeline grew beyond transcription.

ShadyCut began combining streamer speech with gameplay activity, game audio, visual signals, speaker information and other session-level evidence.

Streamer camera activity was also added, helping the system recognize when the creator’s reaction itself was part of the story.

Story generation was rebuilt

The early versions mostly focused on identifying strong moments.

That evolved into something much closer to actual story construction.

The streamer’s speech remains the main narrative backbone, while gameplay and other signals provide context and evidence.

From story to edit

Finding the story was only half the problem.

The Director layer evolved to turn that story into a proposed edit: selecting source material, arranging scenes, trimming unnecessary sections, identifying missing context and adding editorial guidance where the cut needs help.

Continuity became a major focus as well. A scene can be individually strong and still fail if the viewer does not understand what caused it, what a reaction refers to, or how one scene connects to the next.

Editable output

The final step was making the result useful outside ShadyCut itself.

The system now produces an editable Premiere-ready timeline with scene selections, editorial notes and structured output that can be reviewed and refined instead of forcing an editor to start again from the original stream.

The first pilot: Chowds

The Chowds Arena Breakout session became ShadyCut’s first beta pilot.

The goal was not yet to prove that every editorial decision was perfect. The goal was to prove that the complete pipeline could run from a long raw stream all the way to a structured story and usable editing output.

It did.

That first end-to-end run exposed plenty of weaknesses, but it also proved that the core idea worked.

A long stream could be analysed, understood, structured and turned into something an editor could actually begin working with.

The first real test: 1ceStream

The next major test was 1ceStream’s Genshin Impact session built around Odette.

By that point, ShadyCut was already a very different system.

It could see more of the session, understand more of the game, use streamer reactions more effectively, build a stronger narrative structure and provide much more detailed editorial guidance.

The difference showed clearly in the output.

This test also revealed the next class of problems: not whether ShadyCut could find the story, but whether the final cut preserved enough context and continuity for a viewer who had never seen the original stream.

That led to another round of improvements around story requirements, scene boundaries, missing setup, unresolved intentions, reaction context and final Director continuity checks.

One month in

The target for August was to get to the first real output.

We got there.

But the more important result was learning what ShadyCut actually needed to become.

It was not enough to detect highlights.

It was not enough to summarize a stream.

And it was not enough to identify good scenes.

The system needed to understand why those scenes mattered, how they connected, and what the viewer needed in order for the final video to feel intentional.