Grand BladeMedia Start a conversation

Selected work

Work

Four cases from channels and systems we run ourselves: Shorts on what happened, a 24/7 livestream, a one-hour film and the tools behind them. There are no client logos on this page because there are no clients yet. Each case shows its numbers, how it runs and what missed.

These are channels and systems we own and run. They are not client results and are not a forecast of yours.

Short-form video

Carousel Studios & Lab

Shorts about what happened, up to four a day.

Public videos in window
50
14 days, Sep 18 – Oct 1, 2026 · as of Oct 2, 2026 · YouTube Data API read via our own tooling
Median views per video
676
Same 50 videos, Sep 18 – Oct 1, 2026 · as of Oct 2, 2026 · YouTube Data API read via our own tooling
Best single video: King Tut's Tomb Might Have a Secret Door
1,492
Public views, Sep 18 – Oct 1, 2026 · as of Oct 2, 2026 · YouTube Data API read via our own tooling

A Shorts system, public since Aug 11, 2026, reached a median of 676 views per video in its latest window.

Three Shorts from the window

Public views are each video’s lifetime count as of Oct 2, 2026, read from the YouTube Data API through our own tooling. All three were published inside the window of Sep 18 – Oct 1, 2026, whose median was 676 views per video, so these three sit above it. We chose them.

How it runs

  1. Automated. Public trend feeds and web research choose the story. A hosted AI model writes the script from dated facts, and an automated check reviews it.
  2. Automated. A synthetic voice narrates. Visuals are generated, or licensed photos credited on screen, and an automated check judges each still.
  3. Directed by a person. The format, the voice, the look, the rules and the safety rails.
  4. AI disclosure. Carousel Shorts are presented by an AI version of Justin, and the narration is his cloned voice. Each Short’s YouTube description says so. The pictures are mostly photos credited on screen, except when the AI version of Justin is speaking. On Shorts the production system uploads, YouTube’s altered-content label is set at upload.
  5. Safety rail. Up to four Shorts a day, a QA hold if automated repair fails, and a kill switch the run re-reads partway through.
  • Short-form video
  • Trend research
  • Scripting and voice
  • Publishing with QA holds
Open the channel

Livestream

The 24/7 livestream

24/7 livestream since August 2026.

Public stream entries in window
22
49 days, Aug 15 – Oct 2, 2026 · as of Oct 2, 2026 · YouTube Data API read via our own tooling
Public views across those entries
1,022
Lifetime views of each entry, Aug 15 – Oct 2, 2026 · as of Oct 2, 2026 · YouTube Data API read via our own tooling
Formats it has streamed in
Horizontal and vertical
We make no claim about a vertical feed today · as of Oct 2, 2026 · our own operating notes

Of the 22 public stream entries in the window, 6 still play back and 16 do not (checked Oct 3, 2026).

How it runs

  1. Automated. A software setup broadcasts the stream on the Justin Gim Ho Cheung channel, and YouTube lists public stream entries for it.
  2. Directed by a person. What is on screen, on a personal build-in-public channel that is not marketing work.
  3. What we do not claim. We print no day count, total hours or uptime, because the entries overlap and repeat their titles. We do not claim a vertical feed is live today.
  • Livestreaming
  • Horizontal and vertical formats
  • Playback checks
Open the channel’s Live tab

Long-form film and music

The Night Edition

One hour, 12 scenes, one looped track.

Runtime
60:04
Published Sep 26, 2026 · as of Oct 2, 2026 · YouTube Data API read via our own tooling
Scenes of about five minutes each
12
Published Sep 26, 2026 · as of Oct 2, 2026 · our own production notes
Public views
11
Since Sep 26, 2026 · as of Oct 2, 2026 · YouTube Data API read via our own tooling

One desktop machine made a one-hour film of 12 scenes in about two hours of GPU time.

How it runs

  1. Automated. An AI-made lofi and jazz track loops over 12 neo-noir scenes. Each scene is a generated still turned into a short animated loop, graded, and dissolved into the next. The render took close to two hours of GPU time on one machine.
  2. Directed by a person. The look, and the motion prompts, which say only what should happen: a negated prompt such as “no people” puts people in the scene.
  3. AI disclosure. Published with YouTube’s AI label on.
  • Long-form video
  • Music
  • Generated imagery
Watch the film on YouTube

Production systems

The tools behind the channels

Software for making and posting video.

Tools and apps built in-house
40+
Everything built to date · as of Oct 2, 2026 · Justin’s own project index
Graphics memory in the one desktop machine
16 GB
A single graphics card · as of Oct 2, 2026 · our own setup notes
What the ones behind the channels do
Make and post the videos
Our own account · as of Oct 2, 2026

The tools behind the channels make and post the videos. Production runs on one desktop machine, and scripting uses a hosted AI model.

Six of the 40+ tools

  1. Trend-to-video pipeline. Finds a story in public feeds and turns it into a finished Short.
  2. Voice comparison studio. Builds synthetic voices and sets them side by side.
  3. Clip extractor. Cuts clips out of long recordings.
  4. Thumbnail-variant generator. Makes several thumbnail options for one video.
  5. Cross-platform posting hub. One place to upload and track posts across channels.
  6. Job queue. Chains the others, so a finished step starts the next.

How it runs

  1. Automated. The queue hands work from one tool to the next on one desktop machine, with a single 16 GB graphics card. Scripting uses a hosted AI model.
  2. Directed by a person. What gets built, what runs, and what is allowed to publish.
  3. Safety rail. The Shorts system’s publishing step keeps a QA hold and a kill switch, and the Shorts it uploads have YouTube’s altered-content label set at upload.
  • Production systems
  • Automation with QA holds
  • Tooling
Read how the Shorts system works

The honest limit

What this does and does not prove

It proves that these methods have run in public, on channels we own, and that the figures here are dated and sourced.

It does not prove they will work for your brand, because none of this was done for a client, and it is not a forecast.

And it is a small sample: two channels, short windows and several modest numbers.

Read the numbers in full.

The Lab sets out the method, the windows, the definitions and the misses behind the channel figures above.

See the numbers