MeridianBrand Film2025
Atlas
A four-minute origin film for a navigation company, shot entirely with a custom-trained model and a cast that had to stay consistent across 312 shots.
End-to-end AI video production
9AM Studio — the end-to-end AI video production studio
Stories that survive the machine.
We develop the story, build the characters, train the models, shoot the film, score it, finish it and ship it to every screen it needs to live on. Seven stages. One studio.
- Stages, one studio
- 7
- Stages, one studio
- Models trained
- 40+
- Models trained
- Shots per month
- 600+
- Shots per month
- Story development
- Character systems
- LoRA training
- AI cinematography
- Animation
- Voice & score
- Sound design
- Compositing
- Colour grade
- Localisation
- YouTube operations
- Enterprise automation
Scene 01 — 01 / 07
Story Lab
Every frame starts as an argument about what the story is.
We develop premise, structure and beat sheets before a single pixel is generated. Narrative architects work alongside prompt engineers so the story survives contact with the model — not the other way around.
- Concept & premise development
- Beat sheets and shooting scripts
- Story architecture for serialized formats
- Narrative QA against brand canon
3–5 daysConcept to locked beat sheet
Explore
Scene 02 — 02 / 07
Character Consistency Lab
A character the audience can recognise in shot 400.
Identity drift is the failure mode that kills AI film. We build reference sets, identity locks and continuity sheets so faces, wardrobe and proportion hold across every shot, episode and season.
- Reference set construction
- Identity locking across shots
- Wardrobe & prop continuity sheets
- Multi-character scene blocking
98.6%Identity match across a 400-shot cut
Explore
Scene 03 — 03 / 07
LoRA Training Center
Models that know your world before you brief them.
We train custom LoRAs and fine-tunes on your characters, products, sets and visual language. The result is a private model that renders your brand correctly on the first pass instead of the fortieth.
- Character & product LoRA training
- Style and look fine-tunes
- Dataset curation and tagging
- Evaluation harness and version control
40+Production models trained and maintained
Explore
Scene 04 — 04 / 07
AI Production Studio
Cinematography, not generation.
Shot lists, lens language, blocking and coverage — directed by people who have shot film. The model is a camera crew, and we run it like one: dailies, selects, and a cut that holds together.
- Shot design and coverage planning
- AI cinematography and lens language
- Animation and motion direction
- Dailies, selects and editorial
6×Faster than a comparable live-action schedule
Explore
Scene 05 — 05 / 07
Audio Studio
Half the picture is what you hear.
Voice casting and cloning with consent, original score, dialogue editing, foley and full mix. We deliver stems and a broadcast-ready master, not a temp track bolted on at the end.
- Voice design, casting and licensed cloning
- Original score and sound design
- Dialogue edit, ADR and foley
- Mix, stems and loudness-compliant masters
-14 LUFSPlatform-compliant delivery masters
Explore
Scene 06 — 06 / 07
VFX Laboratory
Where generated footage becomes finished footage.
Compositing, cleanup, rotoscoping, simulation and grade. This is the stage that separates a convincing film from an interesting demo — and it is still, stubbornly, craft.
- Compositing and cleanup
- Roto, paint and beauty work
- Simulation, particles and environments
- Colour grade and finishing
4K / HDRFinishing and delivery standard
Explore
Scene 07 — 07 / 07
Distribution Center
One story, every screen it needs to live on.
Versioning, aspect ratios, localisation, thumbnails and platform-native cutdowns — plus the publishing automation that keeps a long-form channel shipping on schedule.
- Multi-aspect versioning and cutdowns
- Localisation and subtitling
- YouTube and long-form channel operations
- Publishing automation and analytics loops
18Deliverable formats from a single master
Explore
Selected work
Films that shipped.
Nova LabsEpisodic Series2025The Quiet Hours
Six-episode animated series with a recurring cast, built on a character system that held identity across two seasons of production.
Harbor FinancialCampaign2024Small Numbers
A 14-asset campaign built from one master film, versioned across broadcast, social and in-branch displays.
KestrelProduct Film2025Field Guide
Product films rendered from a LoRA trained on CAD geometry, so every hero shot matched the shipping hardware exactly.
Orchard MediaLong-Form Channel2024—2025The Long Way
A weekly long-form YouTube channel producing 40-minute documentaries on an automated pipeline with a two-person creative team.
Vantage HealthEnterprise2025Protocol
An internal training library of 90 modules, generated from a governed pipeline the client's own team now operates.
Custom AI · Nothing leaves our infrastructure
Your material never reaches a third-party model.
Producing AI video normally means handing your material to someone else's model. The terms behind those services often allow retention, human review and reuse for training — and they change without asking you.
Ownership, retention and non-training terms are written into the agreement before work starts, so this is something you can enforce rather than something you have to take on trust.
This site itself runs no analytics, no cookies and no third-party trackers. The only information it collects is what you choose to type into the contact form, which is emailed to us and nowhere else.
- 01
No third-party generation APIs
Every frame is generated on models we train and run on infrastructure we control. Your footage, scripts, reference boards and talent likenesses are never uploaded to an external generation service.
- 02
No vendor terms to inherit
With no third party in the loop there is no external retention clause, no human-review provision, and no terms of service that can be rewritten after your project ships.
- 03
Your model is yours alone
Trained in isolation on your material. Not pooled with other accounts, not reused between clients, and never published back into a public model.
- 04
Deleted on your schedule
Project assets, datasets and trained weights are removed from working storage on a retention window you set — or archived under your control if you would rather keep them.
Start at 9am
Tell us what you want to make.
Send a brief, a script, a reference board or a sentence. We will come back with an approach, a schedule and a number.