Our Technology
IntuiTV runs on MediaStream AI's MOTHER models - a sovereign UK AI stack trained in-house from scratch on MSAI-controlled UK compute. Weights, data and inference stay under UK control.
Sovereign by Design
Media Stream AI is a UK sovereign AI company: sovereignty is ownership across every layer - data centres, power, foundational models and products. Designed, owned and operated in Britain.
Trained In-House
MOTHER models are trained from scratch on MSAI-controlled UK compute - not fine-tunes. Reproducible, audited checkpoints with a locked inference path.
Built for Compliance
Built for UK/EU data residency and GDPR, with sovereign control over where data lives and how models run.
The MOTHER model family
A family of sovereign, in-house models built and trained in the UK.
MOTHER CORE
Sovereign UK reasoning + agentic model
Our sovereign UK reasoning and agentic model. The production 7B tier (~6.9B params) uses 48 layers, 3,072 hidden dimensions, 24 attention heads / 6 KV heads (GQA), SwiGLU, RMSNorm and RoPE, with a 4,096-token context and a 50,258-token vocabulary. Trained on a 2.4M-record curriculum spanning reasoning, tool-calling, RAG, document and code generation, and safety. A 70B tier is in testing.
View public model card →MOTHER LLM
General language model
Our general-purpose language model, powering scripting, dialogue, narration and intelligent personalization across the platform.
MOTHER CODE
Code & engineering model
A dedicated model for code and engineering tasks, supporting development, automation and technical generation workflows.
MOTHER DeepVision
Sovereign in-house text-to-video
Our sovereign, in-house ViT plus projection layer - the text-to-video generator itself, not a third-party wrapper. DeepVision turns scenes and shots into moving images for long-form output.
MOTHER T2V - sovereign text-to-video
A sovereign, from-scratch latent text-to-video model: MOTHER's own 3D video VAE and flow-matching Diffusion Transformer, conditioned on the frozen MOTHER CORE-7B text backbone. 100% MOTHER-trained - no external generator weights (Sora, Wan, LTX, SVD) are embedded; external models are used only as data teachers.
3D Video VAE
≈118M params
KL-regularised autoencoder with 8× spatial and 4× temporal compression to a 16-channel spatiotemporal latent - the compact space the transformer generates in.
Latent Diffusion Transformer
≈734M params
24 blocks, hidden size 1024, 16 heads, patch 2. Flow-matching velocity objective with classifier-free guidance, adaLN-zero time conditioning, and axial-sinusoidal 3D positions for resolution-agnostic generation.
CORE-7B Text Conditioning
48 layers · 3,072-d
The DiT cross-attends to embeddings from a frozen MOTHER CORE-7B text backbone (3,072-d), grounding every generation in MOTHER's own sovereign language model.
Specification
- Total parameters
- ~0.85B (≈118M VAE + ≈734M DiT)
- Model type
- From-scratch latent text-to-video (3D VAE + flow-matching DiT)
- Text encoder
- Frozen MOTHER CORE-7B (48 layers, 3,072-d)
- Training data
- Trained from scratch on real captioned clips
- Sovereignty
- 100% MOTHER-trained - no Sora / Wan / LTX / SVD weights embedded; UK-controlled
- Resolution
- 720p (upgrading to higher resolutions)
- Inference
- NVIDIA GB10 (Blackwell) inference nodes
- Evaluation
- Official VBench text-to-video benchmark (scores published as measured)
- Delivery
- Adaptive HLS up to 4K
Script to screen
Text-to-video runs on MOTHER DeepVision. From a single brief, MOTHER LLM and CORE expand it into scenes and shots, DeepVision generates the video, and the platform assembles long-form output of up to around 1.5 hours.
Brief
Start from one brief or idea.
Scenes & Shots
MOTHER LLM + CORE plan scenes and shots.
DeepVision T2V
MOTHER DeepVision generates the video.
Assembly
Assembled into long-form, up to ~1.5 hours.
The same pipeline powers the Mini-Drama System and IntuiSTUDIO, with support for over 60 languages and broadcaster-grade HLS playout up to 4K.
UK sovereign infrastructure
MOTHER runs on owned compute and sovereign power, hosted in the UK.
Owned GPU Compute
NVIDIA B300 and H200 GPUs hosted in the UK - Scotland (Dundee) and Manchester (2026).
Sovereign Power
Owned compute and sovereign power, with low-PUE free-cooling for efficient operation.
Data Residency & GDPR
Built for UK/EU data residency and GDPR, with reproducible, audited checkpoints and a locked inference path.
Learn more about MOTHER AI
MOTHER AI is the sovereign intelligence layer behind IntuiTV - orchestrating discovery, personalization, generation and optimization in real time.
Visit MOTHER AI Platform →