Meridian

Use it to generate new camera viewpoints of existing video, vary the timing of action, or create a camera move from a single image.

Availability

Adapters, inference CLI, runtime assets, and installation guide available on Hugging Face

Pricing model

Not established

Deployment

Reference implementation runs on one high-memory CUDA GPU

Privacy

Not established

Licence

Review separate code, weights, and VGGT-Omega licences; VGGT-Omega has noncommercial restrictions

• Meridian authors new observations of existing events using geometry-guided video generation. ^[1]

  **Meridian is a geometry-guided video model for authoring new observations of existing events.**

• It can generate a camera move from a single image. ^[1]

  You can also create a camera move from a single image.

• Meridian ships as two LoRA adapters for the unmodified MiniMax-H3 transformer. ^[1]

  **Meridian ships as two LoRA adapters on the unmodified MiniMax-H3 transformer**, not as a

• The release includes an inference CLI, runtime assets, sample clips, and a prototype Studio. ^[1]

  - **Included tools:** inference CLI, runtime assets, sample clips, and a prototype Studio.

• The reference implementation runs on one high-memory CUDA GPU. ^[1]

  The reference implementation runs on one high-memory CUDA GPU; memory and timings are reported below.

• The prototype Studio has no authentication and should not be exposed directly to the internet. ^[1]

• New models and tools focus on typed decisions and long-context tasks Weekly brief 2026-W39
• Claude Platform adds automatic evaluation for Managed Agents tool calls Daily brief 2026-09-24