Hy-Nomas

Use it to generate and extend interactive video worlds, changing scenes, objects, actions, styles, or camera movement through successive instructions.

Availability

Web demo and downloadable model assets with inference code

Pricing model

Not established

Deployment

Web demo; offline inference on Linux

Privacy

Not established

Licence

Apache-2.0

• A distilled BF16 Diffusers-format model is provided. ^[1]

  This directory provides the distilled model in **BF16 Diffusers format**.

• The example configuration generates at 704 × 1280 and 24 FPS using four denoising steps. ^[1]

  The example below uses **4 denoising steps**, with **704 × 1280** output at **24 FPS**.

• The usage instructions install a Diffusers version with HyNomasModularPipeline support. ^[1]

  - **Diffusers integration:** Load the bundled transformer, text encoder, tokenizer, scheduler, and decoder with `HyNomasModularPipeline`.

• Hy-Nomas generates video worlds through free-form, multi-turn text instructions and camera control. ^{[1] [2]}

  **Hy-Nomas** is a video world generation model that supports free-form, multi-turn interaction through text instructions and camera control.

• Generation proceeds one video segment at a time and reuses a KV cache to carry context between segments. ^{[1] [2]}

  It generates a world one video segment at a time, allowing users to introduce objects, change styles and scenes, describe actions, or control the camera as generation continues.

• The distilled inference configuration uses 8 denoising steps and produces 720 × 1280 output at 24 FPS. ^[2]

  The distilled inference configuration uses **8 denoising steps**, with **720 × 1280** output at **24 FPS**.

• A web demo is available, and the model assets and inference code can be downloaded. ^[2]
• A separate configuration supports video generation on a single NVIDIA GeForce RTX 4090 using quantized components and CPU offload. ^[2]

  A separate configuration supports video generation on a **single NVIDIA GeForce RTX 4090** using INT8 DiT weights, an INT8 KV cache, an FP8 decoder, and CPU offload.

• A speech model adds multilingual transcription and translation Daily brief 2026-10-26