Original Title: Hy-Nomas-Diffusers
Tencent Hy-Nomas is a video world generation model that supports free-form, multi-turn interaction: it generates one segment of video at a time, allowing users to introduce objects, change styles and scenes, describe actions, or provide camera trajectories to control the viewpoint via text instructions during the generation process. The model consists of a segmented autoregressive DiT, a Qwen text encoder, a camera conditioning module, and a PixelUMM video decoder, using persistent KV cache to pass context between segments. This repository contains the distilled BF16 Diffusers weights; the example uses 4 denoising steps and outputs 704 × 1280 at 24 FPS.
Why It Matters Transforming "world models" from merely observable to interactively modifiable while generating: adding objects, changing styles and scenes, describing actions, and controlling viewpoints all happen within a single generation process.
Hugging Face Likes 14 · Hugging Face Trending Score 0 · Hugging Face Downloads 154
License: Conditional Commercial Use · ComfyUI: Mentioned in Model Card · VRAM: ≈21 GB Estimated
Reasons
- Hugging Face tencent account
- ComfyUI: Third-party ComfyUI package: t8star/HY-Nomas-Comfy
- Score 35 (freshness 0 / source 15 / heat 0 / comfy 4 / vram 4 / demo 5 / license 7)
Usage
- Load using HyNomasModularPipeline from Diffusers; you can also try it first in the official Demo
Risks
- The model card lists apache-2.0, but the original LICENSE text includes restrictions on region, scale, or usage; refer to the original text for authoritative terms
- Programmatic VRAM estimation is approximately 21 GB (excluding offloading)