FULL STORY
Tencent Hunyuan Open-Sources Hy-Nomas Video World Model
Tencent Hunyuan fully open-sourced its 8B video world model Hy-Nomas on Hugging Face. Developers soon ported it to native ComfyUI nodes, enabling local inference with just 12GB of VRAM.
2026-10-23 ~ 2026-10-25 · 2 episodes · 12 posts
Episode 1 · Tencent Hunyuan Fully Open-Sources Hy-Nomas, an 8B Interactive Video World Model (2026-10-23, 9 posts)
Between October 23–24, Tencent Hunyuan released and fully open-sourced (code + weights) its video world model Hy-Nomas on Hugging Face. The 8B-parameter model ships under Apache 2.0 and, per a team member cited in a repost, is the product of their first six months on the Hunyuan team.
Confirmed
- Fully open source: code and weights on Hugging Face (tencent/Hy-Nomas), weights in safetensors with Diffusers and PyTorch formats, Apache 2.0 license, plus a 280-example benchmark
- Architecture: segmented autoregressive diffusion Transformer (DiT) with a Qwen text encoder, supporting camera control and autoregressive video generation
- Free-form interaction: multi-turn text prompts and camera trajectories; between segments users can add characters, switch styles, describe actions, or specify camera paths — "directing while generating"
- Long-horizon generation with persistent KV cache "world memory" for world consistency and continuous generation
- Output: 720p / 24FPS
- Low inference barrier: optimized for consumer GPUs, runs on a single RTX 4090
- Lightweight LoRA adapters available for domains such as robotics
Why it matters
Hy-Nomas delivers world-model-style interactive video generation as a fully open-source, consumer-hardware-runnable package, lowering the barrier for research in interactive world simulation and robotics; its KV-cache world memory enabling long-horizon, editable continuous generation is the key differentiator from ordinary video generation models.
- Tencent Hunyuan releases open world model Hy-Nomas: 720p/24FPS, runs on one RTX 4090 — AdinaYakup · 2026-10-23
- Tencent Hunyuan releases open world model Hy-Nomas with multi-turn camera control, runs on single RTX 4090 — AdinaYakup · 2026-10-23
- Tencent Hunyuan open-sources Hy-Nomas, a free-form interactive video world model — TencentHunyuan · 2026-10-24
- Tencent's Open-Source World Model Hy-Nomas: 720p/24fps, Runs on a Single RTX 4090 — mishig25 · 2026-10-24
- Tencent Hunyuan open-sources Hy-Nomas, an 8B interactive video world model — yshan2u · 2026-10-24
- Tencent releases Hy-Nomas, an autoregressive interactive video world model with camera control — tencent · 2026-10-24
- Tencent open-sources Hy-Nomas video world model you can direct mid-generation, runs on one RTX 4090 — lmoroney · 2026-10-25
- Tencent's Hy-Nomas weights land on Hugging Face under Apache 2.0 with 280-case benchmark — lmoroney · 2026-10-25
- Tencent open-sources Hy-Nomas interactive video world model that runs on a single RTX 4090 — AIFlow_ML · 2026-10-25
Episode 2 · Tencent's Hy-Nomas World Model Runs Natively in ComfyUI with 12GB VRAM (2026-10-24, 3 posts)
A developer has ported Tencent's Hy-Nomas video world model to native ComfyUI nodes, enabling local inference with just 12GB of VRAM. Workflows start from a user image and chain multiple clips with independent text prompts, bringing world models to consumer GPUs.
- Tencent's Hy-Nomas video world model ported to native ComfyUI, runs on 12GB VRAM — LatentSpacer · 2026-10-24
- Tencent's Hy-Nomas video world model runs natively in ComfyUI on 12GB VRAM — LatentSpacer · 2026-10-24
- Native ComfyUI nodes ship for Tencent's Hy-Nomas world model, runs on 12GB GPUs — PurzBeats · 2026-10-24