Tencent released Hy-Nomas weights, code and training and inference recipes.
Tencent releases Hy-Nomas for interactive world generation
2026-10-24Verified announcementTRIAL
Tencent released Hy-Nomas code, training recipes and open weights for multi-turn world generation controlled by text and camera movement. The project includes distilled 4-, 6- and 8-step variants and low-memory paths for a single RTX 4090. Its four-benchmark evaluation is publisher-reported, and no technical report is public.
Availability: public code, training and inference recipes, model weights and Diffusers conversion
What this means for you
For teams prototyping interactive simulation or video agents, Hy-Nomas lowers the artifact barrier from a demo to runnable weights and code. The lack of an independent evaluation and technical report makes local temporal-consistency and control tests essential.
Evidence & related material
Technical details & scores
| Architecture | segment-autoregressive diffusion transformer with persistent KV cache |
|---|---|
| Availability | public code, training and inference recipes, model weights and Diffusers conversion |
| License | Apache-2.0 stated by the repositories |
| Runtime | publisher provides INT8/FP8 low-memory settings for a single RTX 4090 |
| Evaluation | four publisher-run benchmarks covering 280 cases; no independent evaluation |
| Time Basis | official Tencent post timestamp decoded to 2026-10-23T17:00:00Z; repositories were created earlier and do not alone establish release time |
Technical impact 4
Engineering value 4
Adoption signal 2
Maturity 3
Verification cost 4
Risk 3