OPEN LINK ↗
// 1h agoOPENSOURCE RELEASE
Tencent Hy-Nomas Opens Interactive Worlds
Tencent’s Hy-Nomas is an Apache 2.0 interactive video world model that responds to multi-turn text prompts and camera controls while maintaining scene context across generated segments. Its standard setup uses eight GPUs, but INT8 quantization, CPU offload, and bounded history enable inference on a single 24GB RTX 4090.
// ANALYSIS
Hy-Nomas makes open interactive world models significantly more accessible, even if its real-world quality remains unproven until benchmarks and the technical report arrive.
- –Persistent KV-cache memory helps scenes remain coherent as users explore and modify them
- –Single-GPU inference lowers the barrier for researchers and independent developers
- –The 280-case benchmark covers text control, camera navigation, mixed interaction, and spatial memory
- –Robotics and cinematic LoRA adapters point toward simulation and creative-production use cases
- –The main caveat is missing benchmark scores, so impressive demos should not yet be treated as evidence of state-of-the-art performance
// TAGS
hy-nomasopen-sourcevideo-genmultimodalagent-memorygpuinference
DISCOVERED
1h ago
2026-10-23
PUBLISHED
1h ago
2026-10-23
RELEVANCE
9/ 10
AUTHOR
TeksEdge