HunyuanVideo-1.5

HunyuanVideo-1.5 is a lightweight text-to-video and image-to-video generation model developed by Tencent, featuring 8.3 billion parameters while maintaining state-of-the-art visual quality and motion coherence. It is designed to run efficiently on consumer-grade GPUs, making advanced video creation accessible to developers and creators.

Key Features:

• Lightweight Architecture: Combines a 3D causal VAE with a Diffusion Transformer (DiT) to compress spatial and temporal dimensions (16× spatial, 4× temporal).
• Selective and Sliding Tile Attention (SSTA): Reduces computational overhead for long video sequences, achieving a 1.87× speedup in 10-second 720p video synthesis compared to FlashAttention-3.
• Video Super-Resolution: Integrates a lightweight few-step SR network to upscale outputs to 1080p, enhancing sharpness and detail retention.
• Multi-Stage Training: Employs progressive training (pre-training/post-training) and the Muon optimizer for improved motion coherence, aesthetics, and alignment with human preferences.
• Support Text-to-Video (T2V) and Image-to-Video (I2V) generation across varying durations/resolutions.
• System Requirements: NVIDIA GPU with CUDA support (minimum 14 GB GPU memory with offloading enabled).

---

The model is a component of the image generation pipeline, consisting of:

• Upsampler: ~21M (720p_sr_distilled) or ~50M (1080p_sr_distilled) parameters,
• Transformer: ~8.3B parameters,
• VAE: ~1.3B parameters.

Total: ~9.6B parameters

Announce Date: 18.11.2025
Parameters: 8.3B
VRAM requirements: 4.8 GB using 4 bits quantization, 9.6 GB using 8 bits quantization, 19.2 GB using 16 bits quantization
Developer: Tencent
License: tencent-hunyuan-community

Use our pre-built public endpoints for free to test inference and explore HunyuanVideo-1.5 capabilities. You can obtain an API access token on the token management page after registration and verification.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.