HunyuanVideo-1.5
Tencent
Tencent's open-weight video generation model, released with inference code and weights on 20 November 2025. An 8.3B-parameter Diffusion Transformer paired with a 3D causal VAE (16x spatial and 4x temporal compression) handles both text-to-video and image-to-video at 480p and 720p, and a separate few-step super-resolution network upscales output to 1080p. The default generation length is 121 frames, exported at 24 fps. Tencent states a minimum of 14 GB of GPU memory with model offloading enabled, and a step-distilled 480p image-to-video checkpoint that generates on a single RTX 4090 within 75 seconds.
Strengths
- 8.3B-parameter DiT that Tencent says runs from 14 GB of GPU memory with offloading enabled
- One model family covers both text-to-video and image-to-video at 480p and 720p
- Few-step super-resolution network upscales output to 1080p and corrects distortions
- SSTA sparse attention gives a stated 1.87x end-to-end speedup on 10-second 720p synthesis versus FlashAttention-3
- Step-distilled 480p I2V checkpoint runs in 8 or 12 steps, cutting RTX 4090 end-to-end time by 75%
- Bilingual English/Chinese prompt handling via glyph-aware text encoding
- Training code, LoRA tuning script, Diffusers and ComfyUI integrations all published
Caveats
- Community licence, not OSI open source: the licence text excludes the European Union, United Kingdom and South Korea, and the model repo is gated against the EU
- Tencent publishes its T2V/I2V ratings, GSB comparisons and inference-speed results only as chart images, so no numeric benchmark score is citable from the vendor
- Native generation tops out at 720p; 1080p comes only from a separate super-resolution pass
- Announced checkpoints are still missing: the card says "Distillation models and sparse attention models are still coming soon" and the 720p T2V cfg-distilled and sparse rows read "Comming soon"
- Requires Linux, Python 3.10+ and a CUDA NVIDIA GPU; prompt rewriting is on by default and expects an externally served vLLM model such as Qwen3-235B-A22B-Thinking-2507
- Vendor
- Tencent
- License
- Tencent Hunyuan Community License Agreement
- Release Date
- 2025-11-20
- Modalities
- video
Capabilities
Pricing
Weights are free to download under the Tencent Hunyuan Community License Agreement, whose text states it "DOES NOT APPLY IN THE EUROPEAN UNION, UNITED KINGDOM AND SOUTH KOREA AND IS EXPRESSLY LIMITED TO THE TERRITORY"; the Hugging Face repository is correspondingly flagged extra_gated_eu_disallowed. Tencent also runs a hosted version on the official Hunyuan video site.