What is HunyuanVideo 1.5?
HunyuanVideo 1.5 was released on 20 November 2025 and remains Tencent's general-purpose open video model; its later video releases build on it. It is an 8.3-billion-parameter diffusion transformer with a 3D causal VAE, generating text-to-video and image-to-video at 480p or 720p, with step-distilled super-resolution models that upscale to 1080p. The default clip is 121 frames at 24 fps, about 5 seconds, and there is no audio.
Tencent lists 14GB of GPU memory as the minimum with offloading. A step-distilled 480p image-to-video model added in December 2025 runs in 8 or 12 steps and makes a clip in about 75 seconds on an RTX 4090; training code and LoRA scripts followed the same month. The weights use the Tencent Hunyuan Community Licence, which does not apply in the European Union, the United Kingdom or South Korea.
Is HunyuanVideo 1.5 worth it?
HunyuanVideo 1.5 is the practical choice for local video on modest hardware, and little else. With offloading it runs in 14GB of GPU memory, the step-distilled 480p image-to-video model makes a clip in about 75 seconds on an RTX 4090, and Diffusers, ComfyUI, LightX2V and Wan2GP all support it; Wan2GP claims to run it in as little as 6GB.
It has aged. The model makes no sound, while MiniMax H3, LTX-2.5 and MAGI-2 Preview all generate audio with the picture and rank on Artificial Analysis's current text-to-video board, where HunyuanVideo 1.5 does not appear. Tencent's own evidence is side-by-side preference testing on 300 prompts from 2025 rather than arena scores.
The licence narrows it further. The Tencent Hunyuan Community Licence does not apply in the EU, the UK or South Korea, forbids using outputs to improve any other AI model, and requires a separate licence from Tencent for services with more than 100 million monthly active users.
Hosted use is cheap but limited. fal charges $0.075 per second, so a 5-second clip costs about $0.38, but serves only 480p text-to-video and takes about 3 minutes per 121-frame clip. Some promised weights, including the 720p text-to-video distilled and sparse-attention models, are still marked coming soon a year after launch.
Where it earns its place
A freelance animator in Texas with one RTX 4090
Turn concept stills into short motion tests
Run the step-distilled 480p image-to-video model at 8 or 12 steps, about 75 seconds per clip, then upscale the keepers with the 1080p super-resolution model. Sound has to come from another tool.
A university research group in Singapore
Fine-tune a compact video model on a domain dataset
Use the released train.py with LoRA, FSDP and the Muon optimiser on a multi-GPU node; the licence allows derivatives inside its territory but bars using outputs to improve non-Hunyuan models.
Before you commit
- Setup
- Clone the repository, install Flash Attention and optionally SageAttention, download the checkpoints and run the generate script; a prompt-rewriting LLM behind a vLLM endpoint is optional but recommended.
- Integrations
- Diffusers, ComfyUI, LightX2V, Wan2GP, cache accelerators such as DeepCache and TeaCache, and fal's hosted endpoint.
- Data and privacy
- Self-hosted weights keep footage local; the licence bars use outside its territory and use of outputs to train other AI models.
- Cost at scale
- On fal, 1,000 five-second clips cost about $375; self-hosted, one RTX 4090 makes roughly 48 distilled 480p clips an hour at 75 seconds each.
- Limits
- 121 frames at 24 fps by default; native 480p and 720p; no audio; fal offers 480p text-to-video only; some distilled and sparse checkpoints still unreleased.
What’s good
- Runs in 14GB of GPU memory with offloading
- About 75 seconds per distilled 480p clip on an RTX 4090
- Open training code and LoRA scripts
- Supported by Diffusers, ComfyUI and LightX2V
What’s not
- No audio
- Licence excludes the EU, UK and South Korea
- Outputs may not be used to improve other AI models
- Several promised checkpoints still unreleased
Best for
Hobbyists and small teams outside Europe who want silent 5-second clips from a single consumer GPU.
HunyuanVideo 1.5 alternatives
HunyuanVideo 1.5: common questions
- Is HunyuanVideo 1.5 worth it?
- HunyuanVideo 1.5 is the open video model that fits a gaming PC, as long as you need no sound, no European users and nothing newer than 2025's quality. We rate it 3.5 out of 5.
- How much does HunyuanVideo 1.5 cost?
- Free weights under the Tencent Hunyuan Community Licence (not valid in the EU, UK or South Korea; separate licence above 100 million monthly users). fal charges $0.075 per second of 480p video.
- What are the downsides of HunyuanVideo 1.5?
- No audio; licence excludes the EU, UK and South Korea; outputs may not be used to improve other AI models; several promised checkpoints still unreleased.
- Who is HunyuanVideo 1.5 best for?
- Hobbyists and small teams outside Europe who want silent 5-second clips from a single consumer GPU.
- How much does HunyuanVideo 1.5 cost for a team or in production?
- On fal, 1,000 five-second clips cost about $375; self-hosted, one RTX 4090 makes roughly 48 distilled 480p clips an hour at 75 seconds each.
- Is HunyuanVideo 1.5 safe for company data?
- Self-hosted weights keep footage local; the licence bars use outside its territory and use of outputs to train other AI models.
- What are the best HunyuanVideo 1.5 alternatives?
- The closest alternatives we have reviewed are LTX-2.5, Kandinsky 6.0 Video, LongCat-Video and MiniMax H3 (Hailuo 3.0).
Answers checked Nov 7, 2026.