HunyuanVideo 1.5.
HunyuanVideo 1.5 is an open-weight video model from Tencent (Hunyuan team). Windpaint doesn't serve it yet, so there's no model id or price. Request early access to start with the live models.
- Status
- Coming soon
- Task
- Text to video, Image to video
- Input
- Prompt, optional start frame
- Output
- Clip at 480p or 720p, 121 frames by default
- Developer
- Tencent (Hunyuan team)
- Parameters
- 8.3B
- Released
- November 2025
What the model does.
Paraphrased from the official model card.
HunyuanVideo 1.5 is Tencent's lightweight video model. At 8.3B parameters it does text-to-video and image-to-video, and Tencent built it to run on a single consumer GPU.
In December 2025 Tencent added a step-distilled 480p image-to-video model that runs in 8 or 12 steps, and published training code and a LoRA tuning script.
Coming soon to Windpaint.
HunyuanVideo 1.5 isn't served, so it has no model id, no price and no request example here. When it goes live, this page gets its tiers, prices and request fields like the live models have.
Want this model? Request early access. We're adding models as we add capacity.
Is HunyuanVideo 1.5 licensed for commercial use?
Licence: Tencent Hunyuan Community License.
Conditional. The licence allows commercial use only outside the EU, the UK and South Korea, and a licensee whose products had more than 100 million monthly active users in the month before the November 21, 2025 release needs a separate licence from Tencent. If Windpaint serves it, it won't be available to organizations in those three regions.
- The agreement does not apply in the EU, UK or South Korea, and the licence covers the Territory only. Section 5(c): using, reproducing, modifying, distributing or displaying the works, or their Output, outside the Territory is unlicensed. HunyuanVideo-1.5 repo, LICENSE
- Section 4: if, on the HunyuanVideo 1.5 release date (November 21, 2025), the licensee's products or services had more than 100 million monthly active users in the preceding calendar month, the licensee must request a licence from Tencent. HunyuanVideo-1.5 repo, LICENSE
- Section 3(a): third-party recipients of the works, or of products or services using them, must be given a copy of the agreement. HunyuanVideo-1.5 repo, LICENSE
- Section 3(e): any service built on it must clearly disclose to end users the full legal name of the actual provider, and must state that Tencent is not affiliated with, sponsoring or endorsing it. HunyuanVideo-1.5 repo, LICENSE
- Section 5(a): the use restrictions in 5(a) and 5(b) must be included as an enforceable provision in any agreement governing use or distribution (for example terms of use), and subsequent users you distribute to must be told about them. HunyuanVideo-1.5 repo, LICENSE
- Section 5(b): the works and their Output can't be used to improve any other AI model, except Tencent Hunyuan or its Model Derivatives. HunyuanVideo-1.5 repo, LICENSE
- Section 6(d): Tencent claims no rights in Outputs; you and your users are responsible for them. Section 3(c) encourages, but does not require, marking products "Powered by Tencent Hunyuan". HunyuanVideo-1.5 repo, LICENSE
- The Acceptable Use Policy bars placing generated content in any public context without expressly and conspicuously identifying it as machine generated, and also bars military use and high-stakes automated decisions. Governing law is Hong Kong. HunyuanVideo-1.5 repo, LICENSE
This is our reading of the licence, not legal advice. Read the licence text before you rely on it.
HunyuanVideo 1.5 VRAM requirements.
Published by Tencent (Hunyuan team) for running the weights yourself. These are the developer's figures, not Windpaint's.
- Published VRAM
- 14 GB minimum
- GPUs named
- NVIDIA GPU with CUDA
- Applies to
- Measured with model offloading on. Linux, Python 3.10 or newer.
- NVIDIA GPU with CUDA support. Minimum GPU memory 14 GB, measured with model offloading enabled; the README says you may disable offloading if your GPU has enough memory. HunyuanVideo-1.5 repo
- OS: Linux. Python 3.10 or higher. HunyuanVideo-1.5 repo
- If you hit OOM on a GPU with more than 14GB, the README suggests
PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:128; if CPU memory is limited, pass--overlap_group_offloading false. HunyuanVideo-1.5 repo - Sparse attention requires H-series GPUs, and only the 720p models have it. HunyuanVideo-1.5 repo
- Published checkpoints: 480p and 720p T2V and I2V, 480p T2V and I2V CFG-distilled, 720p I2V CFG-distilled, 480p I2V step-distilled, 720p I2V sparse CFG-distilled, and 720p and 1080p super-resolution step-distilled models. FP8 GEMM inference for the transformer was added on Dec 23, 2025 and needs sgl-kernel. HunyuanVideo-1.5 repo
- Supported in Diffusers via
HunyuanVideo15Pipeline; the README's example loads weights from the hunyuanvideo-community organisation. HunyuanVideo-1.5 repo
HunyuanVideo 1.5 native resolution.
What the developer's own code and examples use.
--resolutiontakes 480p or 720p; super-resolution models upscale 480p to 720p and 720p to 1080p. HunyuanVideo-1.5 repo- Aspect ratios: 16:9, 4:3, 1:1, 3:4, 9:16. HunyuanVideo-1.5 repo, HunyuanVideo_1_5_Prompt_Handbook_EN.md
--aspect_ratiodefaults to 16:9. HunyuanVideo-1.5 repo- Default 121 frames;
generate.pywarns that 121 frames is the optimal value. Output is written at 24 fps. HunyuanVideo-1.5 repo, generate.py
Recommended settings for HunyuanVideo 1.5.
Steps, guidance and defaults from the developer's card and code, for running the weights yourself.
- Official configs (CFG / flow shift / steps, embedded CFG none): 480p T2V 6/5/50; 480p I2V 6/5/50; 720p T2V 6/9/50; 720p I2V 6/7/50; 480p T2V and I2V CFG-distilled 1/5/50; 480p I2V step-distilled 1/7/8 or 12 (recommended); 720p T2V CFG-distilled 1/9/50; 720p I2V CFG-distilled 1/7/50; 480→720 SR 1/2/6; 720→1080 SR 1/2/8. HunyuanVideo-1.5 repo
- The README says the CFG-distilled models must use 50 steps to produce correct results. HunyuanVideo-1.5 repo
- Prompt rewriting is on by default (it runs without remote rewriting if no vLLM endpoint is configured). Suggested rewrite models are Qwen3-235B-A22B-Thinking-2507 for T2V and Qwen3-VL-235B-A22B-Instruct for I2V. Super resolution is on by default. HunyuanVideo-1.5 repo
- Diffusers example:
num_frames=121,num_inference_steps=50, exportfps=24, withenable_model_cpu_offload()andvae.enable_tiling(). HunyuanVideo-1.5 repo
HunyuanVideo 1.5 prompt guide.
From Tencent (Hunyuan team)'s own prompt guide, prompt tooling and examples.
- The README encourages longer, more detailed prompts and points to Tencent's Prompt Handbook. HunyuanVideo-1.5 repo
- Text-to-video formula: Subject + Motion + Scene + [Shot Type] + [Camera Movement] + [Lighting] + [Style] + [Atmosphere]; bracketed parts are optional. HunyuanVideo-1.5 repo, HunyuanVideo_1_5_Prompt_Handbook_EN.md
- Image-to-video formula: Subject Motion Dynamics + Scene Motion Dynamics + [Camera Movement]. The first frame comes from the uploaded image. HunyuanVideo-1.5 repo, HunyuanVideo_1_5_Prompt_Handbook_EN.md
- Camera phrases from the handbook: "the camera moves forward" (dolly in), "moves back" (dolly out), "moves to the left/right", "pans to the left/right", "tilts up/down", "circles around", "rotates 360 degrees", "follows", "remains static". HunyuanVideo-1.5 repo, HunyuanVideo_1_5_Prompt_Handbook_EN.md
- For on-screen text, put the text in quotation marks: English double quotes for English prompts, Chinese double quotes for Chinese prompts. HunyuanVideo-1.5 repo, HunyuanVideo_1_5_Prompt_Handbook_EN.md
- Describe action as a sequence ("first... then... next... finally..."), turn abstract emotion words into concrete action details, use simple directional words (left/right of frame, foreground/background), and distinguish multiple characters by attribute or position. HunyuanVideo-1.5 repo, HunyuanVideo_1_5_Prompt_Handbook_EN.md
- Prompts can be in Chinese or English; keep vocabulary and grammar simple and direct. HunyuanVideo-1.5 repo, HunyuanVideo_1_5_Prompt_Handbook_EN.md
HunyuanVideo 1.5 limitations.
As the developer states them.
- The README says disabling prompt rewriting "may result in lower quality video generation". HunyuanVideo-1.5 repo
- The 720p T2V CFG-distilled and 720p T2V sparse CFG-distilled checkpoints are listed as coming soon. HunyuanVideo-1.5 repo
- The step-distilled generation model (
--enable_step_distill) is for 480p I2V only. HunyuanVideo-1.5 repo
Questions.
Can I call HunyuanVideo 1.5 on Windpaint today?
No. HunyuanVideo 1.5 isn't served yet, so it has no model id, no price and no entry in the catalog endpoint. There's no date for HunyuanVideo 1.5.
How much VRAM does HunyuanVideo 1.5 need?
Tencent lists 14 GB of GPU memory as the minimum, measured with model offloading on. With enough memory you can turn offloading off.
What are the HunyuanVideo 1.5 system requirements?
An NVIDIA GPU with CUDA and at least 14 GB of memory with offloading, Linux, and Python 3.10 or newer.
What resolution does HunyuanVideo 1.5 generate?
480p or 720p, with super-resolution models that upscale to 720p or 1080p. The default aspect ratio is 16:9, and the default clip is 121 frames, exported at 24 fps in the official examples.
Is there a HunyuanVideo 1.5 prompt guide?
Yes, Tencent publishes a Prompt Handbook. For text-to-video it suggests subject, motion and scene, then optional shot type, camera movement, lighting, style and atmosphere. For image-to-video, describe how the subject and scene move, plus any camera movement.
Does HunyuanVideo 1.5 do image to video?
Yes. There are separate 480p and 720p checkpoints for text-to-video and image-to-video, plus a step-distilled 480p image-to-video model that uses 8 or 12 steps.
Does HunyuanVideo 1.5 support FP8?
Yes. Since December 23, 2025 the official code supports FP8 GEMM inference for the transformer, which needs sgl-kernel installed.
Is HunyuanVideo 1.5 open source?
No, it's open-weight. The weights are downloadable, but its licence (Tencent Hunyuan Community License) restricts how they can be used, so it isn't an open-source licence.
Can HunyuanVideo 1.5 be used commercially?
Conditional. The licence allows commercial use only outside the EU, the UK and South Korea, and a licensee whose products had more than 100 million monthly active users in the month before the November 21, 2025 release needs a separate licence from Tencent. If Windpaint serves it, it won't be available to organizations in those three regions. Its licence is Tencent Hunyuan Community License. That's our reading of the licence, not legal advice; check the licence text before you rely on it.
How do I ask for HunyuanVideo 1.5?
Request early access at app.windpaint.ai/register. We're letting people in gradually while we add models and capacity.