How much does HunyuanVideo cost via Railwail?

No monthly minimum, no subscription. Start with €5 free credits.

What is the context window of HunyuanVideo?

HunyuanVideo supports a unknown context window — enough for typical AI workloads.

How fast is HunyuanVideo?

Latency depends on prompt length and load — typically 200ms to 2s for short prompts. We measure p50/p95 in real-time on /rankings.

Is HunyuanVideo better than Google Veo 2?

It depends on your use case. HunyuanVideo (Tencent) and Google Veo 2 (Google DeepMind) are both strong choices in video generation. Compare them side-by-side at /compare/hunyuanvideo-tencent-vs-google-veo-2.

HunyuanVideo

Name: HunyuanVideo
Brand: Replicate
SKU: hunyuanvideo-tencent
Availability: InStock

Tencent

Video Generation

Tencent's 13B open-weights video diffusion transformer. SOTA among open video models at release.

Queue video with HunyuanVideo

Video generation runs asynchronously — we'll queue a job and you can track it in your history.

Generates as an async job — typically 30 s to 2 min.

TL;DR·Last updated May 16, 2026

HunyuanVideo is video generation AI model from Tencent, priced at €0.000 per 1M input tokens with a unknown context window.

Try HunyuanVideo

Prompt

Duration

Aspect Ratio

Pricing

Price per Generation

Per generationFree

API Integration

Use our OpenAI-compatible API to integrate HunyuanVideo into your application.

Install

npm install railwail

JavaScript / TypeScript

import railwail from "railwail";

const rw = railwail("YOUR_API_KEY");

// Simple — just pass a string
const reply = await rw.run("hunyuanvideo-tencent", "Hello! What can you do?");
console.log(reply);

// With message history
const reply2 = await rw.run("hunyuanvideo-tencent", [
  { role: "system", content: "You are a helpful assistant." },
  { role: "user", content: "Explain quantum computing simply." },
]);
console.log(reply2);

// Full response with usage info
const res = await rw.chat("hunyuanvideo-tencent", [
  { role: "user", content: "Hello!" },
], { temperature: 0.7, max_tokens: 500 });
console.log(res.choices[0].message.content);
console.log(res.usage);

Specifications

Developer

Tencent

Deep dive — Tencent's HunyuanVideo

About Tencent

Founded 1998 · Shenzhen, China

Tencent Holdings, founded 1998 by Ma Huateng and four co-founders in Shenzhen, is one of the world's largest internet conglomerates and runs WeChat, Tencent Games and Tencent Cloud. Tencent's Hunyuan team developed the company's foundation-model family (Hunyuan Large for text, Hunyuan DiT for image, Hunyuan3D, and HunyuanVideo for video). HunyuanVideo (December 2024) was the first 13B-parameter open-weight video diffusion model and remains one of the largest publicly released video generators. This deep-dive entry is the Tencent-platform-facing presentation of HunyuanVideo, covering both the open release and the Tencent-hosted commercial API surfaced through Hunyuan Cloud, WeChat creator tools and Tencent's own ad platforms.

Visit Tencent →

Architecture

Diffusion Transformer (13B) with dual-stream text/video design and 3D causal VAE

HunyuanVideo combines a 3D causal Variational Autoencoder for video compression with a 13B-parameter Diffusion Transformer denoiser. The denoiser adopts a FLUX-1-style dual-stream architecture: separate text and video streams that process each modality independently before merging into single-stream joint self-attention. Text conditioning fuses signals from a CLIP-style vision-language encoder and a large multimodal LLM, which the Tencent team found materially improves prompt fidelity on long captions. Training uses Flow Matching with 3D Rotary Position Embeddings and a progressive curriculum (images, low-res videos, high-res videos) on a heavily filtered multi-billion-clip multilingual corpus with dense bilingual captions. The Tencent-hosted commercial endpoint adds longer-duration generation (via chained extensions), 1080p upscaling, image-to-video conditioning and a 'sound' module producing aligned audio.

Parameters: 13 billion
Context: unknown

What it can do

13B open-weight base model with permissive Hunyuan Community Licence
5-second 720p / 24 fps native clips; extended to 10-15s on commercial endpoint
Strong bilingual (Chinese/English) prompt understanding via MLLM text encoder
Dual-stream DiT with high prompt adherence on dense captions
Image-to-video and audio-sync available on commercial Tencent API
Massive open-source ecosystem of LoRAs, control nets and pipelines
Available on Hugging Face, GitHub and Tencent Hunyuan Cloud
Strong on cinematic motion, lighting and dynamic camera
Best for: research, open pipelines, Chinese-market creative tools, branded campaigns.

Training & License

Multi-billion-clip curated video corpus with hierarchical filtering for aesthetics, motion and caption alignment, plus dense bilingual captions from an in-house MLLM captioner. Exact size disclosed only in summary form in the paper.

License: Tencent Hunyuan Community Licence on weights (free for research and limited commercial use with attribution; thresholds for very large deployers). Commercial API governed by Tencent Cloud terms.

Known limitations