HunyuanVideo 1.5
Open-Source Video on a 4090
Tencent's open-weight video model — only 8.3B parameters, designed to run on a single consumer RTX 4090. Released under Apache 2.0 in late 2025, it's the entry point for hobbyists and academics to do real video generation locally without datacenter access.
Cost
Free
Open weights — self-host
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on
Artificial Analysis. Cost buckets are fixed dollar
thresholds keyed off output-token price ($/M out).
Intelligence
- Top 1%≤ 1%
- Top 5%≤ 5%
- Top 10%≤ 10%
- Good≤ 25%
- Medium≤ 50%
- Below avg> 50%
Speed
- Top 1%≥ 345 tok/s
- Top 5%≥ 237 tok/s
- Top 10%≥ 196 tok/s
- Good≥ 146 tok/s
- Medium≥ 90 tok/s
- Slow< 90 tok/s
Cost
- Freeopen weights · self-host
- Low< $1 / M out
- Moderate$1–5 / M out
- High≥ $5 / M out
Why it matters
Made consumer-GPU video generation real. Prior open-source video models were either too small (talking-head only) or required 8×A100 setups. HunyuanVideo fits the entire frontier-quality pipeline on a $1500 GPU.
Core Capabilities
Generative
Produces images, video, audio, or other media.
Multimodal
Combines text, vision, and audio in one model.
Vision
Understands images, scenes, and visual context.
Context Window
Context window not disclosed.
Availability
API
Available
Product / App
Not available
Open Source
Released
Enterprise
Contact sales
Pricing Model
Free / self-host
Open weights — pay only for compute.
Self-host What it feels like
- First open-weights video model genuinely competitive with Runway Gen-3 and Luma 1.6
- 13B params (8.3B in HunyuanVideo-1.5) — runs on consumer GPUs unlike most frontier video models
- Human evaluators preferred Hunyuan's motion quality but Gen-3's visual quality — split decision
- Outperformed all 3 top Chinese commercial video models in blind professional review
- 5-second clips at 480p / 580p / 720p — short-form social-video sweet spot
- Apache-2.0 weights are a big deal — first credible open alternative to closed video APIs
Best use cases
- Self-hosted video generation pipelines (privacy / cost / customisation)
- Research and fine-tuning experiments on video diffusion
- Short social-media clips (5s, 720p) at zero per-clip cost after hardware
- Image-to-video workflows via the I2V variant
Tools to try
Not ideal for
- Long-form (>5s) coherent narrative clips
- Cinematic-quality work where Sora 2 / Veo 3 photorealism matters
- Synced audio generation — Hunyuan Video is silent (use Veo 3 or Kling 2.6 for sound)