whatmodel.sh / video / HunyuanVideo 1.5 (I2V)

HunyuanVideo 1.5 (I2V)

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

HunyuanVideo 1.5 (I2V) is an open-weights video model from Tencent, released November 2025. Its published size is 8.3B parameters. Licence: tencent-hunyuan-community.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

8.3B = diffusion transformer; the card describes one unified framework for T2V and I2V. Excludes text encoder and VAE.

52.7±2.1#44 of 93 video modelsfrom 8 benchmark results across 8 sources

$ how it compares

The video models ranked either side of HunyuanVideo 1.5 (I2V), out of 139 with a score, then more from Tencent.

#modelscorewhereRAMreleased
53LTX-2.5 Lightricks53±6api $0.12/sec +local~14GB2026-08
54Vidu Q2 Pro ShengShu (Vidu)53±3comparison only2025-09
55HunyuanVideo 1.5 (I2V) Tencent53±2api $0.075/sec +local~66GB2025-11
56Kling 2.1 Master Kuaishou (Kling AI)53±2api $0.28/sec2025-05
57Kandinsky 5.0 Pro Sber52±2api $0.040/sec +local~76GB2025-11
more from Tencent
58HY-OmniWeaving Tencent52±4comparison only2026-03
59HunyuanVideo 1.5 Tencent52±2api $0.075/sec +local~20GB2025-11
90HunyuanVideo-I2V Tencent45±4local~57GB2025-03

see all video models · find the ones that fit your machine

$ what it's good at

image to video▁▂·▂▅█▅▄▅▆▃▄ #39 of 50 · 77% of its score · 5 tests

Trails Cosmos3-Nano, Kling 2.1 Master here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
WorldReasonBench human Elo (image+text-to-video world-state prediction)911.0 ✎▃▃··█··▃▃▃·▃ beats 3/8avg 875▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%sharp
Cosmos HUE human evaluation, image-to-video (Cosmos 3 report)85.6% ✎▄·█··▄··▄▄▄█ beats 3/8easy · avg 84%▇▇▇▇ 11%typical
VGI-Bench visual reasoning (I2V)19.1% ✎▄▄···▄▄█·█·▄ beats 0/8hard · avg 20%▇▇▇ 7.2%typical
What-If World causal-intervention video generation (APEO paired average, Gemini 3.1 Pro judge)27.0% ✎█·▄····▄···█ beats 0/5hard · avg 27%▇▇ 3.9%blunt · at its level
Physics-IQ Verified (I2V, best-practice prompts)33.4% ✎██···█··█·██ beats 4/5hard · avg 25%▇ 3.0%blunt · at its level
other tests3 not in a task · 21% of its score
testresultvs the others in its rundifficultycountswhy
MindWorldBench intention accuracy (I2V mental-state reasoning)23.0% ✎▅·▃▅·█·▃···▃ beats 4/9hard · avg 21%▇▇▇▇ 11%typical
VideoArgus-Bench TI2V (rubric-grounded score)81.1% ✎█··▄▄▄▄····▄ beats 4/6easy · avg 77%▇▇▇ 6.3%blunt
MechVerse mechanical-motion I2V (WorldModelBench-judge sum)75.6% ✎▂···▂······█ beats 2/5medium · avg 65%▇▇ 4.1%blunt · at its level

$ where to run it

Through an API

api · fal.ai · $0.075/sec · fal.ai ↗ · price sourceUSD per second of generated video, 480p (the only resolution offered on this endpoint). Product: fal HunyuanVideo 1.5 image-to-video endpoint. API model: fal-ai/hunyuan-video-v1.5/image-to-video. Price basis: rate printed by fal ("$0.075 per second"), matching the endpoint billing record. Currency: USD. fal.ai model page, observed 2026-09-25.

Other ways to run it

  1. Run it as published · needs ~65.6 GB, NVIDIA only · from huggingface.co/hunyuanvideo-community
    uv run --with torch --with diffusers --with transformers --with accelerate --with imageio-ffmpeg python -c "import torch; from diffusers import HunyuanVideo15ImageToVideoPipeline; from diffusers.utils import export_to_video, load_image; p=HunyuanVideo15ImageToVideoPipeline.from_pretrained('hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-480p_i2v', torch_dtype=torch.bfloat16); p.enable_model_cpu_offload(); p.vae.enable_tiling(); image=load_image('https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/wan_i2v_input.JPG'); export_to_video(p(prompt='Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard.', image=image, generator=torch.Generator(device='cuda:0').manual_seed(1), num_frames=121, num_inference_steps=50).frames[0], 'output.mp4', fps=24)"

Official weights and documentation: huggingface.co/tencent ↗

$ how the score was worked out · WorldReasonBench: Human-Align… 52%, MindWorldBench: Evaluating Me… 11%, Cosmos 3: Omnimodal World Mod… 11% and 5 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors (arXiv v1)1▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%CC-BY-4.0
MindWorldBench: Evaluating Mental-State-to-Behavior Reasoning in Image-to-Video Generation (arXiv:2609.39147v1)1▇▇▇ 11%CC-BY-4.0
Cosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)1▇▇▇ 11%CC-BY-4.0
VGI-Bench: Probing Visual Intelligence in Video Generation Models1▇▇ 7%CC-BY-4.0
VideoArgus: Agentic Rubric-Grounded Unified Evaluation for Video Generation and Editing (arXiv:2608.05485v1)1▇▇ 6%CC-BY-4.0
MechVerse: Evaluating Physical Motion Consistency in Video Generation Models (arXiv:2605.14843v1)1▇ 4%CC-BY-4.0
What-If World: A Causal Benchmark for General World Models in Embodied Scenarios (arXiv:2605.27589v1), Table 21▇ 4%CC BY 4.0
Physics-IQ Verified1▇ 3%CC-BY-4.0
prior · pull toward the middle▇ 2%
weighted average= 52.7

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json