whatmodel.sh · find the best video models you can run

Best text-to-video models to run locally on a 32GB graphics card

A leaderboard of the best text-to-video models you can run locally on a 32GB graphics card (with 32GB of system RAM). MiniMax H3 is #1 right now, followed by Cosmos3-Nano and HunyuanVideo 1.5. Models are ranked on human votes and video generation tests. Each model shows the download that fits and the command to run it.

34 models · scores combine 25 benchmarks · updated how it works · find a leaderboard · benchmark your machine
task[all tasks][text to video][image to video][video editing]

Scores use reviewed text to video evaluations only. Compare scores within this task. Models without task measurements are excluded.

Models for 32GB RAM · 32GB graphics card → ~32GB fast · local only · reset
+ more
card[RTX 5090][RTX 5000 Ada][RTX PRO 4500 Blackwell]
set[all sets][LMArena video · 94][VABench T2AV micro Visual QA · 3]

Filter the model list while keeping task scores. Models without a shared score scale are not yet calibrated; select a set to see its own scores.

Detect your exact chip or graphics card:

curl -G whatmodel.sh/video -d max=0 --data-urlencode "mac=$(sysctl -n hw.memsize machdep.cpu.brand_string)"Mac

curl -G whatmodel.sh/video -d max=0 --data-urlencode "ram=$(awk '/MemT/{print $2}' /proc/meminfo)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>/dev/null)"Linux

curl.exe -G whatmodel.sh/video -d max=0 --data-urlencode "ram=$((gcim Win32_ComputerSystem).TotalPhysicalMemory)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null)"Windows

$ terminal

detecting your OS…

Run curl "whatmodel.sh/video?ram=32&gpu=card&vram=32&max=0"to get this list in your terminal (add &format=json for agents).

ram[8][16][24][32][48][64][96][128][192][256]GB
gpu[apple silicon][graphics card][none]
vram[1][2][4][8][12][16][24][32][48][64][80][96][128][160]GB
price[local][$0.02][$0.05][$0.1][$0.25][any]max per sec

detecting your OS…

Score vs memory to run

#modeltask score↓where↓speed↓RAM↓get itsources
#modeltask score↓where↓speed↓RAM↓get itsources
8MiniMax H3 · gguf-q4_k_m fl2va + Qwen3-VL 32B q4_k_m (sd.cpp) MiniMax76±2*local (slow)~52GBhow to run › LMArena leaderboard datasetSci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains
35Cosmos3-Nano · bf16 NVIDIA54±4*local (slow)~42GBhow to run › Cosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)Sci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains
36HunyuanVideo 1.5 · bf16 Tencent54±2local~20GBhow to run › LMArena leaderboard datasetCosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)WorldJen: An End-to-End Multi-Dimensional Benchmark for Generative Video ModelsSci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science DomainsHumanScore: Benchmarking Human Motions in Generated Videos
42Wan 2.2 A14B · gguf-q4_k_m ×2 experts + umt5 q8_0 (sd.cpp) Alibaba52±2local~31GBhow to run › LMArena leaderboard datasetCosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)WorldJen: An End-to-End Multi-Dimensional Benchmark for Generative Video ModelsVideoVerse: Does Your T2V Generator Have World Model Capability to Synthesize Videos?
44Wan 2.1 14B · gguf-q8_0 + umt5 q8_0 (sd.cpp) Alibaba50±2local~27GBhow to run › VideoPhy / VideoPhy-2 official repository leaderboardsSkyReels-V2: Infinite-length Film Generative ModelCosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)T2VPhysBench: A First-Principles Benchmark for Physical Consistency in Text-to-Video GenerationAlibaba Wan technical report (arXiv:2503.20314v2), VBench table (own models)
46Kandinsky 5.0 T2V Lite · bf16 Sber49±2*local~29GBhow to run › LMArena leaderboard dataset
48Cosmos-1.0-Diffusion-7B-Text2World · bf16 NVIDIA48±3local~30GBhow to run › VideoPhy / VideoPhy-2 official repository leaderboardsCosmos World Foundation Model Platform for Physical AI (arXiv:2501.03575v3)
52Wan 2.2 TI2V 5B · fp16 + umt5 q8_0 (sd.cpp) Alibaba47±4local~21GBhow to run › Cosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)VABench: A Comprehensive Benchmark for Audio-Video Generation (arXiv:2512.09299v2)Sci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains
53CogVideoX-5B · bf16 Zhipu AI (THUDM)47±2local~13GBhow to run › VideoPhy / VideoPhy-2 official repository leaderboardsGenAI-Arena (TIGER-Lab) text-to-video leaderboardHumanScore: Benchmarking Human Motions in Generated Videos
57Pyramid Flow (SD3) · bf16 Peking University / Kuaishou45±3*local~8GBhow to run › GenAI-Arena (TIGER-Lab) text-to-video leaderboard
58T2V-Turbo-v2 · fp32 UC Santa Barbara45±7*†local~9.4GBhow to run › T2V-Turbo-v2: Enhancing Video Generation Model Post-Training through Data, Reward, and Conditional Guidance Design (arXiv 2410.05677v3)
61Wan 2.1 T2V 1.3B · fp16 + umt5 q8_0 (sd.cpp) Alibaba43±4local~12GBhow to run › WorldJen: An End-to-End Multi-Dimensional Benchmark for Generative Video ModelsAlibaba Wan technical report (arXiv:2503.20314v2), VBench table (own models)Training-Free Efficient Video Generation via Dynamic Token Carving
62LTX-2.3 · gguf-ud-q4_k_m 22B dev + Gemma 3 12B q4 (sd.cpp) Lightricks43±7*local (slow)~34GBhow to run › Sci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains
63Mochi 1 · bf16 Genmo42±2local~25GBhow to run › LMArena leaderboard datasetGenAI-Arena (TIGER-Lab) text-to-video leaderboardVideoPhy / VideoPhy-2 official repository leaderboardsT2VPhysBench: A First-Principles Benchmark for Physical Consistency in Text-to-Video Generation
65HunyuanVideo · bf16 Tencent41±3local (slow)~32GBhow to run › VideoPhy / VideoPhy-2 official repository leaderboardsSkyReels-V2: Infinite-length Film Generative ModelVideoVerse: Does Your T2V Generator Have World Model Capability to Synthesize Videos?Training-Free Efficient Video Generation via Dynamic Token Carving
68T2V-Turbo (VC2) · fp32 UC Santa Barbara41±3†local~9.4GBhow to run › GenAI-Arena (TIGER-Lab) text-to-video leaderboardT2V-Turbo-v2: Enhancing Video Generation Model Post-Training through Data, Reward, and Conditional Guidance Design (arXiv 2410.05677v3)T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback (arXiv:2405.18750v2)
70CogVideoX-2B · fp16 Zhipu AI (THUDM)39±2local~5.3GBhow to run › VideoPhy / VideoPhy-2 official repository leaderboardsGenAI-Arena (TIGER-Lab) text-to-video leaderboardChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation
71CogVideoX1.5-5B · bf16 + sequential CPU offload Zhipu AI (THUDM)38±3local (slow)~38GBhow to run › LOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)VideoVerse: Does Your T2V Generator Have World Model Capability to Synthesize Videos?SkyReels-V2: Infinite-length Film Generative ModelSci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains
72Allegro · bf16 Rhymes AI38±2local~9.3GBhow to run › LOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)GenAI-Arena (TIGER-Lab) text-to-video leaderboard
73VideoCrafter2 · fp32 Tencent AI Lab37±2local~9.4GBhow to run › VideoPhy / VideoPhy-2 official repository leaderboardsLOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)GenAI-Arena (TIGER-Lab) text-to-video leaderboardChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation
76LaVie (base T2V) · fp16 Shanghai AI Lab (Vchitect)35±2local~4.9GBhow to run › LOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)VideoPhy / VideoPhy-2 official repository leaderboardsGenAI-Arena (TIGER-Lab) text-to-video leaderboardChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation
77LTX-Video 0.9 (2B) · bf16 Lightricks35±3*local~5.3GBhow to run › GenAI-Arena (TIGER-Lab) text-to-video leaderboard
79Open-Sora 1.1 · bf16 HPC-AI Tech34±6*local~14GBhow to run › ChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation
81NOVA-0.6B T2V · fp16 BAAI32±3*local~9.3GBhow to run › LOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)
82Hotshot-XL · fp16 Hotshot31±2local~10GBhow to run › EvalCrafter: original 2023 text-to-video evaluationsLOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)
85ZeroScope v2 576w · fp16 cerspense30±3local~7.9GBhow to run › VideoPhy / VideoPhy-2 official repository leaderboardsChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video GenerationFETV: A Benchmark for Fine-Grained Evaluation of Open-Domain Text-to-Video Generation (arXiv:2311.01813v3)
87VideoCrafter1 T2V · official script (checkpoint as stored) Tencent AI Lab / HKUST29±2local~8.1GBhow to run › EvalCrafter: original 2023 text-to-video evaluations
89Open-Sora 1.2 · bf16 HPC-AI Tech28±3*local (slow)~39GBhow to run › GenAI-Arena (TIGER-Lab) text-to-video leaderboardChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation
90ModelScope T2V 1.7B · fp16 Alibaba DAMO Academy28±2local~4.6GBhow to run › EvalCrafter: original 2023 text-to-video evaluationsLOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)GenAI-Arena (TIGER-Lab) text-to-video leaderboardChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video GenerationFETV: A Benchmark for Fine-Grained Evaluation of Open-Domain Text-to-Video Generation (arXiv:2311.01813v3)
91LVDM (text-to-video) · fp32 HKUST / Tencent AI Lab28±3*local~6.1GBhow to run › LOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)
93ZeroScope v2 XL (576w → XL pipeline) · fp16 + model CPU offload (576w base, then XL vid2vid) cerspense27±2local~15GBhow to run › EvalCrafter: original 2023 text-to-video evaluations
94CogVideo 9B (original) · fp16 (--fp16), both stages Tsinghua / Zhipu26±7*local (slow)~52GBhow to run › FETV: A Benchmark for Fine-Grained Evaluation of Open-Domain Text-to-Video Generation (arXiv:2311.01813v3)
—Kandinsky 6.0 Video Lite · bf16 + model CPU offload Sbernot yet calibratedlocal (slow)~34GBhow to run › Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation (arXiv:2610.05608v1)
—LTX-2.5 · fp8 Lightricksnot yet calibratedlocal~14GBhow to run › Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation (arXiv:2610.05608v1)

+ 63 more don't fit your machine or budget · show them

97 of 97 catalogue models listed here · sources and licences

* new or thinly tested: ranked on too few benchmark results · † only the maker’s own numbers so far · ~RAM is estimated; full context may need more memory