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

Best image-to-video models to run locally on a 16GB graphics card

A leaderboard of the best image-to-video models you can run locally on a 16GB graphics card (with 32GB of system RAM). Wan 2.2 I2V A14B is #1 right now, followed by HunyuanVideo 1.5 and Wan 2.2 TI2V 5B. Models are ranked on human votes and tests that turn an image into a video. Each model shows the download that fits and the command to run it.

5 models · scores combine 16 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 image to video evaluations only. Compare scores within this task. Models without task measurements are excluded.

Models for 32GB RAM · 16GB graphics card → ~16GB fast · local only · reset
+ more
card[RTX 4060 Ti 16GB][RTX 4070 Ti Super][RTX 4080][RTX 4080 Super][RTX 5060 Ti 16GB][RTX 5070 Ti][RTX 5080][RTX 3080 Laptop][RTX 3080 Ti Laptop][RTX 4090 Laptop][RTX 5080 Laptop][RTX A4000][RX 7600 XT][RX 7800 XT][RX 7900 GRE][RX 9060 XT][RX 9070][RX 9070 XT]

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=16&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
37Wan 2.2 I2V A14B · gguf-q4_k_m ×2 experts + umt5 q8_0 (sd.cpp) Alibaba48±2local (slow)~31GBhow to run › WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors (arXiv v1)PAI-Bench: A Comprehensive Benchmark For Physical AI (arXiv:2512.01989v1)CRONOS: Benchmarking Counterfactual Physical Consistency in Video Models (arXiv:2605.23699v1), Table 1Cosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)VGI-Bench: Probing Visual Intelligence in Video Generation ModelsPhysics-IQ Verified
57HunyuanVideo 1.5 · fp16 720p T2V + Qwen2.5-VL q8_0 (sd.cpp) Tencent40±4*local (slow)~34GBhow to run › LMArena leaderboard dataset
58Wan 2.2 TI2V 5B · bf16 Alibaba40±4local~13GBhow to run › Cosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)Target-Bench: Can Video World Models Achieve Mapless Path Planning with Semantic Targets?PAI-Bench: A Comprehensive Benchmark For Physical AI (arXiv:2512.01989v1)What-If World: A Causal Benchmark for General World Models in Embodied Scenarios (arXiv:2605.27589v1), Table 2
64Wan 2.2 A14B · gguf-q4_k_m ×2 experts + umt5 q8_0 (sd.cpp) Alibaba37±4*local (slow)~31GBhow to run › LMArena leaderboard dataset
71LTX-2.3 · gguf-ud-q4_k_m 22B dev + Gemma 3 12B q4 (sd.cpp) Lightricks23±3*local (slow)~34GBhow to run › WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors (arXiv v1)

+ 69 more don't fit your machine or budget · show them

74 of 74 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