Wan 2.2 vs HunyuanVideo for local video
Two open text-to-video models and the GPU memory they document
Facts from the reviewed catalog
| Specification | Wan2.2 T2V-A14B | HunyuanVideo |
|---|---|---|
| Task | Video | Video |
| Parameters | 27B | 13B |
| License | Apache 2.0 | Publisher terms: Tencent Hunyuan Community License Agreement |
| Commercial use | Allowedlicense text | Allowed with conditionsNot licensed in the EU, UK or South Korea; over 100M monthly users needs a Tencent license · license text |
| Quality · LMArena 2026-09-22 · text-to-video | 1,132 ± 151,116–1,147 · 10,458 votes · rank 42 · source | Arena rated hunyuan-video-1.5; this entry is the first HunyuanVideo · source |
| Memory | publisher's documented path from 80 GB (CUDA, conditional) | publisher's documented path from 45 GB (CUDA) |
| Checkpoint | official card | official card |
Only Wan2.2 T2V-A14B has a result in LMArena 2026-09-22 · text-to-video (1,132 ± 15); HunyuanVideo has none, so the two are not ranked from this source and no score is inferred for it. Blind pairwise votes by people on hosted outputs, CC BY 4.0. A rating is relative; overlapping intervals mean the order is not settled. Scale: Arena rating ± half the confidence interval.
Memory on three machines
Each machine's own reserve is added (graphics cards 2 GB; unified and system memory 25%, 2 to 8 GB). Planning estimates, not measured runs.
| Machine | Wan2.2 T2V-A14B | HunyuanVideo |
|---|---|---|
| RTX 4090 · 24 GB VRAM | 80 GBBelow the documented minimum | 45 GBBelow the documented minimum |
| RTX 5090 · 32 GB VRAM | 80 GBBelow the documented minimum | 45 GBBelow the documented minimum |
| DGX Spark · 128 GB unified | 80 GBWithin a conditional path (offload or quantization) | 45 GBWithin a conditional path (offload or quantization) |
Wan2.2 T2V-A14B: publisher's documented path from 80 GB (CUDA, conditional). HunyuanVideo: publisher's documented path from 45 GB (CUDA). Formulas
Everything that runs on each machine: RTX 4090 · 24 GB VRAM · RTX 5090 · 32 GB VRAM · DGX Spark · 128 GB unified
Which to choose
Local video needs far more memory than images. Wan2.2 T2V-A14B (27 billion parameters in two experts, Apache 2.0) documents an 80 GB conditional path; HunyuanVideo (13 billion) documents 45 GB and 60 GB paths, and Tencent's licence excludes the EU, the UK and South Korea. None of the consumer cards below reaches either documented minimum. Wan 2.2 has a rating in the LMArena text-to-video release; HunyuanVideo has none, because the Arena rated HunyuanVideo 1.5, a later checkpoint.
Files and run commands
The default file of the most downloaded collected source, or the commands each card documents. Not run by this site.
Wan2.2 T2V-A14B
Documented ways to run it
Copied as printed in the linked source (read 2026-10-18); not run by this site.
python generate.py --task t2v-A14B --size 1280*720 --ckpt_dir ./Wan2.2-T2V-A14B --offload_model True --convert_model_dtype --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage."torchrun --nproc_per_node=8 generate.py --task t2v-A14B --size 1280*720 --ckpt_dir ./Wan2.2-T2V-A14B --dit_fsdp --t5_fsdp --ulysses_size 8 --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage."HunyuanVideo
Documented ways to run it
Copied as printed in the linked source (read 2026-10-18); not run by this site.
cd HunyuanVideo
python3 sample_video.py \
--video-size 720 1280 \
--video-length 129 \
--infer-steps 50 \
--prompt "A cat walks on the grass, realistic style." \
--flow-reverse \
--use-cpu-offload \
--save-path ./resultscd HunyuanVideo
torchrun --nproc_per_node=8 sample_video.py \
--video-size 1280 720 \
--video-length 129 \
--infer-steps 50 \
--prompt "A cat walks on the grass, realistic style." \
--flow-reverse \
--seed 42 \
--ulysses-degree 8 \
--ring-degree 1 \
--save-path ./resultspython3 gradio_server.py --flow-reverse
# set SERVER_NAME and SERVER_PORT manually
# SERVER_NAME=0.0.0.0 SERVER_PORT=8081 python3 gradio_server.py --flow-reverseNumbers each publisher reports
Each card's own results with its own settings. Publishers choose different tests and settings, so these values are never compared across publishers here.
Wan2.2 T2V-A14B · reported by Wan-AI
HunyuanVideo · reported by Tencent Hunyuan
Limits and dates
The published quality benchmark does not test a local quantized build. Memory estimates are not speed measurements or guaranteed hardware fits. Repository size, when available, is not VRAM. Formulas
Page updated · Catalog reviewed · LMArena 2026-09-22 · text-to-video