Wan2.2 VRAM requirements
The next Wan release: mixture-of-experts models for text-to-video and image-to-video with 27 billion parameters, 14 billion of them active at each step, a 14B speech-to-video model, and TI2V-5B, a 5 billion parameter model for text or image to 720P video. Wan-AI's figures run from at least 24 GB for TI2V-5B to at least 80 GB for T2V-A14B, I2V-A14B, S2V-14B and TI2V-5B without offload. The cheapest single GPU for the first is the 24 GB A30 at Massed Compute ($0.35 an hour); for the second, the 80 GB A100 at Thunder Compute ($1.09).
What Wan-AI says each setting needs
| Model and setting | Stated GPU memory | Cheapest single GPU with that much | Per hour | |
|---|---|---|---|---|
TI2V-5B720P (1280×704) on one GPU, with --offload_model True --convert_model_dtype --t5_cpu | at least 24 GB | A30, 24 GB, Massed Compute | $0.35 | Rent → |
T2V-A14B720P (1280×720) on one GPU, with --offload_model True --convert_model_dtype | at least 80 GB | A100, 80 GB, Thunder Compute | $1.09 | Rent → |
I2V-A14B720P on one GPU, with --offload_model True --convert_model_dtype | at least 80 GB | A100, 80 GB, Thunder Compute | $1.09 | Rent → |
S2V-14B1024×704 speech-to-video on one GPU, with --offload_model True --convert_model_dtype | at least 80 GB | A100, 80 GB, Thunder Compute | $1.09 | Rent → |
TI2V-5B without offload720P on one GPU, faster: the same job without --offload_model True --convert_model_dtype --t5_cpu | at least 80 GB | A100, 80 GB, Thunder Compute | $1.09 | Rent → |
The memory figures are Wan-AI's, not our estimate. The price is the lowest on-demand rate for one GPU with at least that much memory, from our latest check; a card that holds a model is not a promise of its speed.
Before you rent
Each figure holds only with the options listed beside it, which move parts of the model to the CPU or convert its weights. TI2V-5B is the one model with two rows: it fits a 24 GB card with those options, and on a card with at least 80 GB the README says you can drop them to run faster, so compare the two hourly prices against how long your renders take.
In Wan-AI's words
- TI2V-5B: “This command can run on a GPU with at least 24GB VRAM (e.g, RTX 4090 GPU).”
- T2V-A14B: “This command can run on a GPU with at least 80GB VRAM.”
- I2V-A14B: “This command can run on a GPU with at least 80GB VRAM.”
- S2V-14B: “This command can run on a GPU with at least 80GB VRAM.”
- TI2V-5B without offload: “If you are running on a GPU with at least 80GB VRAM, you can remove the `--offload_model True`, `--convert_model_dtype` and `--t5_cpu` options to speed up execution.”
Source: the Wan2.2 README on GitHub, read 17 October 2026.
Text models are sized by a formula instead: LLM VRAM requirements for every open language model, or the LLM VRAM calculator for your own context and batch.
More models sized by their publishers' figures
- Whisper (speech recognition, OpenAI)
- Stable Diffusion 3.5 Medium (image generation, Stability AI)
- Wan2.1 (video generation, Wan-AI)
- HunyuanVideo (video generation, Tencent)
- olmOCR (OCR and PDF conversion, Ai2)
- MinerU (OCR and document parsing, OpenDataLab)
- GTE-Qwen2 (text embeddings, Alibaba)