Two-expert MoE diffusion design
According to the model card, the A14B series pairs a high-noise expert for early layout with a low-noise expert for later detail, totaling 27B parameters with 14B active per step.
Open Source Model Profile · Wan-AI
Wan2.2-I2V-A14B
Wan2.2-I2V-A14B is an image-to-video generative model from Wan-AI. According to the model card, it uses a Mixture-of-Experts design and supports both 480P and 720P generation.
- Publisher
- Wan-AI
- Task
- image-to-video
- Model type
- Unknown
- License
- apache-2.0
- Library
- wan2.2
- Publication status
- Accepted · not indexed
Model overview
Wan2.2-I2V-A14B is published by Wan-AI for image-to-video generation. According to the model card, it is built with a Mixture-of-Experts architecture and supports 480P and 720P resolutions with more stable synthesis and reduced unrealistic camera movement. The card describes a two-expert design with about 14B parameters per expert, totaling 27B with 14B active per step.
Recorded capabilities
480P and 720P image-to-video
According to the model card, this repository's I2V-A14B model generates video from images at 480P and 720P resolutions.
Diffusers integration
According to the model card, Wan2.2 text-to-video, image-to-video, and text-image-to-video variants have been integrated into Diffusers.
Documented 80GB single-GPU path
According to the model card, the documented single-GPU 1280x720 command can run on a GPU with at least 80GB of VRAM.
Use cases in the source record
- Image-to-video generation at 480P or 720P, with optional text guidance or prompt extension that generates a prompt from the input image.
- Diffusers-integrated image-to-video workflows, which the publisher notes cover the I2V-A14B variant.
- Single-GPU runs on at least 80GB of VRAM or multi-GPU FSDP with DeepSpeed Ulysses runs, following the model card's documented commands.
Limitations and unknowns
- No architecture or parameter count was captured in config metadata for this record.
- Publisher performance claims, such as top performance among open and closed models, are model-card statements and were not independently verified by Ethen.
- Provider state is historical snapshot data and should be refreshed before being presented as current.
Source and provenance
Source: Wan-AI/Wan2.2-I2V-A14B
Captured: Unknown. Processed: 2026-09-07T19:34:38.530356+00:00.
Wan2.2 💜 Wan ｜ 🖥️ GitHub | 🤗 Hugging Face | 🤖 ModelScope | 📑 Technical Report | 📑 Blog | 💬 WeChat Group | 📖 Discord Wan: Open and Advanced Large-Scale Video Generative Models We are excited to introduce Wan2.2 , a major upgrade to our foundational video models. With Wan2.2 , we have focused on incorporating the following innovations: 👍 Effective MoE Architecture : Wan2.2 introduces a Mixture-of-Experts (MoE) architecture into video diffusion models. By separating the denoising process cross timesteps with specialized powerful expert models, this enlarges the overall model capacity while maintaining the same computational co…
F001F002F003F004F007F008F010F012F013F015F017F018F019F024