Wan2.2-TI2V-5B-Diffusers

It is a 5 billion-parameter text-to-video (TI2V) generative model designed for high-definition video generation at 720P resolution (1280×704 or 704×1280) with 24fps. Built using the Wan2.2-VAE architecture, it achieves a compression ratio of 16×16×4, enabling efficient deployment on consumer-grade GPUs like the RTX 4090. The model supports both text-to-video and image-to-video generation within a unified framework.

Key Features

• Tasks - Text-to-video and image-to-video generation.
• Mixture-of-Experts (MoE) Architecture - Employs specialized experts for denoising across different stages of the diffusion process. A high-noise expert handles early-stage layout generation, while a low-noise expert refines details in later stages, doubling model capacity without increasing computational cost.
• Enhanced Aesthetics and Motion - Trained on expanded datasets (65.6% more images and 83.2% more videos than Wan2.1), improving motion generation, semantic diversity, and cinematic-style aesthetics (lighting, composition, color tone).
• Performance - Achieves top performance among open-source and closed-source models on the Wan-Bench 2.0 benchmark.
• Efficient Inference - Generates 5-second 720P videos in ~9 minutes on a single GPU. Supports offloading models to CPU and mixed-precision for memory efficiency. Multi-GPU inference is enabled via FSDP + DeepSpeed Ulysses (--dit_fsdp, --t5_fsdp).
• Deployment - Runs on consumer GPUs with ≥24GB VRAM (e.g., RTX 4090). Multi-GPU configurations reduce inference time further.

---

The model is a component of the video generation pipeline, consisting of:

• Text encoder: ~6B parameters,
• Transformer: ~5B parameters,
• VAE: ~705M parameters.

Total: ~11B parameters

Announce Date: 28.07.2025
Parameters: 5B
Developer: Alibaba Wan Team
Diffusers Version: 0.35.0.dev0
vLLM-Omni Version: 0.12.0rc1
License: Apache 2.0

Use our pre-built public endpoints for free to test inference and explore Wan2.2-TI2V-5B-Diffusers capabilities. You can obtain an API access token on the token management page after registration and verification.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.