Maximize Productivity with Novita AI’s Wan 2.2 I2V API

Novita AI proudly introduces the Wan 2.2 I2V API, a cutting-edge tool for image-to-video (I2V) generation that revolutionizes video content creation. As an extension of Alibaba’s Wan 2.2 T2V, this API leverages Mixture-of-Experts (MoE) architecture and advanced compression techniques to deliver 720P videos at 24fps, optimized for consumer-grade GPUs. This article dives into what Wan 2.2 I2V is, its features, and how it can transform video creation workflows.

What is Wan 2.2 I2V?

Wan 2.2 I2V is an advanced AI-driven video generator that converts text or image inputs into short video clips. The term “I2V” stands for image-to-video, indicating one of its generation modes (it also supports text-to-video). Wan 2.2 represents the second major release of the Wan model series, bringing significant upgrades over version 2.1. It uses a cutting-edge Mixture-of-Experts (MoE) diffusion architecture to achieve high-quality 720p resolution video output from prompts. The model is open-source (Apache 2.0 licensed) and designed to deliver professional-looking results on standard consumer hardware.

Compact and Versatile TI2V Solution: Wan2.2 introduces an open-source 5B model powered by its advanced Wan2.2-VAE, achieving an impressive 16×16×4 compression ratio. This lightweight model seamlessly supports both text-to-video (T2V) and image-to-video (I2V) generation at 720P resolution with 24fps. Optimized for consumer-grade GPUs like the NVIDIA 4090, it stands as one of the fastest 720P@24fps models available, making it an ideal solution for both industrial applications and academic research.

Wan 2.2 I2V Architecture and Image Understanding

Two Types of MOE

The Mixture-of-Experts (MoE) diffusion model in Wan 2.2 utilizes both high-noise and low-noise expert networks to better handle varying complexities in image inputs. For instance, the high-noise network excels at processing intricate image details, while the low-noise network focuses on overall scene composition. This division of labor enhances the model’s ability to analyze and interpret image content effectively.

Compression and Spatio-Temporal Consistency

The model employs Wan-VAE (Variational Autoencoder) for spatio-temporal compression, achieving 64× compression (4× temporally, 16×16 spatially). This enables efficient encoding and decoding of video frames while preserving essential details and temporal coherence. This compression technique not only improves generation efficiency but also ensures a smooth and natural transition from static images to dynamic videos.

Maintaining temporal consistency is crucial when generating videos from images, especially for aspects like lighting changes and object movements. The 3D compression architecture of Wan-VAE guarantees visual fluidity and correctly extends image content over time, ensuring high-quality video outputs.

Wan 2.2 I2V Key Features

Feature	Description
🎥 Cinematic Aesthetic Controls	Provides cinematic-level aesthetic control with professional film-style parameters. Prompts can specify lighting, color tones, camera angles, and composition details to influence the look of the generated video.
🤖 Complex Motion & Stability	Excels at reproducing large-scale, complex motions smoothly. Handles fast camera movements (pans, tilts, zooms) and multiple moving subjects with improved stability. Thanks to MoE experts, it yields smoother motion with fewer jitter or continuity issues.
🎯 Precise Semantic Compliance	Demonstrates a better understanding of complex scenes and multi-object interactions, generating outputs that closely match the user’s prompt intent. Expanded training data and refined diffusion strategies improve consistency and reliability.

What Work Process Optimizations are in Wan 2.2?

Wan 2.2 I2V vs Wan 2.1 I2V

Wan 2.2 I2V vs Wan 2.1 I2V: Architecture

Category	Wan 2.1	Wan 2.2
Diffusion Model	Dense diffusion architecture: A single model handled all denoising timesteps.	Mixture-of-Experts (MoE) diffusion: Two specialized sub-models handle different noise levels, with one processing high-noise early timesteps and the other handling low-noise later timesteps. This improves detail and coherence.
Model Size & Parameters	~14B parameters for text-to-video and image-to-video tasks. Smaller variants (e.g., 1.3B) were available for quicker prototyping.	~27B parameters (2×14B experts), but only one expert is active at a time. Introduced a new 5B hybrid model for TI2V (text and image conditioning) capable of 720p output, filling the role of 2.1’s smaller model but with better fidelity.
Training Data & Aesthetic Labels	Limited dataset with basic descriptors for prompt control.	Trained on a dataset with 65% more images and 83% more video clips. Introduced cinematic tags (e.g., lighting, color, composition) to enable finer style control compared to 2.1’s basic descriptors.
Underlying Components	Used Wan-VAE for 1080p encodings, focusing on maintaining temporal consistency.	Improved Wan-VAE and MoE diffusion integration for a better balance between quality and resource use. Added FlashAttention for faster transformer operations, enhancing performance compared to 2.1.
Features	Supported T2V, I2V, and editing with the VACE framework. LoRA fine-tuning was fully supported.	Supported T2V, I2V, and improved style transfer. No VACE framework yet and only limited LoRA compatibility.

Wan 2.2 I2V vs Wan 2.1 I2V: Perfromance

Wan 2.2 I2V vs Wan 2.1 I2V: Generation

Cost and Access of Wan 2.2 I2V

Hardware Costs

• I2V 5B Model:
  • Minimum VRAM Requirement: 24GB.
  • Minimum GPU Model: NVIDIA RTX 4090.
  • Minimum GPU Quantity: 1.
  • Single GPU Speed: Approximately 524.8 seconds at 720P resolution.
  • Approximate GPU Price: The NVIDIA RTX 4090 was released on October 12, 2022, with a starting price of $1,599.
• I2V A14B Model:
  • 480P Resolution:
    • Minimum VRAM Requirement: 40GB.
    • Minimum GPU Model: NVIDIA A100 40GB.
    • Minimum GPU Quantity: 1.
    • Single GPU Speed: Approximately 810.0 seconds.
    • Approximate GPU Price: The NVIDIA A100 40GB is listed at $13,135.
  • 720P Resolution:
    • Minimum VRAM Requirement: 80GB.
    • Minimum GPU Model: NVIDIA H100 80GB.
    • Minimum GPU Quantity: 1.
    • Single GPU Speed: Approximately 1,055.9 seconds.
    • Approximate GPU Price: Pricing information for the NVIDIA H100 80GB is not available in the provided sources.

API Costs

Novita AI is an AI cloud platform that offers developers an easy way to deploy AI models using our simple API, while also providing the affordable and reliable GPU cloud for building and scaling.

Model	Price	Resolution	Generation Time
Wan 2.1 I2V	$0.3/video	1280*720	5s
Wan 2.2 I2V	$0.4/video	1080P	5s

Wan 2.2 I2V Access Guide

Step 1: Log In and Access the Model Library

Log in to your account and click on the Model Library button.

Step 2: Choose Your Model

Browse through the available options and select the model that suits your needs.

Step 3: Get Your API Key

To authenticate with the API, we will provide you with a new API key. Entering the “Settings“ page, you can copy the API key as indicated in the image.

Step 4: Install the API

Install API using the package manager specific to your programming language.

After installation, import the necessary libraries into your development environment. Initialize the API with your API key to start interacting with Novita AI LLM. This is an example of using chat completions API for python users.

import requests

url = "https://api.novita.ai/v3/async/wan-2.2-i2v"

payload = {
    "input": {
        "prompt": "<string>",
        "negative_prompt": "<string>",
        "img_url": "<string>"
    },
    "parameters": {
        "resolution": "<string>",
        "duration": 123,
        "prompt_extend": True,
        "seed": 123
    }
}
headers = {
    "Content-Type": "<content-type>",
    "Authorization": "<authorization>"
}

response = requests.post(url, json=payload, headers=headers)

print(response.json())

Wan 2.2 I2V: Common Issues & Fixes

Issue	Fix
Flickering Frames	Increase diffusion steps or frame rate; use I2V mode; stabilize in post-processing.
Slow/Out-of-Memory	Use 5B model or lower resolution; enable memory optimizations; consider cloud GPUs.
Prompt Mismatch	Simplify prompts; use negative prompts; refine iteratively for better results.
Blurry Output	Use “DetailZ” LoRA; request sharper details in prompts; sharpen or upscale in post.
Inconsistent Objects	Use reference images in I2V mode; generate shorter clips and chain them; keep prompts steady.
No Audio	Add audio in post-production; use AI tools for music or voiceover and sync with visuals.

Pros and Cons of Wan 2.2 I2V for Small Businesses

Pros:

1. Lower Content Production Costs: No need for filming or a production team, saving budget. Ideal for startups with limited resources.
2. Faster Creative Turnaround: Videos can be generated in minutes, allowing quick responses to trends and fast prototyping.
3. Accessible on Consumer Hardware: Runs on standard PCs with decent GPUs, avoiding the need for costly specialized hardware.
4. Creative Flexibility: Supports various styles and scenes, catering to diverse needs by simply adjusting prompts.
5. Open Source & Evolving Tool: Community support ensures continuous updates, reducing the risk of obsolescence.

Cons:

1. Learning Curve and Expertise: Requires AI knowledge or time to learn prompt crafting, making it challenging for non-tech-savvy users.
2. Computational Costs: Large-scale video generation incurs ongoing GPU and energy costs, which must be budgeted.
3. Quality Limitations: Outputs are limited to 720p and may require post-editing for high-quality needs.
4. Consistency and Branding: Generated content may lack consistency across videos, needing extra curation for brand alignment.
5. Ethical and Legal Considerations: Issues like copyright, transparency, and audience trust must be carefully managed.

Future Trends in Wan 2.2 I2V Technology

Trend	Description
Higher Resolution	Support for 1080p+ resolution and longer video durations (10-15 seconds or full short films).
Audio & Interaction	Integration of audio generation and interactive editing (e.g., video-to-video enhancements).
Greater Control	Tools for storyboards, frame control, and consistent characters/branding across scenes.
Faster & Accessible	Near real-time video generation with optimized models and hardware advances (e.g., GPUs, cloud).
Broader Adoption	Use in entertainment, education, and advertising, with an ecosystem of plugins and community styles.
Competition & Collaboration	Open-source Wan leverages research advancements, driving innovation and hybrid models for quality.

The Wan 2.2 I2V API sets a new standard for video generation, offering cinematic aesthetic controls, precise motion handling, and unmatched efficiency. Whether you’re a creator, marketer, or researcher, Wan 2.2’s capabilities simplify workflows, reduce costs, and open up new creative possibilities. With its open-source foundation and robust API, Wan 2.2 I2V is the future of accessible and powerful video creation.

Frequently Asked Questions

Novita AI is the All-in-one cloud platform that empowers your AI ambitions. Integrated APIs, serverless, GPU Instance — the cost-effective tools you need. Eliminate infrastructure, start free, and make your AI vision a reality.

Recommend Reading

• Unleash Your Creativity: YouTube Videos Voiceovers Mastery
• 2024 Youtube Video Notes Taker AI Market and Leading Players
• Transforming Images with Ease: Image to Video AI API