Native 4K Cinematic Output
Uses a dual-stage diffusion process to generate 1080p content with an integrated 4K refiner for sharp textures.
Wan 2.5
Alibaba Wan 2.5 is a flagship unified multimodal diffusion model for high-fidelity video. It offers native 4K upscaling and flow-latent consistency to minimize background morphing, delivering pro-grade cinematic results via the GPTProto API.
API Usage Examples
curl --request POST "https://gptproto.com/api/v3/alibaba/wan-2.5/image-to-video" \
--header "Authorization: Bearer $GPTPROTO_API_KEY" \
--header "Content-Type: application/json" \
--data '{
"image": "",
"audio": "",
"prompt": "A tiny origami fox sailing a teacup across a moonlit puddle",
"negative_prompt": "",
"resolution": "720p",
"duration": 5,
"enable_prompt_expansion": false,
"seed": -1
}'Wan 2.5 pricing
Start from the cost of a single sample and pick a testing budget. GPTProto rates are 10% below list price.
Your usage
$0.45 / image
Related Models
All Models
| Model | Resolution | Input → Output | ||
|---|---|---|---|---|
| $0.04 per second | 480p · 720p · 1080p | Input: TextInput: Image Output: ImageOutput: Video | |
| $0.22 per time | 2K · more | Input: TextInput: Image Output: Video | ||
| $0.09 per time | more | Input: TextInput: ImageInput: VideoInput: Audio Output: Video | ||
| — | $0.11 per second | 480p · 720p | Input: TextInput: ImageInput: VideoInput: Audio Output: Video | |
| $0.34 per second | 4k | Input: TextInput: Image Output: Video | ||
| — | $0.02 per second | 480p · 720p | Input: TextInput: ImageInput: VideoInput: Audio Output: Video | |
| $0.34 per second | 4k | Input: TextInput: ImageInput: VideoInput: Audio Output: Video | ||
| — | $0.04 per second | 480p · 720p | Input: TextInput: ImageInput: VideoInput: Audio Output: Video | |
| $0.02 per second | 720p · 1080p · more | Input: ImageInput: VideoInput: Audio Output: Video | ||
| $0.09 per second | — | Input: TextInput: ImageInput: VideoInput: Audio Output: Video | ||
| $0.03 per second | 540p · 720p · 1080p | Input: TextInput: Image Output: Video | ||
| $0.04 per second | 540p · 720p · 1080p | Input: TextInput: Image Output: Video | ||
| $0.09 per second | 720p · 1080p · 2k | Input: TextInput: ImageInput: VideoInput: Audio Output: Video | ||
| — | $0.05 per time | — | Input: Video Output: Video | |
| — | $1.20 per time | — | Input: TextInput: ImageInput: Video Output: Video | |
| — | $3.20 per time | — | Input: TextInput: ImageInput: Video Output: Video | |
| $0.03 per second | 1080p · more | Input: Image Output: Video | ||
| $0.44 per time | — | Input: TextInput: Image Output: Video | ||
| $0.09 per time | 1K · more | Input: TextInput: Image Output: Video | ||
| — | $0.50 per time | — | Input: TextInput: ImageInput: VideoInput: Audio Output: Video | |
| — | $0.30 per second | 720p · 1080p | Input: TextInput: Image Output: Video | |
| — | $0.10 per second | 720p | Input: TextInput: Image Output: Video | |
| $0.28 per time | — | Input: Image Output: Video | ||
| $0.09 per time | — | Input: Image Output: Video | ||
| $0.39 per time | — | Input: Image Output: Video |
Alibaba Wan 2.5 Standout Features
Harness the technical advantages of the Alibaba Wan 2.5 unified diffusion architecture.
Direct Camera Control
Provides granular API access to dolly, pan, tilt, and zoom with sub-degree precision for professional directing.
Flow-Latent Consistency
Reduces background morphing by 30% using a refined flow-matching architecture that preserves static geometry.
High-Fidelity Subject Identity
Maintains 98% visual identity preservation in Image-to-Video tasks, ideal for consistent brand storytelling.
Alibaba Wan 2.5 Frequently Asked Questions
Get technical insights and integration support for Alibaba Wan 2.5, the industry-leading video generation model.
How does Alibaba Wan 2.5 compare to the official API?
Using Alibaba Wan 2.5 through GPTProto provides a unified USD billing system and automatic failover across global high-availability clusters. While the core Alibaba model remains the same, our platform simplifies the SDK experience and offers enterprise-grade reliability and 5% volume discounts that aren't always available through direct retail channels.
Is data sent to Alibaba Wan 2.5 used for training?
No. We prioritize your privacy and intellectual property. All data transmitted to the Alibaba Wan 2.5 model via our aggregation platform is strictly excluded from any training cycles. Your proprietary prompts, images, and generated video assets remain confidential and are only used to fulfill your specific inference requests.
What is the typical generation time for Alibaba Wan 2.5?
Alibaba Wan 2.5 is optimized for high-throughput enterprise needs. On our standard NVIDIA H100 clusters, generating a 5-second high-definition 1080p video clip typically takes between 40 and 60 seconds. High-res 4K upscaling or longer 30-second segments will naturally increase the processing time accordingly.
Can I migrate from OpenAI Sora to Alibaba Wan 2.5?
Yes, migration is straightforward. Alibaba Wan 2.5 on our platform utilizes an OpenAI-compatible request structure. You can simply update your base URL and model parameter to 'wan-2.5'. Our gateway handles the internal parameter mapping, allowing you to leverage Alibaba's superior temporal stability without rewriting your entire codebase.
Does Alibaba Wan 2.5 support fine-tuning?
Currently, character-consistent fine-tuning (similar to DreamBooth for Video) is available for Alibaba Wan 2.5 exclusively for Enterprise Tier customers. This allows brands to train the model on specific product assets or spokespeople to ensure 100% visual consistency across large-scale marketing campaigns and social media content.
What cinematic controls does Alibaba Wan 2.5 offer?
Alibaba Wan 2.5 features Direct Camera Control (DCC), giving you granular API access to camera parameters. You can specify precise dolly, pan, tilt, and zoom movements. This is possible because Alibaba trained the model on over 100,000 hours of professionally directed cinema data, allowing it to understand complex directorial intent.
Related Articles
Guides, comparisons, and updates related to this model.
All Articles
wan2.2-animate: Quality Over Speed in AI
While other models rush output, wan2.2-animate focuses on cinematic visual fidelity and prompt adherence. Learn how to optimize your video workflow today.
Wan 2.2 Animate: Real Image-to-Video
Stop struggling with AI video flickering. Learn how to configure wan 2.2 animate for precise character consistency and fluid motion. Read the full guide.
WAN 2.5 Guide: Deep Dive into the Newest AI Video Model and its Features
Explore the massive shift in generative video with WAN 2.5. Learn about its 1080p capabilities, the closed-source controversy, and how it compares to Sora in terms of cost and quality for professional creators.
Wan 2.5: The Future of 4K AI Video Generation
Discover how Wan 2.5 transforms images into 4K cinematic videos with AI. Learn features, pricing, and GPT Proto API options.