Wan 2.5 Preview
Alibaba's open-weight video generation model with native audio synchronization, available in text-to-video and image-to-video variants.
Overview
What is Wan 2.5 Preview
Wan 2.5 Preview is Alibaba Cloud's open-weight AI video model, released in September 2025, that generates up to 10-second clips from text or a still image, with synchronized audio and support for 1080p resolution. It powers both text-to-video and image-to-video workflows under the same architecture.
Running it privately on Venice
On Venice, Wan 2.5 Preview runs with full prompt privacy — your inputs are anonymized and not stored. This means you can generate videos without surveillance or data retention, ideal for creators and enterprises prioritizing sovereignty. As an open and uncensored model, it enables permissionless innovation while maintaining audio-visual fidelity.
Assessment
Strengths and limitations
- Native audio-video synchronization: generates ambient sound, music, and lip-synced speech in one pass, eliminating post-production dubbing.
- Dual input modes: supports both text-to-video and image-to-video generation, enabling animation of stills with motion cues.
- Open weights allow self-hosting, fine-tuning, and auditability — rare among high-fidelity video models.
- Strong prompt adherence with cinematic control language (e.g., 'dolly shot', 'slow motion').
- Available on Venice with zero retention: prompts are not stored or profiled.
- Limited to short clips (5s or 10s), making it unsuitable for long-form content.
- Audio quality can be robotic in complex dialogue; background music and effects are more reliable.
- Faces and fine details may drift during motion, especially in image-to-video mode.
- No support for custom motion curves or frame-level editing in the preview version.
Samples
Sample outputs
Generated on Venice with our standard prompt suite — the same prompts we run through every model of this type, so you can judge it like-for-like.
Slow aerial drone shot gliding over a misty mountain valley at golden hour, sunlight piercing clouds onto a winding river, ancient pine forests on either side, ultra-smooth motion, professional color grading, atmospheric haze, 4K
Calm ocean waves rolling onto a black-sand beach at sunrise, golden light on wet sand, foam dissolving into the shore, a single silhouetted figure at the waterline, smooth continuous forward push, serene cinematic atmosphere
A slow tracking shot through a rain-soaked Tokyo street at night, neon reflecting in puddles, steam rising from a food cart, a person with a translucent umbrella, shallow depth of field, teal-and-orange grade, smooth steady camera
Capabilities
What it supports
- Text to video
- Image to video
- Reference to video
- Native audio generation
Variants
Wan 2.5 Preview model variants
Wan 2.5 Preview runs on Venice as 2 variants of the same underlying model. Pick by what you're starting from: a written prompt, a still image, reference images, or an existing clip. Each variant is its own model id on the API; the generation quality is the same across the family.
| Variant | What it is | Clip lengths | Resolutions | Aspect ratios | Audio | Model ID |
|---|---|---|---|---|---|---|
| Text to Videoflagship | Generate a clip from a written prompt | 5s, 10s | 1080p, 720p, 480p | 16:9, 9:16, 1:1 | wan-2.5-preview-text-to-video | |
| Image to Video | Animate a still image into motion | 5s, 10s | 1080p, 720p, 480p | — | wan-2.5-preview-image-to-video |
Capability data comes straight from the Venice model API and refreshes with every catalog ingest. The specs and pricing on this page are captured from the flagship variant; pass the model id of the variant you want to the API.
Wan 2.5 Preview Text to Video
Generate a clip from a written prompt. Supports clips of 5s, 10s, 1080p, 720p, 480p output, 16:9, 9:16, 1:1 aspect ratios, with native audio.
wan-2.5-preview-text-to-videoWan 2.5 Preview Image to Video
Animate a still image into motion. Supports clips of 5s, 10s, 1080p, 720p, 480p output, with native audio.
wan-2.5-preview-image-to-videoSpecifications
Datasheet
- Maker
- Alibaba Cloud (Tongyi Labs)
- Released
- September 2025
- Modality
- Text-to-video, image-to-video
- Max resolution
- 1080p
- Resolutions
- 1080p, 720p, 480p
- Clip lengths
- 5s, 10s
- Mode
- text-to-video
- Aspect ratios
- 16:9, 9:16, 1:1
- Audio
- Yes
- Privacy on Venice
- Anonymized — prompts not stored
- Available on Venice since
- Sep 2025
- License
- Open weights (permissive license)
API
Call it from your code
Venice exposes this model through the REST API. Queue a generation with the model id.
curl https://api.venice.ai/api/v1/video/queue \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "wan-2.5-preview-text-to-video",
"prompt": "Aerial drone shot over a misty mountain valley at golden hour"
}'
# Use the returned queue_id with https://api.venice.ai/api/v1/video/retrieve.
# Call /video/complete after downloading if needed.Pricing
What it costs on Venice
Pay per clip on Venice — price scales with resolution and duration (5s–10s), from $0.28.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
Alternatives
How it compares
| Model | Max resolution | Strongest at | Open weights | Price (Venice) |
|---|---|---|---|---|
| Wan 2.5 Preview | 1080p | Open, audio-synced clips | Yes | from $0.28 |
| Wan 2.7 | 1080p | Longer clips, higher fidelity | Yes | from $0.55 |
| Kling O3 Pro | 4K | Cinematic realism | No | from $0.46 |
| Vidu Q3 | 1080p | Fast generation | No | from $0.27 |
Only open-weight model here with native audio sync — ideal for private, self-hostable video workflows.
Use cases
What it is good for
- 01Social media shorts with built-in audio — TikTok, Instagram Reels, YouTube Shorts.
- 02Product animations from still images with voiceover or sound effects.
- 03Rapid prototyping of ad concepts with synchronized audio and visuals.
- 04AI-generated storyboards or pitch reels with minimal post-processing.
- 05Privacy-sensitive video creation where data sovereignty is required.
Prompting
Getting better results
Use cinematic terms like 'slow zoom', 'overhead shot', or 'time-lapse' to improve camera control.
For image-to-video, describe motion explicitly: 'the character turns head left, waves hand'.
Include audio cues in quotes: 'with upbeat jazz music' or 'voiceover saying welcome to summer'.
Start at 480p for fast iteration, then scale to 1080p for final output.
Version history
Predecessor with lower fidelity and no native audio.
Current — adds audio sync and improved realism.
FAQ
Frequently asked questions
Wan 2.5 Preview is Alibaba Cloud's open-weight AI video model released in September 2025. It generates up to 10-second video clips from text or still images, with synchronized audio and support for 1080p resolution. It powers both text-to-video and image-to-video workflows.
On Venice, pricing starts at $0.28 per 480p, 5-second clip, scaling up to $0.83 for 1080p at the same duration. You pay per clip — no subscription required.
It is not free, but it is open weights — meaning the model parameters are available under a permissive license, allowing self-hosting, fine-tuning, and modification without vendor lock-in.
Yes. It natively generates synchronized audio, including ambient sound, music, and speech with lip-sync, all within a single generation pass — no separate dubbing needed.
It supports 1080p, 720p, and 480p resolutions, with aspect ratios including 16:9 (landscape), 9:16 (portrait), and 1:1 (square).
Yes. The variant 'wan-2.5-preview-image-to-video' animates still images into 5s or 10s clips with motion and audio, based on a text prompt describing the desired action.
No. On Venice, your prompts are anonymized and not stored — ensuring full privacy and zero retention. Your data remains yours.
Yes. Venice runs Wan 2.5 Preview with anonymized privacy — no prompts are stored, profiled, or used for training. This makes it ideal for sensitive or commercial use cases requiring data sovereignty.
No. Wan 2.5 Preview is a generative video model and does not support tool use, function calling, or reasoning over external APIs. It generates video from prompts or images only.
Yes. The image-to-video variant 'wan-2.5-preview-image-to-video' accepts a still image as input and animates it into motion based on a text prompt.
Wan 2.7 is a newer version with longer clip support, improved motion consistency, and higher fidelity. Wan 2.5 Preview is an earlier release optimized for short, audio-synced clips and open deployment.
No. Wan 2.5 Preview does not perform web search or access real-time information. It generates video based solely on the provided prompt or image input.
It has limited reasoning — it follows cinematic language and prompt structure well, but does not perform logical inference or multi-step planning like a language model.
It generates clips of exactly 5 seconds or 10 seconds — no intermediate durations. This is fixed in the current preview version.
The model ID for the image-to-video variant is 'wan-2.5-preview-image-to-video'.
Yes. Since it has open weights, you can self-host and run Wan 2.5 Preview on your own infrastructure, subject to the license terms.
Run Wan 2.5 Preview privately
No prompt logging. No data used for training.