Powered by Alibaba's open-source Wan series — built for Wan 3.0 the day it ships. Type a prompt, pick aspect and duration, watch it render in under 60 seconds. Commercial use, no watermark, 9 languages.
wan3pro.video is a browser-based interface to Alibaba's open-source Wan video series (Apache 2.0, on Hugging Face under Wan-AI), built to run Wan 3.0 the day Alibaba releases it. Generate up to 4K cinematic video with synchronized audio from text, image, or video input — no GPU, no waitlist, no watermark on paid plans. Jump straight to text-to-video and image-to-video.
About Wan 3.0
Wan 3.0 is the anticipated next release in Alibaba Tongyi Lab's open-source Wan video series. Based on the Wan 2.x trajectory and pre-release reports, it is expected to generate up to 30-second 4K clips with in-pass synchronized audio from text, image, or video input, using a Mixture-of-Experts (MoE) architecture. These figures are expected and not yet officially confirmed by Alibaba. wan3pro.video runs the latest open Wan model in your browser today and will serve Wan 3.0 the day it's released — commercial use included, no watermark on paid plans.
| Model name | Wan 3.0 |
|---|---|
| Developer | Alibaba (Tongyi Lab) |
| Status | Expected 2026 — not yet officially released by Alibaba |
| Architecture | Mixture-of-Experts (MoE); parameter count not yet officially confirmed |
| Max resolution | 4K native (no upscaling) |
| Max duration | 30 seconds per clip |
| Max frame rate | 60 fps |
| Input modes | Text, image, video, audio — up to 12 reference assets |
| Audio | Native synchronized generation (in-pass) |
| Lip sync | Phoneme-level, 12 languages |
| License | Apache 2.0 |
| Open weights | Yes — published on Hugging Face (Wan-AI) |
| UI languages on wan3pro.video | 9 — Arabic, German, English, Spanish, French, Japanese, Korean, Norwegian, Chinese |
Sources: Wan-AI on Hugging Face · Alibaba Tongyi Lab
Open source
Most state-of-the-art AI video models — Sora, Veo, Runway — ship as closed APIs. You pay per generation and you cannot inspect, fine-tune, or self-host them. The Wan series is different: Alibaba Tongyi Lab publishes its video model weights under Apache 2.0 on Hugging Face under the Wan-AI organization — Wan 2.1 and 2.2 are already public, and Wan 3.0 is expected to follow the same open license. That means you can audit how the model handles a prompt, fine-tune it on your own dataset, run a private deployment behind a VPN, or build derivative products without negotiating a license. wan3pro.video runs on these open Wan weights — we just operate the GPUs and the multilingual UI so you do not have to. If you ever want to leave the hosted product, the underlying open Wan weights are yours to take. Open source is not a marketing label here; it is the legal status of the model you are using.
Sources: Wan-AI weights on Hugging Face · Apache 2.0 License
Why Wan 3.0
Wan 3.0 is positioned as a leap, not an increment — expected to be the first open-weight video model to ship 4K, native audio, and multi-shot identity tracking in a single forward pass. Here's what that's set to unlock:
Wan 3.0 renders at 4K natively. Other models top out at 1080p and bolt on a separate upscaler that softens edges and introduces flicker. The MoE backbone allocates compute per region, keeping skin texture, fabric weave, and water surfaces sharp at full resolution.
Cloth folds under gravity, liquids splash and settle, smoke dissipates naturally, rigid objects collide with realistic impact. Wan 3.0's physics is part of the base model, not a post-hoc filter — so momentum and material response stay consistent across every frame.
Wan 3.0 generates dialogue, ambient sound, and Foley effects conditioned on the visual scene during the same forward pass as the video. No separate audio session, no manual sync — footsteps, metal stress, and room tone line up with what's on screen.
Direct up to 6 shots in a single clip with the same character across cuts. Wan 3.0's cross-shot identity tracking keeps face, wardrobe, and lighting consistent — solving the character drift that breaks every other model at the 2-shot mark.
Prompt library
Four short prompts we run on wan3pro.video most weeks — each rendered on the latest open Wan model with the parameters listed underneath. Copy any one into the generator above to reproduce, then tune the camera, light, and pacing to taste.
A glass perfume bottle on a marble countertop, soft window light from the left, slow 5-second push-in dolly, shallow depth of field, photorealistic. Render at 9:16, 1080p, 8s — exports clean as a wan 3 video for Reels or TikTok ads.
Two friends in a Tokyo café: shot 1 close-up on Anna laughing, shot 2 over-shoulder on Ben replying, shot 3 wide of the café interior. 6-shot Identity Lock keeps both faces and wardrobe consistent across cuts. 16:9, 4K, 30s.
Hyper-saturated street fashion walk, neon Tokyo backstreet at night, handheld tracking shot, synced footsteps and rain ambience generated in-pass. 9:16, 1080p, 10s.
A clear glass shatters on a wooden table in 240fps slow motion, shards scatter under gravity, dust catches the side light. No post-processing, no compositing — the neural physics engine handles it natively. 16:9, 4K, 6s.
Model comparison
Capability snapshot across the major AI video models. Competitor figures verified against vendor docs (2026-06); Wan 3.0 figures are expected, pending Alibaba's official release.
| Capability | Wan 3.0 | Sora 2 (OpenAI) | Veo 3 (Google) | Kling 3.0 (Kuaishou) |
|---|---|---|---|---|
| Max resolution | 4K native | 1080p | 4K (upscale) | 1080p |
| Max duration | 30s | 20s | 8s | 10s |
| Native audio (in-pass) | ✓ | ✓ | ✓ | Post-only |
| Multi-shot consistency | ✓ 6-shot Identity Lock | Limited | Limited | Motion-brush |
| Open source | ✓ Apache 2.0 | — | — | — |
| Browser-based (no GPU) | ✓ wan3pro.video | ✓ | ✓ (Gemini) | ✓ |
| Commercial use included | ✓ | Plan-dependent | ✓ | ✓ |
| Watermark on paid plans | None | None | None | None |
Sources: vendor documentation as of 2026-06. Capabilities marked ✓ refer to features available on each vendor's hosted service or open-weight release.
Version history
Where Wan 3.0 is expected to land versus the shipping versions in the Wan series. Wan 3.0 figures are pre-release expectations, not official specs.
| Capability | Wan 2.2 | Wan 2.7 | Wan 3.0 |
|---|---|---|---|
| Max resolution | 1080p | 1080p | 4K native |
| Max frame rate | 24 fps | 24 fps | 60 fps |
| Max video length | 16s | 16s | 30s |
| Multi-shot | Basic | Improved | 6-shot Identity Lock |
| Physics simulation | Minimal | Partial | Full neural physics |
| Audio generation | — | — | In-pass synchronized |
| Input modes | Text | Text + image | Text + image + video + audio |
| Open weights | — | — | ✓ Apache 2.0 |
From idea to watermark-free MP4 in under two minutes.
Describe the scene, camera move, and mood. Optionally drag in up to 12 reference assets (image, video, or audio) and tag them in your prompt as <code>@image1</code>, <code>@video1</code>, etc. Strong prompts usually describe subject, camera, light, and motion in that order — the model parses early tokens as priority cues.
Choose resolution (720p / 1080p / 4K), duration (5–30s), aspect ratio (16:9 / 9:16 / 1:1 / 4:3), and quality tier (Fast, Standard, or Pro). Higher resolution and longer clips consume more credits — many creators run a 720p Fast draft pass first, then commit to 4K Pro only on the take they like.
Generation typically completes in 30–60 seconds. Preview in the browser, refine specific regions with the mask editor, and export a watermark-free MP4. Mask edits target only the region you draw, so a single weak shot in a multi-shot sequence does not force you to regenerate the whole clip. Commercial use included on all paid plans.
Use cases
Six audiences already shipping AI video from wan3pro.video to clients, channels, and storefronts.
9:16 vertical shorts at 60fps with native audio. No music licensing headaches, no third-party stock footage, no second-screen audio session.
A/B test 20 creative variants in a morning at a fraction of a typical Meta or TikTok ad shoot. Identity Lock keeps your product hero consistent across every cut.
Turn a single product photo into 360° rotations, lifestyle scenes, or contextual environments for Shopify, Amazon, or your own DTC site. Native 4K means you crop without quality loss.
Pre-viz storyboards, plate generation, and multi-shot scenes without a film crew or a Resolve license. 6-shot Identity Lock makes character continuity actually work.
Diagrams in motion, lip-synced narration in 12 languages, character-consistent walkthroughs — ready for direct LMS or Notion embed.
Open Apache 2.0 weights on Hugging Face for self-hosting, fine-tuning, and academic citation — no vendor lock-in, and the weights are yours to keep.
Direct answers — useful for both humans and the AI assistants that quote this page.
Wan 3.0 is the anticipated next release in Alibaba Tongyi Lab's open-source Wan video series, expected in 2026 under Apache 2.0. It is not yet officially released. Based on the Wan 2.x trajectory, it is expected to generate up to 30-second 4K clips with in-pass synchronized audio from text, image, or video input.
The Wan series weights are free and open-source on Hugging Face under Apache 2.0 — you can self-host if you have the GPU capacity. wan3pro.video is a paid hosted service that runs the latest open Wan model for you (and Wan 3.0 as soon as it's released) so you don't need a local GPU. Paid plans start at $15 one-time (Mini Pack).
The Wan video series ships under Apache 2.0 with public weights on Hugging Face (Wan-AI), and Wan 3.0 is expected to follow the same open license. You can download, modify, fine-tune, and redistribute these models. wan3pro.video is an independent third-party service operating on top of the open models — we're not affiliated with Alibaba.
It is expected to jump from 1080p to 4K native, from 24fps to 60fps, from 16s to 30s clips, add in-pass audio generation, and ship 6-shot Identity Lock for cross-cut character consistency. The release is also expected to add video and audio as input modes alongside text and image.
Wan 3.0 is the only one of these models expected to ship with open Apache 2.0 weights. On expected specs it leads on max resolution (4K native vs 1080p), max duration (30s), and multi-shot identity tracking. See the comparison table above for the full breakdown.
Yes — Apache 2.0 grants commercial use rights to the model itself. wan3pro.video also includes a commercial license for all generated output on every paid plan, with no watermark.
Yes. The model generates dialogue, sound effects, and ambient audio conditioned on the visual scene during the same forward pass as the video — so what you hear matches what you see, with no separate sync step.
Text prompts, reference images, reference video clips, and reference audio. You can combine up to 12 reference assets in one generation and tag them by type and number in your prompt.
Resolutions: 720p, 1080p, and 4K. Aspect ratios: 16:9, 9:16, 1:1, and 4:3. Higher resolutions consume more credits per second.
Typical generation runs 30–60 seconds for a 5–10s clip on Fast/Standard tier. 4K Pro renders take longer (up to a few minutes). Generation time scales with resolution and duration.
No waitlist. No GPU. No watermark on paid plans. Open the generator or compare options on pricing. The Mini Pack at $15 gives you a low-risk way to validate the model against your own brief before deciding on a subscription. Every paid output is yours to deploy commercially across client work, paid social, or product pages with no attribution required.
Yearly plans cut your effective monthly cost by ~50% and lock pricing for 12 months. Monthly is flexible, credit packs fit irregular usage.
Save ~50% per month with a 12-month commitment. Credits release monthly and reset each cycle.
Pay month-to-month. Switch to yearly if your usage is steady — save ~50% per month.
Pay once, no auto-renew, credits valid for 12 months. Best for one-off bursts.
$14.9 / / month
Ideal for your first repeat paid workflow
$39.9 / / month
636+ AI videos a year — the sweet spot for active creators
$74.9 / / month
Perfect for most ongoing creator use cases
$29.9 / / month
Ideal for your first repeat paid workflow
$79.9 / / month
Up to 53 videos/month. More credits, more flexibility, cancel anytime.
$149.9 / / month
Perfect for most ongoing creator use cases
$15 / / pack
The lowest-risk way to start paid generation without committing to a subscription.
$39 / / pack
1,000 credits for one-off projects without a subscription.
Credits consumed per second of video generation, with separate rates for runs with and without video input.
Everything you need to go from idea to finished video.
AI Video Generator
Start with a text prompt or upload an image. Generate cinematic clips in under 60 seconds with 1080p output.
Text-to-Video Prompting
Learn to write effective prompts — subject, motion, camera, lighting, and style — for reliable text-to-video results.
Image-to-Video Workflows
Upload product shots, portraits, or concept art. Animate them with motion while preserving identity and style.
Wan 3.0 vs Kling, Sora & Veo
Side-by-side comparison of resolution, duration, audio support, and open-source licensing across the major AI video models.
Prompt Engineering Guide
A complete guide to structuring prompts for cinematic output — with examples for ads, social content, and product demos.
Wan 3.0 is the open-source video model series from Alibaba Tongyi Lab (Apache 2.0). wan3pro.video runs the latest open Wan model in your browser — text-to-video, image-to-video, multi-shot storytelling, and lip sync across multiple languages, with output up to 1080p today (4K expected with Wan 3.0) in 16:9, 9:16, and 1:1. Commercial usage rights are included with all paid plans.
