What Alibaba Actually Changed

Alibaba describes Wan 3.0 as a unified creative system rather than a text-to-video model with a few extras. Its official release materials list native 30-second generation with intelligent duration control, integrated sound, precision editing, and up to 20 reference assets across supported workflows.^[1] Fal exposes the parts creators use most through separate text-to-video, image-to-video, and reference-to-video endpoints.^[2][3][4]

On Deep Dream Generator, the launch version supports:

• Text to video from 2 to 30 seconds.
• Image to video with an optional last frame.
• Reference images, motion clips, and audio tracks in one directed generation.
• 480p, 720p, and 1080p quality tiers.
• 1:1, 4:3, 3:4, 16:9, 9:16, and Adaptive composition.
• Generated audio that can be switched on or off without changing Fal's per-second rate.

02 · The screening room

The Five Best Clips From Our 1080p Test

These are not vendor demos. Every video below came from our own Fal.ai requests, was downloaded and technically probed, and is shown without post-production edits. Together they cover vertical hooks, mechanical motion, fabric and anatomy, creature continuity, and a full 30-second environment journey.

01

1080×192030 seconds

The Last Garden Above the World

The flagship. Water, bridges, monumental trees, floating islands, and a dawn-to-violet color journey recur convincingly across the full half-minute.

02

1920×108015 seconds

The Glass Train Across a Mirror Desert

Our strongest luxury landscape. Crystal, gold, water, sunrise, and mirrored arches stay inside one coherent material language.

03

1920×108020 seconds

Aurora Dragon of the Frozen Canyon

The best long-form creature result: a convincing environment-to-emergence-to-flight arc with a remarkably unified cyan and violet grade.

04

1080×192010 seconds

Celestial Whales Above the Clouds

The best instant hook in the batch. The close eye reveal opens into a readable aerial ballet while the lead whale's cosmic surface remains stable.

05

1080×192015 seconds

Dance of Living Light

A fashion-film stress test for fabric, anatomy, and transformation. The overhead floral mandala gives the shot an unusually strong final image.

Open the exact 30-second flagship prompt

A spectacular vertical 30-second continuous cinematic voyage through an impossible floating botanical sanctuary above the clouds. Begin at dawn inside a dew-covered macro flower where tiny droplets contain reflections of the sky. Pull backward to reveal the flower growing on a vast levitating garden island with waterfalls pouring into the clouds. Glide along a turquoise stream past giant white lotus blossoms, iridescent butterflies, and delicate golden bridges toward an immense cathedral grown from living trees. Enter through its open canopy as warm sunbeams illuminate thousands of floating seeds; the seeds rise and transform into a luminous white-and-gold phoenix. Follow the same phoenix upward through the branches and out above the island as sunset turns the clouds coral and violet. In the final moments, orbit wide to reveal a chain of floating garden islands awakening with warm lights while the phoenix crosses a crescent moon. Preserve spatial continuity and recurring landmarks; smooth unbroken camera flight, graceful natural motion, photorealistic plants and water, subtle magical realism, exquisite color progression from dawn gold to violet twilight, emotionally overwhelming but elegant. Synchronized waterfalls, birds, wind through leaves, distant bells, and a cinematic orchestral crescendo. No people, no text, no logos, no cuts.

What Our Controlled Tests Found

Where Wan 3.0 Is Strongest

World continuity is the standout. Wan repeatedly preserved environment motifs, direction of travel, lighting logic, and a central subject over 15–30 seconds. It also handled reflective and translucent materials unusually well: wet black rock, stained glass, crystal trains, ice, aurora, and bioluminescent fabric all produced convincing light interactions.

Vertical composition is a real strength. The whale, dancer, ramen-stall, and floating-garden tests used the height of the frame to reveal space. They did not feel like landscape shots with the sides removed.

Prompt expansion earns its latency. In our fixed-seed A/B, both hummingbird clips were polished, but expansion-on delivered a clearer approach, contact, and “drinking light” action. Leave it enabled for narrative beats, transformations, and camera choreography; disable it for quick single-subject drafts.

Where It Still Breaks

Wan's weakness is not beauty; it is literal event accounting. When we packed a prompt with many mandatory beats, it often preserved the overall journey while simplifying one action. The Moon bazaar clip kept the rover, dome, direction, and Earthrise, but largely skipped the requested glowing-fruit toss. The floating-garden flagship built a coherent sanctuary, yet simplified the seed-to-phoenix transformation into a looser bird motif.

Fine text is also not a reason to choose this model. Decorative signage in our neon scene became plausible glyph-like texture rather than dependable lettering. And while anatomy stayed coherent at the hero-subject level, hands, limbs, wings, and secondary figures can merge during complex motion.

Which Quality Tier Should You Use?

• Turbo · 480p: storyboards, controlled A/B tests, and cheap prompt iteration.
• Pro · 720p: the default for most finished social and web video. This was the best cost-quality balance in our tests.
• Ultra · 1080p: final showcase renders, detailed materials, and clips you intend to feature prominently.

Fal's published rates are $0.05, $0.10, and $0.20 per generated second for 480p, 720p, and 1080p respectively.^[2] DDG converts those provider costs into its normal energy rates with the same target margin used for our other recent video models. For launch, all three Wan tiers are 50% off that normal energy rate.