Video model by Alibaba

Wan 2.7

Wan 2.7 makes clips of up to 10 seconds with sound, and it holds together prompts with a lot of rules, like a storm where only the world stops. It can also take a video you've already made and change the scene, the style or the action from one sentence.

Make a video with Wan 2.7
Made with Wan 2.7

The prompt

16:9, 10 seconds, 720p, sound on. Multi-shot, 5 shots. Look: ultra-photorealistic live-action fantasy, as if shot on location with a real performer and a practical costume. Natural skin texture, heavy fabric with visible fibers, wet moss, rough bark, real rain. 28mm and 50mm lenses, subtle film grain, restrained color: deep forest greens and grays with one vivid red. Character (invented, adult): a mage in her thirties with long copper-red hair and a pale, freckled face, wearing an immense deep-red hooded cloak of heavy wool that drags on the ground, and carrying a tall wooden staff wrapped in living moss. Keep her face, hair, cloak and staff identical in every shot. The rule of the scene: when she raises the staff, the world freezes but she does not. Rain, leaves and debris stop dead in midair, while her hair and cloak keep moving in a faint wind and she moves freely through the frozen storm. Shot 1 (0-2s): Wide, the camera pushing through the trees as a violent supernatural storm tears through an old forest. Branches whip, leaves and bark spiral, rain sheets sideways. Far ahead, the mage stands still, a red figure against dark green. Sound: roaring wind, cracking branches, heavy rain. Shot 2 (2-4s): Low-angle medium shot. She lifts the staff high above her head with both hands, calm and focused, the wind dragging her hair and cloak sideways. Sound: the storm at its loudest, a deep rising tone. Shot 3 (4-7s): Close on her hand gripping the staff. At the peak of the gesture, time stops. The camera glides slowly through the frozen storm: raindrops hang like glass beads, a torn leaf hovers mid-spin, splinters of a shattered oak float in place. Only her cloak still ripples. Sound: sudden silence, then a faint high ringing. Shot 4 (7-9s): Close-up. She reaches out and touches one suspended raindrop with her fingertip; it trembles but stays in the air. She closes her hand into a fist. Sound: a tiny crystal chime, her slow breath. Shot 5 (9-10s): Wide. Everything rewinds at speed: rain flies back up into the sky, leaves return to their branches, and a huge shattered tree behind her pulls its splinters back together and stands whole again, bark sealing over. The forest settles into calm, misty light. Sound: a reversed whoosh building to a deep boom, then soft dripping and birdsong. Negative prompt: CGI look, game-engine render, plastic textures, the mage freezing with the world, her cloak changing color, extra limbs, text, watermark.

Settings

  • 10 seconds
  • 720p
  • 16:9
  • Sound on

Made with Wan 2.7 on October 23, 2026. Trimmed only, nothing else touched.

Try this prompt

A mage in a heavy red cloak raises her staff in a forest storm and time stops around her, rain and leaves frozen in midair while she keeps moving; she closes her hand and the storm rewinds as a shattered tree rebuilds itself. Made with Wan 2.7.

Edit by instruction

Change the world, keep the walk

Give an edit model a clip and one sentence. Here, a Wan 2.7 beach clip became Mars, edited with Wan 2.7, while the astronaut's steps and the camera move stayed exactly where they were.

The edit instruction

Change the beach and the sea into the rust-red surface of Mars: rocky red sand, low dusty dunes, and a pale butterscotch sky with two small moons near the horizon, lit by a cooler, thinner sunset. Keep the astronaut, the spacesuit, the orange patch, every step, the bootprints, the moment they stop and lift their hand, and the camera move exactly the same.

The original clip's prompt

16:9, 10 seconds, 720p, sound on. Single continuous shot. Look: cinematic science fiction at golden hour, warm amber and soft teal, 35mm, light film grain. An astronaut (invented, adult) in a white and silver spacesuit with an orange mission patch on the left shoulder and a mirrored gold visor walks slowly along the wet sand at the edge of a calm sea at sunset, leaving a line of bootprints. Halfway through, the astronaut stops, turns toward the horizon and lifts one gloved hand to shade the visor. Camera: a slow tracking shot from the side at waist height, matching the astronaut's pace, then holding as they stop. Sound: soft waves, wet sand crunching under each boot, the faint hiss of the suit's air supply. Negative prompt: a visible face through the visor, text, logos, extra limbs.

Settings

  • 10 seconds
  • 16:9
  • Edit, no sound

Original clip made with Wan 2.7 on Predis.ai on October 23, 2026. Edited with Wan 2.7 on Predis.ai on October 27, 2026. Nothing retouched.

Try this prompt
Original clip, 10 seconds
After one sentence, edited with Wan 2.7
0:00 / 0:10

An astronaut walks along a beach at sunset, then the same clip edited with one sentence in Wan 2.7 so the beach becomes the red surface of Mars, the walk and camera move unchanged.

What an edit can change
Scene and settingYes
Style, such as anime or paintingYes
How a character looksYes
What a character does or saysYes
Camera framingYes
SoundNo, edited clips come out silent
Clip lengthUp to 10 seconds

Edits come out silent, so keep the original clip's sound or add a track afterward. If the change is big, rewriting the original prompt can be faster than editing, but that's your call.

First and last frame

You pick the opening and the reveal

Upload the frame you want to start on and the frame you want to end on, and Wan 2.7 fills the 10 seconds between them. For a launch, that means the covered car and the final hero shot are both ones you approved.

Hover a frame to see where it shows up.Tap a frame to see where it shows up.

Shoot both frames from the same angle. The closer they match, the smoother the move between them.

What Wan 2.7 made between them

A silk sheet slides off a car in a dark showroom as spotlights snap on, revealing the AERO concept car. Made from a first and a last frame.

The prompt between the frames

16:9, 10 seconds, 720p, sound on. Single continuous shot, starting on the first frame and ending on the last frame. Look: premium automotive launch film, deep blacks, crisp highlights, slow and confident. In the dark showroom, the silk sheet begins to slide off the car toward the back, pulled by unseen wires, rippling and catching the light as it falls away. One by one, spotlights snap on overhead, each lighting more of the pearl-white body. The ice-blue light line glows on from nose to tail. The camera makes a slow, smooth arc to the right and settles on the final framing, the AERO lettering sharp behind the front wheel. Sound: the soft hiss of sliding silk, a heavy click as each spotlight switches on, a low electronic hum rising as the light line glows, then one deep bass note. Negative prompt: people, misspelled text, real car brands, the car changing shape, the sheet passing through the car.

Settings

  • 10 seconds
  • 720p
  • 16:9
  • Sound on

Made with Wan 2.7 on October 23, 2026. Trimmed only, nothing else touched.

Try this prompt
Made with Wan 2.7

Five prompts, three ways in

These are sorted by what went in: a prompt, a photo or a clip that already existed. Open any card to see exactly what was asked.

Text to video

Don't stretch

A drop of sweat cracks concrete, then one sprint crosses a stadium, a city, a desert and the ocean.

Don't stretch

16:9, 10 seconds, 720p, sound on. Multi-shot, 6 shots. Look: mythic sports commercial. High-contrast, desaturated blue-gray light, deep shadows, heavy film grain, every small sound exaggerated. Character (invented, adult): a sprinter in her twenties with dark hair in a ponytail, in a black running top and black shorts, with plain white and silver running shoes and no logos. Keep her identical in every shot. Shot 1 (0-2.5s): Extreme macro of a single drop of sweat falling against a blurred dusk sky. When it hits the wet ground, the concrete cracks outward and a flock of birds bursts into the air. Sound: a heartbeat, then a huge, deep crack. Shot 2 (2.5-4.5s): Low, wide shot in an empty stadium under floodlights as she sprints straight at the camera, the track cracking behind her and storm clouds rolling in. Sound: her breath, the buzz of the lights. Shot 3 (4.5-6.5s): Side tracking shot as she runs down a rain-soaked city street at night. Sound: one enormous drum hit on every footfall. Shot 4 (6.5-8s): The same run across a desert at sunset, sand kicking up behind her. Shot 5 (8-9s): She runs across the surface of the ocean, each footfall slapping the water and holding her weight. Shot 6 (9-10s): She stands still at a red finish-line tape on an empty desert highway. Sound: the drums stop dead. A low, calm female voiceover, British accent, talks through the run. Negative prompt: real brand logos, swooshes, text, extra limbs, cartoon style.

Settings

  • 10 seconds
  • 720p
  • 16:9
  • Sound on

Made with Wan 2.7 on October 23, 2026. Trimmed only, nothing else touched.

Try this prompt
Text to video

Backs to the river

A Han dynasty general holds the line, with the dialogue in Mandarin.

Backs to the river

16:9, 10 seconds, 720p, sound on. Multi-shot, 3 shots. Dialogue in Mandarin Chinese. Look: epic historical war film, realistic Chinese period drama, cold blue-gray natural light, drifting dust, wide anamorphic framing. Characters (invented portrayals, adults): General Han Xin, in his thirties, in dark scaled armor, calm and resolute; a messenger in light armor, out of breath. Shot 1 (0-5s): Extreme wide, low over a fast, churning brown river. Across the plain, the enemy Zhao army lines the hills under rising dust. The messenger runs up to the general on the riverbank and kneels, shouting in Mandarin: "将军！我军背后是大河，无路可退！" Sound: the roaring river, banners snapping in the wind, distant war drums. Shot 2 (5-8s): Medium shot of Han Xin in front of ranks of soldiers and black banners, eyes on the enemy. He says in Mandarin, low and certain: "传令下去，置之死地而后生！" Sound: the river, the drums growing louder. Shot 3 (8-10s): Over his shoulder toward the Zhao army massed on the hills, dust drifting across the plain. Sound: war drums. Negative prompt: subtitles, captions, text, modern objects, gore, fantasy creatures, cartoon style.

Settings

  • 10 seconds
  • 720p
  • 16:9
  • Sound on

Made with Wan 2.7 on October 23, 2026. Trimmed only, nothing else touched.

Try this prompt
Text to video

Neon blade

One fighter, a ring of enemies and a fast pull-back through a cyberpunk alley.

Neon blade

16:9, 10 seconds, 480p, sound on. Single continuous shot. Look: stylized cyberpunk game cinematic. A rain-slick alley at night, red and blue neon signs with no readable text, haze, hard rim light, glossy reflections. Character (invented, adult): a lean fighter with long wet black hair and glowing pale-blue eyes, in a black armored bodysuit, holding a glowing blue energy blade. Keep the fighter identical throughout. The camera starts in an extreme close-up on the fighter's eyes behind the glowing blade, then pulls back fast to reveal a ring of armored enemies with glowing red visors closing in. The fighter moves fluidly, deflecting and striking, blurring across short distances in streaks of blue light, and the enemies fall one by one. In the final seconds, the fighter stands alone among the fallen, blade lowered and humming. Sound: steady rain, a low synth bass pulse, the electric hum and crackle of the blade, sharp impact hits. Negative prompt: blood, gore, real weapons, readable text, extra limbs, heavy motion smearing.

Settings

  • 10 seconds
  • 480p
  • 16:9
  • Sound on

Made with Wan 2.7 on October 23, 2026. Trimmed only, nothing else touched.

Try this prompt
Edits

Same fight, in ink

The neon blade clip, restyled as an ink-wash painting with one sentence in Wan 2.7.

Same fight, in ink

Turn the whole video into a traditional Chinese ink-wash painting on aged rice paper: black ink brushstrokes, soft gray washes and visible paper texture, with the energy blade as a single stroke of bright red ink. Keep the fighter, every move, every enemy, the camera pull-back and the timing exactly the same.

Edited from

The Neon blade clip, edited in Wan 2.7 with the instruction above.

Settings

  • 10 seconds
  • 16:9
  • No sound

Edited with Wan 2.7 on Predis.ai on October 27, 2026, from the Wan 2.7 clip. It plays without sound.

Try this prompt
Image to video

Jellyfish drones

One festival photo becomes glowing drones drifting over the crowd.

Jellyfish drones

9:16, 10 seconds, 720p, sound on. Start from the uploaded photo as the first frame. Look: dreamlike sci-fi, deep violet and electric-blue palette, soft bloom on every light. High above the crowd, dozens of glowing drones shaped like jellyfish descend slowly out of the dark sky, their translucent domes pulsing pink and cyan, long ribbons of light trailing below them like tentacles. They drift and sway in formation above the raised hands, and the crowd turns its faces upward as the drones' light washes over them. The camera tilts up slowly from the crowd to the drones. Sound: muffled bass from the distant stage, a delighted cheer rising from the crowd, a soft shimmering hum from the drones. Negative prompt: readable text, logos, people merging together, recognizable faces, daylight.

First frame

A night photo taken from inside a festival crowd, with no recognizable faces.

Settings

  • 10 seconds
  • 720p
  • 9:16
  • Sound on

Made with Wan 2.7 on October 23, 2026. Trimmed only, nothing else touched.

Try this prompt
Step by step

How to make a video from a first and last frame

  1. Wan 2.7 selected in the Predis.ai Model Studio, with the first and last frame slots under the prompt
    01

    Choose Wan 2.7

    Open the Model Studio and choose Wan 2.7. The first and last frame slots sit under the prompt.

  2. A first frame and a last frame uploaded side by side
    02

    Upload the two frames

    Add the frame you want to open on, then the one you want to end on. Making the last frame as an edit of the first keeps the angle and the room the same.

  3. A prompt describing what happens between the two frames
    03

    Describe what happens between them

    Say what moves, in what order, and where the camera ends up. A 10-second clip took about 4 minutes in our tests.

We write every prompt between two frames in the same shape: where it starts, what changes, and where the camera lands.

Start on the first frame: [what we see].
Then [what moves or changes], [in what order].
The camera [the move], ending on the last frame: [what we see].
Sound: [effects], [music].

Wan 2.7, by the numbers

  • Made byAlibaba (Tongyi Lab)
  • ReleasedApril 2026
  • Clip lengthUp to 15 seconds, or 10 for an edit
  • Resolution720p or 1080p
  • Ways inA prompt, a photo, a first and last frame, or an existing clip to edit
  • SoundMade with the video; edits come back silent
  • Open weightsNot released for 2.7
  • Formats16:9, 9:16, 1:1, 4:3 and 3:4
Last checked on October 23, 2026.
Fine print

What it won't do yet

Most of these come straight from Alibaba's own documentation. Each one has a way around it, so it's worth a read before your first edit.

Edits stop at 10 seconds

New clips run up to 15 seconds, but Wan 2.7 VideoEdit takes at most 10.

FixTrim the clip to the 10 seconds you want changed before you edit it.

Edits come back silent

An edited clip has no sound at all.

FixKeep the original audio and lay it back over the edit.

No 4K

It tops out at 1080p.

FixFor 4K, use Google Veo 3.1.

Not for your own computer

Wan 2.7 isn't released as open weights, so it can't run locally in tools like ComfyUI.

FixUse it through Predis.ai; for local work, the open Wan 2.2 models are the current option.

Busy prompts take longer

Prompts with many rules take longer, since the model reportedly plans the scene before it starts.

FixDraft with a simple prompt, then add detail for the final.

What people ask about Wan 2.7

What is Wan 2.7?

It's Alibaba's video model, made of four parts: text-to-video, image-to-video, reference-to-video and video editing. It makes clips at 720p or 1080p with sound, up to 15 seconds long on Predis.ai, or 10 for an edit.

Not yet. The open-weight Wan models on GitHub and Hugging Face are earlier versions such as Wan 2.2, while Wan 2.7 runs through Alibaba Cloud and the platforms built on it.

Only once its weights are public, which hasn't happened yet. Until then, ComfyUI workflows labeled Wan run the open Wan 2.2 models.

Alibaba released it in spring 2026, with API access reported from March and the full set of models in April.

On Predis.ai it's a paid model you can try during the 3-day trial, which needs a card and moves to a paid plan unless you cancel.

Yes, with Wan 2.7 VideoEdit in the Model Studio. Upload a clip up to 10 seconds and describe the change in a sentence, and the edited clip comes back without sound.

Up to 10 seconds on Predis.ai, for new clips and edits alike. Through Alibaba Cloud's own API it goes up to 15 seconds from a prompt or an image.

Yes, it generates sound with the video by default, or you can upload your own track. Edited clips are the exception and come back silent.

No. Wan 2.7 Image is a separate Alibaba model for still images, and this page covers the video model.

Yes, through Alibaba Cloud Model Studio and providers that build on it. On Predis.ai you don't need one, since it's built into the Model Studio.

It adds first and last frame control, editing by instruction and voice references, and makes 1080p clips up to 15 seconds.

Make one, then change it

Start from any prompt on this page, then edit the result with one sentence.

Make a video with Wan 2.7
Make a video with Wan 2.7

Wan is a trademark of Alibaba. Predis.ai is not affiliated with or endorsed by Alibaba.