Skip to content

UlazAI - AI Image & Video Tools

New in UlazAI Video Studio

H3 LiteMiniMax H3 video for 6 credits a second

A distilled build of MiniMax H3. It makes 5 to 15 second clips with sound, from a prompt, a photo, or a first and last frame. The full MiniMax H3 costs 36.5 UlazAI credits a second, so a 10-second clip drops from 365 to 60 credits.

Before we added it to the studio we ran 33 clips through it. The best 22 are on this page with their prompts, plus six where it did not do what we asked.

A paper boat, from gutter to open sea. Text to video, 2K, 15 seconds, one take. See the prompt
Up to 2K
6 credits a second
4K
30 credits a second
Length
5 to 15 seconds, with sound
Our test
33 of 33 clips rendered

How much cheaper is H3 Lite than MiniMax H3?

About six times. Both bill per second of video, in UlazAI credits. H3 Lite charges 6 UlazAI credits a second at 768p, 1080p and 2K, and 30 at 4K. MiniMax H3 charges 36.5 UlazAI credits a second at 2K.

What you give up: this is a distilled, post-trained version of H3, and its 2K output is upscaled. Clips start at 5 seconds instead of 4, you get up to four reference images and no audio references. Need those, or the full model? Use MiniMax H3.

UlazAI credits per clip at 2K
5 s
H3 Lite30
MiniMax H3183
10 s
H3 Lite60
MiniMax H3365
15 s
H3 Lite90
MiniMax H3548

H3 Lite at 4K: 150 credits for 5 s, 300 credits for 10 s, 450 credits for 15 s.

Hover or tap a clip to play it. Every clip has its full prompt, the settings, and how long it took from submit to download. Text prompts open straight in the studio.

A paper boat, from gutter to open sea

Text to video · 2K · 16:9 · 15 s · rendered in 5:31

One unbroken take through rain, a canal and a harbour. The sun breaks through right on the 8-second mark we asked for.

Prompt

One continuous tracking shot following a small folded paper boat. 0-4s: low close shot, the paper boat rides a stream of rainwater down a street gutter in an old city, raindrops splash around it. 4-8s: it drops over a small edge into a canal, bobs, and keeps sailing under a stone bridge while the camera glides just above the water. 8-12s: the canal opens into a harbor, the rain stops, sunlight breaks through the clouds. 12-15s: the camera rises as the tiny boat sails past the harbor lights toward the open sea, golden light on the waves. Sound: rain and gutter water, an echo under the bridge, then gulls and open sea. No cuts, no text, no logos.

Try this prompt

A crane shot over a city made of glass

Text to video · 2K · 16:9 · 12 s · rendered in 5:26

Starts on one raindrop, rises over the roof edge and glides between the towers. No cut anywhere in the 12 seconds.

Prompt

One continuous unbroken shot at blue hour. 0-2s: extreme close-up of a single raindrop sliding down a glass rooftop, city lights blurred behind it. 2-6s: the camera cranes up and over the edge of the roof, revealing a miniature city made entirely of glass, every tower lit from inside in warm amber and cyan. 6-10s: the camera glides between the towers at rooftop height, a tiny glass tram crosses a bridge below, its lights reflecting in wet streets. 10-12s: the camera rises slowly and settles on a wide view of the whole glowing glass city under a deep blue sky. Sound: soft rain on glass, a distant tram bell at 7s, low ambient hum. Smooth crane movement, no cuts, no text, no logos.

Try this prompt

From one lit window to the whole valley

Text to video · 2K · 21:9 · 12 s · rendered in 5:42

A slow pull-out that keeps the village, the smoke and the falling snow consistent all the way back.

Prompt

Blue hour in winter. 0-3s: close shot of a single warmly lit window with frost on the glass, through it a family sets a table, candles flicker. 3-12s: one slow continuous pull out: the camera drifts backward and upward, revealing the snowy wooden house, then the whole alpine village with dozens of lit windows, smoke rising from chimneys, a small church, and finally the dark mountains around the valley under the first stars. Snow falls gently throughout. Sound: muffled laughter from inside, then wind and a distant church bell. Continuous movement, no cuts, no text.

Try this prompt

Rack focus from a dew drop to a bee

Text to video · 4K · 16:9 · 6 s · rendered in 5:45

Focus moves from the drop to the bee as asked. At 4K this clip cost 30 credits a second instead of 6.

Prompt

Macro nature shot in a lavender field at sunrise. 0-2s: extreme close-up, razor thin depth of field, sharp focus on a single dew drop hanging from a lavender flower, the field a soft purple blur behind. 2-4s: rack focus: the dew drop goes soft as focus shifts to a honeybee landing on the next flower, fine hairs and pollen grains on its legs sharply visible. 4-6s: the bee crawls into the flower, the lavender sways gently in the breeze. Sound: close buzzing of the bee, soft morning birds. Static macro camera, no cuts, no text.

Try this prompt

Rising over a glacier at sunrise

Text to video · 4K · 21:9 · 8 s · rendered in 7:50

Crevasses up close, then the peaks. At 4K this took almost 8 minutes, one of the slowest renders of the run.

Prompt

Aerial crane shot over a blue glacier at sunrise. 0-8s: the camera starts low over deep blue crevasses, ice textures and meltwater streams sharply detailed, then rises and tilts up to reveal a vast ice field leading to jagged snowy peaks lit pink by the first sun, a tiny line of three climbers roped together crosses the ice in the middle distance. Sound: wind, creaking ice, a distant crack. Smooth rising movement, no cuts, no text.

Try this prompt

Street wok, shot vertical for social

Text to video · 2K · 9:16 · 10 s · rendered in 8:17

Fire, pull-back and the plate pushed to camera, all in one vertical shot with kitchen sound.

Prompt

Vertical handheld street-food shot at night in Bangkok. 0-3s: close shot of a cook tossing noodles high from a flaming wok, fire bursts up and lights his face orange. 3-7s: the camera pulls back to show the busy stall, steam, hanging bulbs, customers on plastic stools laughing. 7-10s: he slides the noodles onto a plate and pushes it toward the camera with a grin. Sound: roaring wok fire, sizzling, clanging spatula, street chatter and scooters. No cuts, no readable text, no logos.

Try this prompt

A hand writes OPEN in chalk

Text to video · 2K · 16:9 · 6 s · rendered in 4:22

The word appears letter by letter and is spelled right in every frame. Text in video often goes wrong; here it held.

Prompt

Static medium shot of a black chalkboard sign on the sidewalk outside a bakery in morning sun. A hand finishes writing the single word OPEN in large white chalk capital letters, then steps out of frame. The word OPEN stays clearly readable and correctly spelled for the whole shot. Sound: chalk scratching, distant street sounds. No cuts, no other text, no logos.

Try this prompt

An old fisherman says one line

Text to video · 2K · 16:9 · 8 s · rendered in 2:47

Speech-to-text gives back the exact line: "The sea gives, the sea takes. Tonight it gives." The lips follow it.

Prompt

An old fisherman with a weathered face and a grey beard sits on an upturned wooden crate at a small harbor at dusk, side light from a single lantern on his left. 0-2s: medium shot, he coils a rope and looks out to sea. 2-8s: slow push-in to a close-up as he turns to the camera and says in a low, calm English voice: "The sea gives. The sea takes. Tonight, it gives." His lips move in sync with the words, a small knowing smile at the end. Behind him boats rock gently, gulls cry. Sound: his voice clearly, water lapping, distant gulls. No cuts, no subtitles, no text.

Try this prompt

Two people, two lines

Text to video · 2K · 16:9 · 10 s · rendered in 6:48

Both lines come back word for word from speech-to-text, plus the laugh. The listener reacts instead of mouthing along.

Prompt

A small Amsterdam brown cafe in the afternoon, warm practical light. Two-shot across a wooden table: on the left an older man with a flat cap, on the right a young woman with round glasses. 0-4s: the man leans in and says in English: "You moved here for the bikes, didn't you?" 4-8s: she laughs and answers: "No. For the bread." 8-10s: they both laugh and clink their coffee cups. Only the person speaking moves their lips, the other listens and reacts. Sound: both voices clearly, cafe murmur, cups clinking at 9s. No cuts, no subtitles, no text.

Try this prompt

Drift, crash zoom, hands on the wheel

Text to video · 2K · 21:9 · 8 s · rendered in 8:23

Three shot sizes in one prompt: aerial, crash zoom into the window, close-up of the hands.

Prompt

Golden hour on a winding mountain road with a steep valley below. 0-3s: wide aerial shot of a red vintage sports car drifting through a hairpin, tire smoke trailing behind it. 3-4s: a sudden crash zoom into the driver's side window. 4-8s: close shot of the driver's hands turning the wheel and shifting gears, then the car straightens and accelerates away toward the sun. Sound: engine roar, tire screech, a whoosh on the crash zoom. No cuts, no readable text, no logos.

Try this prompt

Colour powder in slow motion

Text to video · 2K · 16:9 · 8 s · rendered in 3:25

Backlit pigment clouds look great. The switch to slow motion comes about a second later than the prompt asked.

Prompt

A contemporary dancer in a loose white outfit on a black stage, one strong backlight behind her creating a rim of light. 0-3s: at normal speed she spins twice, holding handfuls of colored powder. 3s: she throws her arms open. 3-8s: extreme slow motion: magenta, cyan and saffron pigment explodes outward in huge glowing clouds lit from behind, individual grains visible, her hair and fabric floating, the camera slowly pushes in. Sound: a deep drum hit at 3s, then a slowed-down airy whoosh and soft reverb. No cuts, no text, no logos.

Try this prompt

Whip pan from the wheels to her face

Text to video · 2K · 9:16 · 8 s · rendered in 3:57

The whip pan is there, with motion blur. During the pan the body morphs for a few frames.

Prompt

Golden hour on an empty coastal road. 0-3s: low tracking shot beside the urethane wheels of a skateboard rolling fast over warm asphalt, long shadows, sun flares through palm trees. 3-4s: a sudden whip pan upward with strong motion blur. 4-8s: the camera lands on the skater's face, a young woman with curly hair and a yellow windbreaker, laughing as she carves a wide turn, hair blowing, the sea glittering behind her. Sound: rolling wheels on asphalt, a whoosh during the whip pan, distant waves and gulls. No cuts, no text, no logos.

Try this prompt

Broth poured in slow motion

Text to video · 2K · 1:1 · 6 s · rendered in 2:01

Hard top light, steam through the beam. Our fastest render: two minutes.

Prompt

Low key food shot on a dark wooden counter, one hard spotlight from above. A black ceramic bowl holds noodles, a soft-boiled egg cut in half, scallions and chashu. 0-6s: in slow motion a ladle pours glossy golden broth into the bowl, the broth swirls around the noodles, fat droplets catch the light, thick steam curls up through the spotlight beam. Static camera at a slight top angle. Sound: pouring liquid, a gentle sizzle, quiet kitchen ambience. No cuts, no text.

Try this prompt

A puppy runs at the lens through powder snow

Text to video · 2K · 16:9 · 8 s · rendered in 3:58

Backlit snow spray and floppy ears. The leap at the end did not happen; it keeps running.

Prompt

Low tracking shot at snow level in a sunny winter forest. A golden retriever puppy bounds through deep fresh powder toward the camera in slow motion, snow spraying from its paws, ears flopping, tongue out. At 5s it leaps and lands in a puff of sparkling snow right in front of the lens. Backlit by low sun, snow crystals glitter. Sound: soft crunch of snow, happy panting, a small bark at the end. No cuts, no text.

Try this prompt

A six-word prompt, with the planner on

Text to video · 2K · 16:9 · 6 s · rendered in 2:09

Prompt: "a cat DJ at a rooftop party". With the prompt planner on, you get the cat. Without it, see below.

Prompt

a cat DJ at a rooftop party

Try this prompt

Product shot: the laces tie themselves

Image to video · 2K · 16:9 · 8 s · rendered in 2:30

Starts on our product photo. The shoe keeps its knit, colours and shape while the laces pull into a bow.

Input image 1 for Product shot: the laces tie themselvesstart frame
Prompt

Start exactly on the provided frame. 0-3s: slow push in toward the sneaker as a thin line of lime light sweeps across the knit from left to right, revealing the texture. 3-6s: the concrete pedestal slowly rotates the sneaker a quarter turn, the laces tighten by themselves into a neat bow. 6-8s: a soft burst of fine dust lifts from the pedestal in the rim light, the sneaker stays perfectly unchanged in shape and color. Sound: a low synth swell, soft fabric creak as the laces tighten, a deep bass hit at 6s. No cuts, no text, no logos.

Open image to video

Orbit around a vintage radio

Image to video · 2K · 1:1 · 8 s · rendered in 6:11

Square product clip. The dial, strap and grille stay true to the source photo during the orbit.

Input image 1 for Orbit around a vintage radiostart frame
Prompt

Start exactly on the provided product frame. 0-3s: slow orbit around the vintage teal radio as soft light glides over the enamel and chrome dial. 3-5s: the dial needle sweeps across the scale and the speaker grille glows faintly warm. 5-8s: the camera settles front-on, the radio stays perfectly unchanged in shape, color and proportions. Sound: radio static tuning into a warm 1960s jazz tune. No cuts, no text, no logos.

Open image to video

A baker speaks Dutch

Image to video · 2K · 9:16 · 8 s · rendered in 4:05

Speech-to-text returns "Elke ochtend om 5 uur. Nog nooit overgeslagen." Same face as the photo throughout.

Input image 1 for A baker speaks Dutchstart frame
Prompt

Start exactly on the provided frame. 0-3s: she presses and folds a ball of dough on the counter with both hands, flour puffs up in the lamp light. 3-8s: she looks into the camera and says warmly in Dutch: "Elke ochtend om vijf uur. Nog nooit overgeslagen." Her lips match the words, a proud small smile. Behind her the window slowly gets lighter. Sound: her voice, dough on wood, an old radio playing softly. Keep her face and clothing identical to the frame. No cuts, no subtitles, no text.

Open image to video

An oil painting that starts to move

Image to video · 2K · 16:9 · 8 s · rendered in 3:09

The windmill turns and the camera flies in, and it stays a painting. No slide into photo realism.

Input image 1 for An oil painting that starts to movestart frame
Prompt

Start exactly on the provided oil painting. 0-8s: the painting comes alive while keeping its oil paint brush strokes and canvas texture: the windmill sails start to turn, the huge clouds drift slowly to the left, light and shadow move over the polder, the cows graze and one looks up, ripples run along the canal. The camera slowly cranes up and forward into the landscape. Sound: wind, the creak of the windmill, a distant cow, birds. Keep the painted style throughout, no photorealism, no cuts, no text.

Open image to video

Seedling to jungle, first and last frame

First + last frame · 2K · 16:9 · 8 s · rendered in 4:12

Starts on frame one, ends on frame two. The time-lapse in between grows a tree out of the pot.

Input image 1 for Seedling to jungle, first and last frameInput image 2 for Seedling to jungle, first and last framefirst and last frame
Prompt

Static locked-off camera, the exact framing of both frames. A time-lapse over many days: the tiny sprout in the terracotta pot grows into a young tree, plants spread across the oak floor, vines climb the walls and hang from the ceiling, leaves unfurl, the light through the window cycles from morning to golden afternoon, finally butterflies arrive. End exactly on the last frame. Sound: a soft rising ambient pad, rustling leaves, birds at the end. No cuts, no text.

A city square from noon to night

First + last frame · 2K · 16:9 · 8 s · rendered in 6:20

Two photos of the same square, one at noon and one at night after rain. The model fills in shadows, a shower and the lights switching on.

Input image 1 for A city square from noon to nightInput image 2 for A city square from noon to nightfirst and last frame
Prompt

Static locked-off camera, the exact framing of both frames. A time-lapse from noon to night: shadows sweep across the square, people stream past as motion blur, the sky turns gold then deep blue, a short rain shower makes the stones wet, cafe lights and string lights switch on one by one, the fountain lights up. End exactly on the last frame. Sound: city ambience, a passing tram bell, rain, then a quiet evening hum. No cuts, no text.

Two references: a robot and a radio

Reference image · 2K · 16:9 · 10 s · rendered in 2:57

Both objects keep their design from the reference photos. The framing cuts the robot off at the left edge.

Input image 1 for Two references: a robot and a radioInput image 2 for Two references: a robot and a radiotwo reference images
Prompt

Use the first reference image for the robot's exact design and the second reference image for the radio's exact design. Golden hour on a city rooftop with string lights. 0-3s: the small orange robot carries the teal radio and sets it down on the edge of the roof, then taps the dial. 3-10s: music starts and the robot dances happily next to the radio, bouncing, spinning once, waving its arms, its round white eyes squinting with joy; the camera tracks slowly around it, the city skyline glowing behind. Keep both objects identical to the references. Sound: the click of the dial, then upbeat retro funk from the radio, the robot's small servo sounds. No cuts, no text, no logos.

Open reference image

Six clips where H3 Lite got it wrong

A miss is billed like any other clip of the same length and resolution. Knowing where it breaks saves you the re-runs.

The same six words without the planner

Text to video · 2K · 16:9 · 6 s · rendered in 2:47

Prompt planner off, prompt unchanged: you get a human DJ and no cat. A short prompt needs the planner. That is why UlazAI turns it on for prompts under 300 characters and off for a written-out shot list.

Prompt

a cat DJ at a rooftop party

A reference video beats a reference image

Reference image + video · 2K · 16:9 · 8 s · rendered in 3:18

We sent the robot photo plus the dancer clip as motion reference. The result is the dancer again, no robot. Mixing a video and an image reference did not work in our test.

Prompt

Use the reference image for the robot's exact design and the reference video only for the choreography and camera rhythm. On a black stage with one strong backlight, the small orange robot performs the same spin as the dancer in the reference video, then throws its arms open while bright orange and cyan confetti bursts around it in slow motion. Keep the robot identical to the reference image. Sound: a drum hit on the arm throw, then a slowed whoosh. No cuts, no text, no logos.

The robot came back with two copies

Reference image · 2K · 16:9 · 8 s · rendered in 5:02

Same idea with only the robot photo: the design is right, but the model added two extra robots and a band we never asked for.

Prompt

Use the reference image for the robot's exact design. On a black stage with one strong backlight, the small orange robot spins around twice on one foot, then throws both arms wide open at 4s while bright orange and cyan powder explodes around it in slow motion, its round white eyes squinting with joy. Keep the robot identical to the reference image. Sound: a drum hit at 4s, then a slowed airy whoosh. No cuts, no text, no logos.

Long physics chains fall apart

Text to video · 2K · 16:9 · 12 s · rendered in 4:08

Dominoes, a rolling lemon and a bell in one chain. Pieces pass through each other and the order breaks. Keep cause-and-effect chains short.

Prompt

A bright Scandinavian kitchen in morning light. A long chain of colorful wooden dominoes runs across the counter. 0-12s: a smooth tracking shot follows the falling dominoes from left to right: they topple in sequence, run up a stack of cookbooks, knock a spoon that rolls a lemon down a small ramp, the lemon tips the last domino, which hits a brass service bell at 11s. Every domino falls in the correct order with believable physics. Sound: rhythmic clicking of dominoes, a lemon rolling on wood, a clear bell ding at 11s. No cuts, no text, no logos.

Quality low finished six seconds sooner and added a word

Text to video, quality low · 768p · 16:9 · 8 s · rendered in 2:41

Same prompt as the fisherman above with quality low at 768p. In this run it finished in 161 seconds against 167, which is within normal queue variation. Speech-to-text heard a made-up name before the line. UlazAI always sends quality high.

Prompt

An old fisherman with a weathered face and a grey beard sits on an upturned wooden crate at a small harbor at dusk, side light from a single lantern on his left. 0-2s: medium shot, he coils a rope and looks out to sea. 2-8s: slow push-in to a close-up as he turns to the camera and says in a low, calm English voice: "The sea gives. The sea takes. Tonight, it gives." His lips move in sync with the words, a small knowing smile at the end. Behind him boats rock gently, gulls cry. Sound: his voice clearly, water lapping, distant gulls. No cuts, no subtitles, no text.

Neon signs with fake writing

Text to video · 2K · 16:9 · 8 s · rendered in 3:00

The dolly zoom works, but the alley fills with signs in made-up characters and the face shifts. Ask for no readable signs, or plan text in your edit.

Prompt

Night, a narrow alley in Tokyo lined with small ramen bars and paper lanterns. A young cook in a white headband stands in the middle of the alley holding a steaming bowl. 0-2s: medium shot, neon signs flicker, steam rises. 2-7s: dolly zoom vertigo effect: the camera moves toward him while zooming out, his size stays the same while the alley behind him stretches and the lanterns rush away into depth; his eyes widen and he slowly lowers the bowl. 7-8s: all lanterns go dark at once, only the steam stays lit by a single sign. Sound: sizzling oil, neon buzz, a low rising tone during the zoom, then silence. No cuts, no readable text, no logos.

Prompts that worked in our test

The clips that came out best shared five habits. None of them are tricks, just a clear brief.

  1. Write a shot list with times

    Most of our best clips had seconds in the prompt: "0-3s: close shot of the wheels. 3-4s: whip pan up." The model kept to those marks in most clips.

  2. Name the camera move

    Crane, pull-out, rack focus, whip pan, crash zoom and dolly zoom all showed up on screen when we named them.

  3. Put spoken lines in quotes

    Give the exact sentence and the language. English and Dutch both came back word for word.

  4. Short idea? Keep it short

    Under 300 characters, UlazAI lets the model expand the prompt first. That rescued the cat DJ.

  5. One reference type at a time

    Images or a video, not both. And say in the prompt which reference is for what.

The test run in numbers

14 October 2026, one account, 8 jobs sent at a time. Render time is from submit to finished download, so it includes queueing.

Video clips33 (24 text, 4 image, 2 first + last frame, 3 reference)
Failed jobs0
Source images9, made with an image model, about 40 seconds each
Render time, middle value4 min 5 s
Fastest / slowest2 min 1 s (6 s ramen) / 8 min 23 s (8 s drift)
Spoken lines checked4 clips with speech-to-text. The three at quality high came back word for word, in English and Dutch
Total cost$3.57 at our provider's list price: $3.39 for the 33 clips ($0.01 a second, $0.05 at 4K) and $0.18 for the 9 images

Which mode do you need?

Text to video

Only a prompt. Pick 16:9, 9:16, 1:1, 4:3, 3:4 or 21:9 and 5 to 15 seconds.

Image to video

Upload one photo as the first frame. Add a second photo to steer where the clip ends. In both of our tests it landed on that frame.

Reference to video

Up to four reference images, or one reference video. Say in the prompt which image is for what.

Questions about H3 Lite

What is H3 Lite?

A distilled build of MiniMax H3. It makes 5-15 second clips with sound, from text, from a first frame, from a first and last frame, or from up to four reference images.

How is it different from MiniMax H3?

It costs 6 UlazAI credits a second up to 2K, against 36.5 for MiniMax H3. It is a distilled, post-trained version, and 2K output is upscaled. The shortest clip is 5 seconds, and there is no audio reference input.

Does it make sound?

Yes, every clip has sound made in the same pass: ambience, music, effects and speech. We checked the spoken lines in our test with speech-to-text.

How long does a clip take?

In our test the middle value was about 4 minutes from submit to download, with 8 jobs running at the same time. The fastest took 2 minutes, a 4K clip almost 8.

What does 4K cost?

30 UlazAI credits a second, so a 6-second 4K clip is 180 credits. Up to 2K it stays at 6 credits a second.

Try it on your own idea

A 5-second 2K clip costs 30 credits. Start with five seconds; go longer once the shot works.

Music Studio Songs, instrumentals and your own lyrics. AI song generator