Steering Video Diffusion Transformers
with Massive Activations

Xianhang Cheng1, Yujian Zheng1,†, Zhenyu Xie1, Tingting Liao1, Hao Li1,2
1MBZUAI    2Pinscreen
Corresponding author

Abstract

In this work, we study the role of Massive Activations (MAs), which are rare, high-magnitude spikes confined to a few fixed hidden dimensions in video diffusion transformers (DiTs). We uncover a structured positional hierarchy: MA magnitudes peak at first-frame tokens and recur at the spatial boundary tokens of latent frames, with this pattern being most pronounced during early denoising. We trace this organization to an encoding asymmetry of the video VAEs, whose causal temporal padding and zero spatial padding cause the first latent frame and frame borders to carry reduced content load. Elevated MAs consistently align with these lower-content structural positions. To understand their function, we analyze intermediate representations and find that MAs act as implicit rescalers of residual computation: enlarging MAs suppresses the corresponding self-attention and feed-forward updates, while erasing them amplifies these updates. Together, these observations suggest that MAs serves as a token-level rescaler of residual computation, which video DiTs deploy unevenly, placing the strongest damping at the encoding-asymmetric structural positions. Motivated by this native rescaling behavior, we propose Structured Activation Steering (STAS), a training-free technique that steers MAs at the observed structural positions toward a scaled, model-derived reference during early denoising. STAS requires no additional forward passes and consistently improves video quality and temporal coherence across text-to-video models with negligible overhead.


Results

Side-by-side comparisons for every qualitative figure in the paper. In each clip the left half is the unmodified backbone and the right half is the same backbone, same prompt and same seed with STAS applied.

Baseline + STAS (Ours) Same prompt · same seed · no extra forward passes Drag or use ‹ › to roll · click a clip to play / pause · click a caption for the full prompt
Overview Wan2.1-1.3B Wan2.2-5B CogVideoX-5B LTX-2 (base)

Figure 7 — Qualitative comparison across backbones

Main paper 4 comparisons

Figure 7

Wan2.1-1.3B+ STAS (Ours)

7aIn a cozy, sunlit kitchen, a vintage chrome toaster sits on a wooden countertop, gleaming under the morning light. Beside it, a plush teddy bear with a red bow tie leans against the toaster, creating an endearing scene. The toaster pops up two perfectly golden slices of bread, and the teddy bear appears to be watching intently, as if anticipating breakfast. The camera zooms in on the teddy bear's soft, stitched features and then pans to the toaster's shiny surface, reflecting the warm, inviting ambiance of the kitchen. The video ends with a close-up of the teddy bear holding a tiny piece of toast, adding a whimsical touch to the charming morning moment.

CogVideoX-5B+ STAS (Ours)

7bA playful panda, with its distinctive black and white fur, sits on a wooden swing set in a lush bamboo forest. The panda's eyes sparkle with joy as it grips the ropes tightly, swaying back and forth. The surrounding greenery and tall bamboo stalks create a serene, natural backdrop. As the swing moves, the panda's playful antics, including a gentle push off the ground with its hind legs, bring a sense of whimsy and delight. The sunlight filters through the leaves, casting dappled shadows on the ground, enhancing the enchanting atmosphere of this playful scene.

Wan2.2-5B+ STAS (Ours)

7cA sleek, emerald-green sports car glistens under the midday sun, parked on a winding coastal road with the ocean's waves crashing in the background. The car's aerodynamic design and polished chrome accents reflect the surrounding scenery. As the camera zooms in, the intricate details of the car's bodywork and the luxurious leather interior become evident. The engine roars to life, and the car speeds down the road, the sunlight catching its vibrant green paint, creating a mesmerizing effect. The video concludes with the car gracefully navigating a sharp turn, showcasing its agility and power against the stunning coastal landscape.

LTX-2 (base)+ STAS (Ours)

7dA cool cat, sporting sleek black sunglasses and a red lifeguard vest, sits confidently on a high lifeguard chair overlooking a sparkling blue pool. The feline's fur is a mix of orange and white, and its tail flicks with authority. In one scene, the cat scans the pool area with a serious expression, its sunglasses reflecting the shimmering water. Next, it holds a tiny whistle in its mouth, ready to spring into action. The final shot shows the cat perched on the edge of the pool, its paw dipping into the water, maintaining a vigilant watch over the swimmers, embodying the perfect blend of charm and responsibility.

Wan2.1-1.3B

MA dim 1188 · Figures 15–16 10 comparisons

Figure 15

Wan2.1-1.3B+ STAS (Ours)

15.1A majestic giraffe strolls gracefully through a sunlit savannah, its long neck swaying gently with each step. The golden grass sways in the breeze, and the distant acacia trees cast elongated shadows. The giraffe's patterned coat glistens under the warm sunlight, highlighting its elegant movements. Birds flutter around, occasionally perching on its back, adding to the serene atmosphere. As it walks, the giraffe pauses to nibble on the tender leaves of a tall tree, its eyes half-closed in contentment. The sky above is a brilliant blue, dotted with fluffy white clouds, completing the tranquil scene.

Wan2.1-1.3B+ STAS (Ours)

15.2A sleek, modern bicycle with a matte black frame and bright red accents stands parked on a quiet, cobblestone street, its design reflecting both elegance and functionality. Nearby, a vintage car with a polished navy blue exterior and chrome details is parked, its classic curves and gleaming surface evoking a sense of nostalgia. The scene transitions to a close-up of the bicycle's intricate gears and the car's shiny hubcaps, highlighting the craftsmanship of both vehicles. As the camera pans out, the bicycle and car are framed against a backdrop of historic buildings and leafy trees, creating a harmonious blend of past and present.

Wan2.1-1.3B+ STAS (Ours)

15.3A sleek, modern airplane, painted in a striking blue and white livery, sits on a sunlit runway, engines roaring to life. The camera captures a close-up of the powerful jet engines as they begin to spool up, emitting a deep, resonant hum. The scene shifts to a side view, showing the aircraft's wheels starting to roll, kicking up small puffs of dust from the tarmac. As the plane gains speed, the background blurs, emphasizing its rapid acceleration. The nose of the aircraft begins to lift slightly, hinting at the imminent takeoff, with the sun glinting off its polished fuselage, creating a sense of anticipation and excitement.

Wan2.1-1.3B+ STAS (Ours)

15.4A playful tabby cat with bright green eyes dashes across a sunlit meadow, its fur gleaming in the golden light. The cat's tail is held high, and its paws barely touch the ground as it sprints with joyous abandon. The scene shifts to a close-up of the cat's face, capturing its wide-eyed excitement and twitching whiskers. Next, the cat leaps over a small stream, its body arched gracefully in mid-air. Finally, it lands softly on the other side, pausing momentarily to look back with a satisfied expression, the vibrant meadow and clear blue sky providing a picturesque backdrop.

Wan2.1-1.3B+ STAS (Ours)

15.5A majestic elephant stands in a sunlit savannah, its massive form casting a long shadow on the golden grass. The elephant, with its rough, gray skin glistening under the intense sun, lifts its trunk high into the air. With a graceful motion, it sprays a refreshing arc of water over its back, droplets catching the sunlight and creating a shimmering mist. The scene captures the elephant's contentment as it cools down, the water cascading over its wrinkled skin and pooling at its feet. In the background, acacia trees and distant mountains frame the serene moment, emphasizing the beauty and tranquility of the African landscape.

Figure 16

Wan2.1-1.3B+ STAS (Ours)

16.1A vibrant red truck, gleaming under the midday sun, rumbles down a quiet, tree-lined suburban street. Its polished chrome accents reflect the surrounding greenery, creating a picturesque scene. Nearby, a vintage blue bicycle with a wicker basket attached to the handlebars leans against a white picket fence, its tires slightly dusty from recent use. The truck slows as it approaches the bicycle, the driver, a middle-aged man in a plaid shirt and baseball cap, glances at the bike with a nostalgic smile. The scene captures a moment of serene coexistence between modern machinery and timeless simplicity, set against the backdrop of a peaceful neighborhood.

Wan2.1-1.3B+ STAS (Ours)

16.2A person with short, curly hair and wearing a cozy, oversized sweater stands in a warmly lit room, their eyes closed in a moment of deep connection. They embrace another individual, whose face is partially visible, showing a gentle smile. The background features soft, ambient lighting and hints of a comfortable living space with a plush sofa and a bookshelf filled with books and plants. The hug is tender and heartfelt, capturing a sense of warmth and intimacy. The scene transitions to a close-up of their hands clasped tightly, emphasizing the bond and emotional depth of the embrace.

Wan2.1-1.3B+ STAS (Ours)

16.3A vibrant purple umbrella opens against a backdrop of a bustling city street, its rich hue standing out amidst the gray, rainy day. The camera zooms in to reveal intricate floral patterns on the umbrella's fabric, glistening with raindrops. As the umbrella twirls, the city lights reflect off its surface, creating a mesmerizing dance of colors. The scene shifts to a close-up of the umbrella's handle, a polished wooden grip, held by a hand adorned with a silver ring. The video concludes with the umbrella being closed, the rain subsiding, and a rainbow appearing in the sky, symbolizing hope and beauty.

Wan2.1-1.3B+ STAS (Ours)

16.4A majestic shark glides through the swirling, vibrant waters of the ocean, depicted in the iconic Van Gogh style. The scene is alive with dynamic, swirling brushstrokes of deep blues, teals, and hints of golden yellows, capturing the movement of the water and the shark's sleek form. The shark's body is rendered with textured, expressive lines, its fins cutting through the water with grace. The ocean around it is a mesmerizing blend of colors and patterns, reminiscent of Van Gogh's "Starry Night," with the waves and currents creating a dreamlike, almost celestial atmosphere. The entire scene feels both surreal and vividly alive, a perfect fusion of marine life and artistic brilliance.

Wan2.1-1.3B+ STAS (Ours)

16.5A refined couple, dressed in elegant evening attire, navigates a bustling street under a heavy downpour. The man, in a tailored black tuxedo, and the woman, in a flowing crimson gown, both hold delicate paper umbrellas adorned with intricate patterns. The scene, reminiscent of Hokusai's Ukiyo-e style, captures the rain's intensity with sweeping lines and dynamic movement. The couple's expressions reflect a mix of surprise and amusement as they hurry along the rain-soaked path, their garments and umbrellas beautifully detailed against the backdrop of traditional Japanese architecture and blurred lantern lights. The rain, depicted with fine, slanting strokes, adds a sense of urgency and romance to their journey home.

Wan2.2-5B

MA dim 1938 · Figures 17–18 10 comparisons

Figure 17

Wan2.2-5B+ STAS (Ours)

17.1A lone bicycle, with its sleek frame and black tires, glides effortlessly through a vast, snow-covered field under a pale winter sky. The rider, bundled in a red parka, black gloves, and a woolen hat, pedals steadily, leaving a delicate trail in the pristine snow. The scene captures the quiet serenity of the landscape, with snowflakes gently falling and the distant silhouette of bare trees lining the horizon. The bicycle's tires crunch softly against the snow, creating a rhythmic sound that complements the peaceful ambiance. As the rider continues, the sun begins to set, casting a warm, golden glow over the snowy expanse, highlighting the beauty of the winter journey.

Wan2.2-5B+ STAS (Ours)

17.2A couple, elegantly dressed in formal evening wear, navigates a bustling city street under a heavy downpour. The man, in a tailored black tuxedo, and the woman, in a flowing emerald gown, hold large, ornate umbrellas that barely shield them from the relentless rain. The scene is painted in the swirling, vibrant brushstrokes of Van Gogh, with the rain depicted as cascading lines of blues and whites. The streetlights cast a golden glow, reflecting off the wet cobblestones, creating a mesmerizing dance of light and shadow. The couple's expressions are a mix of surprise and delight, their attire glistening with raindrops, as they hurry home through the enchanting, rain-soaked cityscape.

Wan2.2-5B+ STAS (Ours)

17.3A majestic zebra stands in the golden savannah, its black and white stripes contrasting vividly against the tall, sunlit grasses. The camera captures a close-up of its face, highlighting the intricate patterns around its eyes and muzzle. As the zebra turns, the scene shifts to a wide shot, revealing a herd grazing peacefully in the distance, with acacia trees dotting the horizon. The zebra then trots gracefully, its mane flowing with each stride, under a sky painted with hues of orange and pink from the setting sun. Finally, the zebra pauses at a watering hole, its reflection shimmering in the clear water, encapsulating the serene beauty of the African landscape.

Wan2.2-5B+ STAS (Ours)

17.4In a rustic barn bathed in the soft morning light, a person in a plaid shirt, denim overalls, and sturdy boots kneels beside a gentle, brown-and-white cow. The person carefully places a metal pail beneath the cow's udder, their hands moving with practiced ease. The cow stands calmly, its large eyes reflecting trust and contentment. The rhythmic sound of milk hitting the pail fills the air, blending with the soft rustling of hay and distant chirping of birds. The scene captures a timeless moment of harmony between human and animal, set against the backdrop of a peaceful, pastoral landscape.

Wan2.2-5B+ STAS (Ours)

17.5A lone rider, clad in a sleek black leather jacket, matching helmet, and dark jeans, navigates a winding mountain road on a powerful motorcycle. The sun sets behind the peaks, casting a golden glow on the rugged landscape. The rider leans into a sharp turn, the bike's engine roaring, echoing through the serene valley. As they accelerate on a straight stretch, the wind whips past, rustling the trees lining the road. The scene shifts to a close-up of the rider's gloved hands gripping the handlebars, the speedometer needle climbing. Finally, the rider pauses at a scenic overlook, the vast expanse of mountains and sky stretching out before them, capturing a moment of freedom and adventure.

Figure 18

Wan2.2-5B+ STAS (Ours)

18.1A sleek, vibrant orange sports car glides effortlessly along a winding coastal road, its glossy finish reflecting the golden hues of the setting sun. The car's aerodynamic design and polished chrome accents catch the light, creating a dazzling display of color and motion. As it speeds past, the roar of its powerful engine echoes against the cliffs, blending with the rhythmic crashing of ocean waves. The camera captures close-up shots of the car's intricate details: the sharp lines of its body, the gleaming alloy wheels, and the luxurious leather interior. The scene transitions to a panoramic view, showcasing the car's journey along the scenic route, with the endless horizon and sparkling sea as a breathtaking backdrop.

Wan2.2-5B+ STAS (Ours)

18.2A serene individual sits comfortably in a cozy, softly lit room, wearing a plush white robe. They gently massage their legs, starting from the calves and moving upwards with slow, deliberate motions. The camera captures the close-up details of their hands, revealing the soothing, rhythmic movements that ease tension and promote relaxation. The background features a warm, inviting ambiance with flickering candles and soft instrumental music playing, enhancing the tranquil atmosphere. The person's face, partially visible, reflects a sense of calm and contentment, emphasizing the therapeutic nature of the massage.

Wan2.2-5B+ STAS (Ours)

18.3A thrill-seeker, clad in a bright red jumpsuit and a secure harness, leaps off a towering cliff, the vast canyon below stretching out in breathtaking detail. The camera captures the moment of freefall, the wind rushing past their exhilarated face, eyes wide with a mix of fear and excitement. As they plummet, the rugged landscape blurs, showcasing the sheer height of the jump. The bungee cord stretches taut, and the person rebounds gracefully, their body arching in a fluid motion against the backdrop of a clear blue sky and jagged rock formations. The scene concludes with a close-up of their triumphant smile, hanging upside down, savoring the adrenaline rush and the stunning natural scenery.

Wan2.2-5B+ STAS (Ours)

18.4A vintage steam locomotive chugs along a winding railway through a picturesque countryside, its billowing smoke blending with the early morning mist. The train, with its polished brass and deep green carriages, glides past fields of golden wheat and vibrant wildflowers. As it crosses an old stone bridge, the sound of the wheels clattering on the tracks echoes through the valley. The scene shifts to a close-up of the train's wheels, showcasing the intricate mechanics and the rhythmic motion. Finally, the train approaches a quaint, rustic station, where a few passengers eagerly await its arrival, their silhouettes framed by the soft glow of the rising sun.

Wan2.2-5B+ STAS (Ours)

18.5A serene individual, dressed in a flowing white shirt and dark trousers, sits cross-legged on a grassy hilltop at sunset, playing a wooden flute. The golden light bathes the scene, casting long shadows and illuminating the musician's focused expression. The camera captures close-ups of their fingers deftly moving over the flute's holes, the gentle breeze rustling their hair. As the melody flows, the surrounding wildflowers sway in harmony, and distant mountains provide a majestic backdrop. The scene transitions to a wider shot, revealing the vast, tranquil landscape, with the flute's soothing notes echoing through the serene evening air.

CogVideoX-5B

MA dim 1982 · Figures 19–20 10 comparisons

Figure 19

CogVideoX-5B+ STAS (Ours)

19.1A majestic chestnut horse with a flowing mane stands at the edge of a crystal-clear river, surrounded by lush greenery and wildflowers. The sunlight filters through the trees, casting a golden glow on the scene. The horse gracefully bends its neck, its reflection shimmering in the gentle ripples of the water. As it drinks, the sound of the flowing river and the rustling leaves create a serene ambiance. The horse's muscles ripple under its glossy coat, and a gentle breeze ruffles its mane, adding to the tranquil beauty of the moment.

CogVideoX-5B+ STAS (Ours)

19.2A massive grizzly bear prowls through a dense, misty forest, its fur glistening with morning dew. The bear's powerful muscles ripple beneath its thick coat as it moves silently, its keen eyes scanning the underbrush for any signs of movement. The forest is alive with the sounds of rustling leaves and distant bird calls, creating an atmosphere of tense anticipation. The bear pauses, sniffing the air, its breath visible in the cool morning mist. Suddenly, it spots a deer grazing nearby, its ears twitching nervously. The bear crouches low, its eyes locked on its prey, and then, with a burst of speed, it charges forward, the forest floor trembling under its weight. The chase is swift and intense, the bear's powerful strides closing the distance between predator and prey.

CogVideoX-5B+ STAS (Ours)

19.3A sophisticated couple, dressed in elegant evening attire, navigates through a bustling city street under a heavy downpour. The man, in a sharp black tuxedo, and the woman, in a stunning red gown, both clutch black umbrellas that struggle against the intense rain and wind. The scene is dramatically intensified by a shaking effect, capturing the chaos of the storm. Raindrops cascade off their umbrellas, and their formal wear clings to them, soaked. Streetlights cast a shimmering glow on the wet pavement, and the couple's determined expressions reveal their resolve to reach home despite the tempestuous weather.

CogVideoX-5B+ STAS (Ours)

19.4A sleek, vibrant orange sports car glides effortlessly along a winding coastal road, its glossy finish reflecting the golden hues of the setting sun. The car's aerodynamic design and polished chrome accents catch the light, creating a dazzling display of color and motion. As it speeds past, the roar of its powerful engine echoes against the cliffs, blending with the rhythmic crashing of ocean waves. The camera captures close-up shots of the car's intricate details: the sharp lines of its body, the gleaming alloy wheels, and the luxurious leather interior. The scene transitions to a panoramic view, showcasing the car's journey along the scenic route, with the endless horizon and sparkling sea as a breathtaking backdrop.

CogVideoX-5B+ STAS (Ours)

19.5A sleek, animated shark glides gracefully through the vibrant, turquoise waters of the ocean. Its streamlined body, adorned with shades of blue and gray, moves effortlessly, creating gentle ripples in its wake. The ocean floor below is a tapestry of colorful coral reefs and swaying seaweed, teeming with diverse marine life. Sunlight filters through the water's surface, casting dappled patterns on the shark's skin. As it swims, schools of fish dart around, adding dynamic movement to the scene. The shark's eyes, animated with a hint of curiosity, scan its surroundings, capturing the essence of the ocean's mysterious depths.

Figure 20

CogVideoX-5B+ STAS (Ours)

20.1A majestic brown bear, with its thick fur glistening in the dappled sunlight, begins its ascent up a towering pine tree in a dense forest. The bear's powerful claws grip the rough bark as it climbs higher, its muscles rippling with each movement. The forest floor below is carpeted with fallen leaves and pine needles, creating a serene, earthy backdrop. As the bear reaches a sturdy branch, it pauses to look around, its intelligent eyes scanning the surroundings. The scene captures the raw strength and grace of the bear, set against the tranquil beauty of the forest.

CogVideoX-5B+ STAS (Ours)

20.2A majestic great white shark glides effortlessly through the crystal-clear, azure waters of the ocean, its powerful body cutting through the waves with grace. As the camera zooms in, the intricate details of its rough, textured skin and the sharpness of its dorsal fin become strikingly visible. The sunlight filters through the water, casting shimmering patterns on the shark's sleek form. Its eyes, dark and mysterious, reveal a sense of ancient wisdom and primal instinct. The surrounding marine life, including schools of colorful fish and swaying seaweed, adds to the vibrant underwater scene, highlighting the shark's dominance in its natural habitat.

CogVideoX-5B+ STAS (Ours)

20.3A colossal iceberg drifts majestically in the frigid, azure waters of the Arctic Ocean, its towering, jagged peaks glistening under the soft, ethereal light of the midnight sun. The iceberg's surface is a mesmerizing blend of pristine white and deep blue, with intricate patterns of cracks and crevices hinting at its ancient origins. Seabirds circle above, their calls echoing in the crisp, cold air, while the gentle lapping of waves against the iceberg's base creates a soothing, rhythmic sound. Occasionally, a chunk of ice breaks off, splashing into the water below, sending ripples across the serene, icy expanse. The scene is both awe-inspiring and tranquil, capturing the raw beauty and power of nature in its purest form.

CogVideoX-5B+ STAS (Ours)

20.4A plush teddy bear, with soft brown fur and a red bow tie, sits on a lush green lawn under a bright, sunny sky. Nearby, a vibrant blue frisbee lies on the grass, hinting at playful moments. The scene transitions to the teddy bear being gently tossed into the air, its limbs flailing joyfully, as the frisbee soars in the background. The bear lands softly, surrounded by daisies, while the frisbee spins to a stop beside it. Finally, the teddy bear is propped up against a tree trunk, holding the frisbee in its lap, creating a heartwarming image of companionship and play.

CogVideoX-5B+ STAS (Ours)

20.5A joyful Corgi with a fluffy coat and perky ears frolics in a sunlit park, its tail wagging energetically. The golden hues of the setting sun cast a warm glow on the lush green grass, creating a picturesque scene. The Corgi leaps and bounds, chasing after a bright red ball, its playful antics bringing smiles to onlookers. As the camera tilts up, the vibrant colors of the sunset fill the sky, with streaks of orange, pink, and purple blending seamlessly. The silhouette of the Corgi against the radiant sky captures the essence of pure happiness and the beauty of a perfect evening.

LTX-2 (base)

MA dim 3999 · Figures 21–22 10 comparisons

Figure 21

LTX-2 (base)+ STAS (Ours)

21.1A golden retriever with a shiny coat strolls leisurely through a sun-dappled forest path, the morning light filtering through the trees casting a warm glow. The dog’s tail wags gently as it sniffs the air, ears perked up, taking in the serene surroundings. The camera captures close-ups of its joyful expression, tongue lolling out, and eyes sparkling with contentment. As it walks, the soft crunch of leaves under its paws adds to the tranquil ambiance. The scene transitions to the dog pausing by a clear, babbling brook, lapping up the cool water, before continuing its peaceful journey through the picturesque woodland.

LTX-2 (base)+ STAS (Ours)

21.2A young woman with long, flowing hair stands in a small, dimly lit bathroom, wearing a casual white t-shirt and jeans. She gazes thoughtfully at an old-fashioned porcelain toilet with a wooden seat, the room's vintage tiles adding a nostalgic touch. The scene shifts to her kneeling beside the toilet, her expression one of curiosity and contemplation. She then reaches out to touch the tank, her fingers tracing its contours as if uncovering a hidden story. Finally, she sits on the closed lid, lost in thought, the soft light casting gentle shadows that enhance the room's intimate and reflective atmosphere.

LTX-2 (base)+ STAS (Ours)

21.3A sleek, black motorcycle with chrome accents stands proudly on a winding mountain road, its polished surface gleaming under the midday sun. The camera zooms in to capture the intricate details of the engine, the leather seat, and the handlebars, showcasing the craftsmanship. The scene shifts to the motorcycle speeding along the road, the rider in a black leather jacket and helmet, leaning into a curve with the majestic mountains and a clear blue sky in the background. The roar of the engine echoes through the serene landscape, emphasizing the power and freedom of the ride. Finally, the motorcycle comes to a stop at a scenic overlook, the rider dismounting to take in the breathtaking view, the machine standing as a symbol of adventure and exploration.

LTX-2 (base)+ STAS (Ours)

21.4A vibrant pink chair sits elegantly in the center of a sunlit room, its plush velvet upholstery catching the light. The chair's sleek, modern design features gently curved armrests and polished wooden legs, adding a touch of sophistication. Surrounding the chair, a cozy ambiance is created by soft, pastel-colored walls adorned with minimalist artwork. A nearby window allows golden rays of sunlight to filter through sheer curtains, casting a warm glow on the chair. The scene transitions to a close-up, highlighting the chair's intricate stitching and luxurious texture, inviting viewers to imagine the comfort and style it brings to the space.

LTX-2 (base)+ STAS (Ours)

21.5A pixel art scene depicts a couple in elegant evening attire, caught in a sudden downpour. The man, in a sharp black tuxedo, holds a black umbrella, while the woman, in a flowing red gown, clutches a white umbrella. Raindrops cascade around them, creating a shimmering effect on the cobblestone street. Their expressions show a mix of surprise and amusement as they navigate the wet path. Streetlights cast a warm glow, reflecting off puddles, and the dark, cloudy sky adds a dramatic backdrop. The couple's attire and the vibrant pixel art style bring a nostalgic charm to the rainy night.

Figure 22

LTX-2 (base)+ STAS (Ours)

22.1In a sunlit meadow, a majestic chestnut horse with a flowing mane grazes peacefully beside a fluffy white sheep. The horse, with its sleek coat glistening in the sunlight, occasionally lifts its head to survey the serene landscape. The sheep, with its woolly fleece, nibbles on the lush green grass, staying close to its equine companion. The scene transitions to a playful moment where the horse gently nudges the sheep, and the sheep responds with a soft bleat. The backdrop of rolling hills and a clear blue sky enhances the idyllic and harmonious interaction between the two animals.

LTX-2 (base)+ STAS (Ours)

22.2A joyful individual, bundled in a red winter coat, knitted hat, and gloves, stands in a snow-covered park, rolling a large snowball to form the base of a snowman. The scene is set against a backdrop of snow-laden trees and a serene, overcast sky. Next, they carefully place a smaller snowball on top, forming the snowman's body, their breath visible in the cold air. The person then adds the finishing touches: a carrot for the nose, coal for the eyes and mouth, and a cozy scarf around the snowman's neck. Finally, they step back, admiring their creation with a satisfied smile, the snowman standing proudly amidst the winter wonderland.

LTX-2 (base)+ STAS (Ours)

22.3A delicate porcelain teacup, adorned with intricate floral patterns in soft pastels, sits on a rustic wooden table. Sunlight streams through a nearby window, casting a warm glow and gentle shadows on the cup's surface. The camera zooms in to reveal the fine details of the painted flowers and the elegant gold trim along the rim. Steam rises gracefully from the cup, indicating a freshly brewed tea inside. The scene transitions to a close-up of a hand gently lifting the cup, showcasing the delicate handle and the smooth, glossy finish. The background remains softly blurred, keeping the focus on the exquisite teacup and the serene moment it represents.

LTX-2 (base)+ STAS (Ours)

22.4A stylish couple, dressed in sleek, futuristic evening wear, navigate a neon-lit cityscape under a heavy downpour. The man, in a sharp, metallic silver suit, and the woman, in a shimmering, holographic gown, hold transparent umbrellas that reflect the vibrant, electric hues of the city lights. Rain cascades around them, creating a mesmerizing dance of colors on the wet pavement. Their expressions are a mix of surprise and amusement as they hurry through the rain-soaked streets, the city's towering skyscrapers and holographic advertisements casting an otherworldly glow. The scene captures the essence of a cyberpunk world, blending elegance with the raw energy of a futuristic metropolis.

LTX-2 (base)+ STAS (Ours)

22.5A pixel art astronaut, clad in a white spacesuit with blue accents and a reflective helmet, floats gracefully through the vast expanse of space. Stars twinkle in the dark, pixelated sky, while distant planets and colorful nebulas add depth to the cosmic scene. The astronaut's suit details, including the oxygen tank and control panel, are meticulously rendered in pixel form. As they drift, their arms and legs move slightly, suggesting the weightlessness of space. The background shifts to reveal a massive, pixelated spaceship and a glowing Earth, emphasizing the grandeur and isolation of their journey.


BibTeX

Please consider citing our paper if you find it useful in your research.

@article{cheng2026stas, title={Steering Video Diffusion Transformers with Massive Activations}, author={Cheng, Xianhang and Zheng, Yujian and Xie, Zhenyu and Liao, Tingting and Li, Hao}, journal={arXiv preprint arXiv:2603.17825}, year={2026} }