Reward Forcing:
Efficient Streaming Video Generation with
Rewarded Distribution Matching Distillation

Yunhong Lu1,2, Yanhong Zeng2, Haobo Li2,4, Hao Ouyang2, Qiuyu Wang2, Ka Leong Cheng2, Jiapeng Zhu2, Hengyuan Cao1, Zhipeng Zhang5, Xing Zhu2, Yujun Shen2, Min Zhang1,3,
1ZJU   2Ant Group   3SIAS-ZJU   4HUST   5SJTU
Paper 🤗 Hugging Face Code arXiv
teaser

Abstract

Efficient streaming video generation is critical for simulating interactive and dynamic worlds. Existing methods distill few-step video diffusion models with sliding window attention, using initial frames as sink tokens to maintain attention performance and reduce error accumulation. However, video frames become overly dependent on these static tokens, resulting in copied initial frames and diminished motion dynamics. To address this, we introduce Reward Forcing, a novel framework with two key designs. First, we propose EMA-Sink, which maintains fixed-size tokens initialized from initial frames and continuously updated by fusing evicted tokens via exponential moving average as they exit the sliding window. Without additional computation cost, EMA-Sink tokens capture both long-term context and recent dynamics, preventing initial frame copying while maintaining long-horizon consistency. Second, to better distill motion dynamics from teacher models, we propose a novel Rewarded Distribution Matching Distillation (Re-DMD). Vanilla distribution matching treats every training sample equally, limiting the model's ability to prioritize dynamic content. Instead, Re-DMD biases the model's output distribution toward high-reward regions by prioritizing samples with greater dynamics rated by a vision-language model. Re-DMD significantly enhances motion quality while preserving data fidelity. We include both quantitative and qualitative experiments to show that Reward Forcing achieves state-of-the-art performance on standard benchmarks while enabling high-quality streaming video generation at 23.1 FPS on a single H100 GPU.

Object Motion Comparison

More accurate and faithful object dynamics.


SkyReels V2
CausVid
Self Forcing
LongLive
Reward Forcing (Ours)
Click to view Prompt
A close-up shot of a Victoria crowned pigeon in a naturalistic wildlife photography style, showcasing its striking blue plumage and red chest. The bird’s crest is adorned with delicate, lacy feathers, and its eye is a striking red color, adding to its regal and majestic appearance. The pigeon’s head is tilted slightly to the side, giving it a regal gaze. The background is blurred, emphasizing the bird’s striking beauty against a soft, muted backdrop. The lighting highlights the bird’s feathers, creating a vibrant and lifelike image.
Click to view Prompt
A dramatic underwater photograph captures a man performing an intense drumming session. He is submerged in clear blue water, with his face partially obscured by bubbles. His arms move rhythmically, striking the drums with powerful strokes. The drums, made of durable material, are suspended above him, reflecting the vibrant underwater environment. The background features a colorful coral reef with fish swimming around, adding to the vividness of the scene. The water has a soft, ethereal quality, creating a mesmerizing effect. A dynamic low-angle shot from below the surface, emphasizing the man's energetic movements and the aquatic surroundings.
Click to view Prompt
A vibrant and dynamic illustration in a thick-line drawing style of a sleek black cat gracefully performing the tango. The cat has large, expressive green eyes and a playful smile, with its fur flowing naturally as it moves. It stands on two legs, one foot lifted, and the other extended out in a tango pose, with a red scarf tied around its neck. The background features a blurred dance floor with a few blurred dancers in the distance, creating a lively and energetic atmosphere. The scene is set in a dimly lit ballroom with chandeliers hanging overhead. A close-up shot from a slightly elevated angle, capturing the cat's fluid movements and joyful expression.

Scene Navigation Comparison

Smoother transitions for continuous visual movement.


SkyReels V2
CausVid
Self Forcing
LongLive
Reward Forcing (Ours)
Click to view Prompt
A dynamic and lively tour through an art gallery, showcasing a diverse array of beautiful works in various styles. The gallery is filled with paintings, sculptures, and installations, each piece telling its own story. One section features impressionistic landscapes with soft brushstrokes and vibrant colors, capturing serene lakes and rolling hills. Nearby, there are realistic portraits with intricate details and lifelike expressions. In another corner, abstract artworks with bold colors and geometric shapes create a sense of movement and energy. The gallery itself has a modern, open design with high ceilings and large windows allowing natural light to flood in. Visitors move gracefully through the space, pausing occasionally to admire the works. The camera captures the gallery from multiple angles—wide shots of the entire room, close-ups of individual pieces, and sweeping pans to show the flow of visitors. The overall atmosphere is one of inspiration and wonder.
Click to view Prompt
A bustling city market scene at dawn, captured in a documentary-style photograph. People move energetically through the crowded streets, setting up colorful stalls filled with fresh fruits, vegetables, and flowers. Shoppers weave through the lively crowd, their faces illuminated by the early morning light as they pick out the best items. The market is alive with the sounds of haggling and the smells of fresh produce. The background features a mix of old and new buildings, with signs in various languages and a variety of street vendors selling everything from spices to handicrafts. A dynamic wide-angle shot with a sense of movement and depth.
Click to view Prompt
A dramatic exploration scene in a dark, mysterious cave, where an intrepid explorer lumbers forward, flashlight beam casting shadows on the ancient murals etched into the stone walls. The explorer, clad in rugged expedition gear, moves cautiously, each step revealing new sections of intricate, faded paintings. The cave walls are adorned with swirling patterns and mystical symbols, casting eerie glows in the flickering light. The background shows jagged rock formations and dripping stalactites, enhancing the sense of isolation and discovery. The camera angle is from behind the explorer, capturing their determined yet weary gait, with a low-angle shot highlighting the vastness of the cave and the mystery it holds.

More Comparisons


SkyReels V2
CausVid
Self Forcing
LongLive
Reward Forcing (Ours)
Click to view Prompt
A dynamic snowboarding scene in the style of a high-energy action shot, featuring a young snowboarder accelerating down a powdery slope. The snowboarder, with a determined expression, weaves expertly between tall pine trees, their trunks partially obscured by the swirling snow. The snow is pristine and fluffy, with the sun casting soft shadows and highlighting the snowboarder's movements. The background showcases a breathtaking mountain vista, with peaks shrouded in mist and a few distant ski lifts visible. The camera angle captures the snowboarder from a slightly behind-the-action perspective, emphasizing their speed and agility.
Click to view Prompt
A stunning mid-afternoon landscape photograph with a low camera angle, showcasing several giant wooly mammoths treading through a snowy meadow. Their long, wooly fur gently billows in the brisk wind as they move, creating a sense of natural movement. Snow-covered trees and dramatic snow-capped mountains loom in the distance, adding to the majestic setting. Wispy clouds and a high sun cast a warm glow over the scene, enhancing the serene and awe-inspiring atmosphere. The depth of field brings out the detailed textures of the mammoths and the snowy environment, capturing every nuance of these prehistoric giants in breathtaking clarity.
Click to view Prompt
A vibrant tropical fish swimming gracefully among colorful coral reefs in a clear, turquoise ocean. The fish has bright blue and yellow scales with a small, distinctive orange spot on its side, its fins moving fluidly. The coral reefs are alive with a variety of marine life, including small schools of colorful fish and sea turtles gliding by. The water is crystal clear, allowing for a view of the sandy ocean floor below. The reef itself is adorned with a mix of hard and soft corals in shades of red, orange, and green. The photo captures the fish from a slightly elevated angle, emphasizing its lively movements and the vivid colors of its surroundings. A close-up shot with dynamic movement.
Click to view Prompt
A documentary-style nature photography shot from a camera truck moving to the left, capturing a crab quickly scurrying into its burrow. The crab has a hard, greenish-brown shell and long claws, moving with determined speed across the sandy ground. Its body is slightly arched as it burrows into the sand, leaving a small trail behind. The background shows a shallow beach with scattered rocks and seashells, and the horizon features a gentle curve of the coastline. The photo has a natural and realistic texture, emphasizing the crab's natural movement and the texture of the sand. A close-up shot from a slightly elevated angle.
Click to view Prompt
A dynamic action shot of a surfer accelerating on a powerful wave, carving through the water with grace and agility. The surfer, with a tanned complexion and muscular build, rides the wave with one hand gripping the board while the other extends outwards for balance. The water splashes behind, creating a foamy trail, and the sun casts a golden glow over the scene. The background features a clear blue ocean and distant white-capped waves, with a few seagulls flying overhead. The surfer's expression is one of exhilaration and focus. A mid-shot from a low-angle perspective capturing the surfer's motion and the wave's power.
Click to view Prompt
A dynamic action shot of a jet ski speeding across the water, generating massive waves and splashes. The jet ski is sleek and powerful, with its engine roaring and smoke trailing behind it. The rider is fully focused, gripping the handlebars tightly, and leaning forward with determination. The water reflects the bright sun, creating a shimmering effect. The background shows a vast, blue sea with white-capped waves and distant sailboats, adding to the sense of motion and adventure. The photo has a vibrant, high-action style, capturing the exhilaration of the moment.
Click to view Prompt
A low-altitude first-person perspective camera tracking shot of a soccer player's feet as they skillfully dribble the ball across the green soccer field. The player, wearing a white jersey with blue shorts, moves with agility and precision, the ball bouncing rhythmically under their feet. The camera follows closely, capturing every detail of the motion, from the subtle movements of the player's legs to the way the grass blades sway gently. The background shows a blurred but recognizable soccer field with distant players and spectators, adding to the dynamic feel of the scene. The video has a sports videography style, with smooth motion tracking and a slightly grainy texture. A first-person perspective tracking shot.

Interactive Videos

Generated using our method (Reward Forcing)


Show Prompts
Prompt A A bird's-eye shot of a glass bowl filled with clear water, perfectly still and transparent.
Prompt B In the same overhead view, a drop of black ink falls into the water, blooming into ethereal wisps and tendrils.
Show Prompts
Prompt A A cinematic wide shot of a serene forest river, its crystal-clear water flowing gently over smooth stones. Sunlight filters through the canopy, creating dancing caustics on the riverbed. The camera tracks slowly alongside the flow.
Prompt B In the same wide shot, a wave of freezing energy passes through the frame. The flowing water instantly crystallizes into a solid, glistening sheet of ice, trapping air bubbles inside. The camera continues its track, now revealing a tiger cautiously stepping onto the frozen surface, its breath visible in the suddenly cold air.
Show Prompts
Prompt A A close-up of a perfect ice cube sitting on a polished granite counter, edges sharp and crystalline.
Prompt B In the same close-up view, the ice cube slowly melts, forming a growing puddle of water around its base.
Show Prompts
Prompt A A macro still-life shot of an empty glass on a rustic wooden table, softly lit by morning light.
Prompt B In the same macro view, dark coffee streams into the empty glass, forming elegant ripples as it fills.

More Videos

Generated using our method (Reward Forcing)


Show Prompts
A chameleon in a spy-themed outfit, blending seamlessly into diverse backgrounds. The chameleon wears a sleek black suit with green and brown camouflage patterns, matching the surrounding environments. It stands in a forest setting, with leaves and branches creating a natural backdrop. The chameleon's posture is alert and poised, with its tail curled slightly behind it. Its large, expressive eyes scan the surroundings, hinting at its espionage activities. The photo captures the chameleon mid-movement, with a slight tilt of its head and a subtle shift of its body, showcasing its agility and adaptability. A dynamic close-up shot from a slightly elevated angle.
Show Prompts
A high-speed action shot of a cheetah in its natural habitat, sprinting at full speed while chasing its prey across the savanna. The cheetah's golden fur glistens under the bright African sun, and its muscular body is stretched out in a powerful run. Its sharp eyes focus intently on the fleeing antelope, and its distinctive black tear marks streak down its face. The background is a blurred landscape with tall grass swaying in the wind, and distant acacia trees. The cheetah's tail is raised high, and its paws leave deep prints in the soft earth. A dynamic mid-shot capturing the intense moment of pursuit.
Show Prompts
A surreal and dreamlike scene in the style of a cyberpunk film, depicting New York City submerged underwater, resembling the mythical city of Atlantis. Fish, whales, sea turtles, and sharks swim through the bustling streets, which now resemble underwater landscapes. The buildings are partially submerged, their facades covered in algae and marine growth. The water is murky and filled with sunlight filtering through from above, casting colorful hues. Pedestrians, now merfolk, move gracefully through the water, interacting with the aquatic creatures. The camera angle is from a low, sweeping shot, capturing the vast expanse of this submerged metropolis.
Show Prompts
A wide-angle shot of a vast desert landscape at sunset, capturing dunes stretching into the distance under a sky ablaze with vibrant oranges, pinks, and purples. The sun is setting, casting long shadows across the golden sand dunes. The background features a dramatic horizon line, with the sky gradually fading to deep indigo as night approaches. The foreground includes some scattered rocks and twisted cacti, adding texture and depth to the scene. The photo has a naturalistic and atmospheric quality, emphasizing the vastness and beauty of the desert at twilight.
Show Prompts
A dynamic first-person view of a cyclist navigating through a bustling city street, weaving skillfully between traffic and pedestrians. The cyclist is a young adult, wearing a helmet and a casual cycling jersey, pedaling energetically with a determined expression. The camera follows the cyclist closely, capturing the rush of the city around them. Buildings line both sides of the street, with shop signs and advertisements visible. Cars honk and pass by, while people walk briskly past. The cyclist's movements are fluid and purposeful, conveying a sense of urgency and determination. The background shows a mix of modern urban architecture and lively street life, with occasional glimpses of sunlight filtering through tall buildings. A close-up shot from a first-person perspective, emphasizing the cyclist's motion and the chaotic yet vibrant city environment.
Show Prompts
A classic black-and-white photograph style image of an older man playing the piano. The man, with a weathered face and kind eyes, sits at an antique piano with his fingers gracefully moving over the keys. The lighting comes from the side, casting dramatic shadows on his face and emphasizing the texture of his hands. His posture is upright and focused, conveying a sense of deep concentration and passion for music. The background is blurred, revealing only hints of a cozy room with wooden floors and old furniture. A close-up shot from a slightly elevated angle, capturing both the man and the piano in detail.
Show Prompts
A close-up 3D animated scene of a short, fluffy monster kneeling beside a melting red candle. The monster has large, wide eyes and an open mouth, gazing at the flame with a look of wonder and curiosity. Its soft, fluffy fur contrasts with the warm, dramatic lighting that highlights every detail of its gentle, innocent expression. The pose conveys a sense of playfulness and exploration, as if the creature is discovering the world for the first time. The background features a cozy, warmly lit room with subtle hints of a fireplace and soft furnishings, enhancing the overall atmosphere. The use of warm colors and dramatic lighting creates a captivating and inviting scene.
Show Prompts
A close-up shot of a bright blue parrot's shimmering feathers, capturing the unique and vibrant colors in the light. The parrot's feathers glisten with a metallic sheen, showcasing a mix of deep indigos, vivid greens, and rich blues. Its eyes sparkle with curiosity, and it appears lively and alert, perched on a branch. The background is blurred, highlighting the parrot against a soft, warm environment. The photo has a naturalistic and lifelike quality, emphasizing the bird's detailed plumage and natural movements.

BibTeX

@misc{lu2025rewardforcingefficientstreaming,
      title={Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillation}, 
      author={Yunhong Lu and Yanhong Zeng and Haobo Li and Hao Ouyang and Qiuyu Wang and Ka Leong Cheng and Jiapeng Zhu and Hengyuan Cao and Zhipeng Zhang and Xing Zhu and Yujun Shen and Min Zhang},
      year={2025},
      eprint={2512.04678},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2512.04678}, 
}