Minghan Li*, Chenxi Xie*, Yichen Wu, Lei Zhang, Mengyu Wang
Harvard University,
Hong Kong Polytechnic University, City University of Hong Kong
Two RF-based Video Editing Models
Video ID: 0002_girl_dog
Source prompt:
A young girl in a wheelchair is moving along a paved path beside a grassy area, accompanied by a gray dog walking alongside her. The environment is a sunny urban setting with a mix of grass and paved walkways. The camera remains stationary, capturing the girl's smooth movement and the dog's relaxed pace.
Target prompt:
Edit1: A gray dog ➡️ A gray rabbit; Edit2: A young girl ➡️ A young alien;
Edit3: A gray dog ➡️ A pink dog; Edit4: A wheelchair ➡️ A wooden wheelchair;
Edit6: accompanied by a gray dog ➡️ not accompanied by a gray dog
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Video ID: 0009_bear
Source prompt:
A large brown bear is walking slowly across a rocky terrain in a zoo enclosure, surrounded by stone walls and scattered greenery. The camera remains fixed, capturing the bear's deliberate movements.
Target prompt:
Edit1: A large brown bear ➡️ A large panda; Edit2: A large brown bear ➡️ A large dinosaur;
Edit3: A large brown bear ➡️ A large purple bear; Edit4: A large brown bear ➡️ A large bronze bear;
Edit5: A large brown bear ➡️ A large brown bear wearing a cap
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Video ID: 0011_lucia
Source prompt:
A woman in a black dress is walking along a paved path in a lush green park, with trees and a wooden bench in the background. The camera remains fixed, capturing her steady movement.
Target prompt:
Edit1: A woman in a black dress ➡️ A man in a black suit; Edit2: A woman in a black dress ➡️ A lion;
Edit3: A woman in a black dress ➡️ A woman in a red dress; Edit4: A woman ➡️ A porcelain woman;
Edit5: A woman in a black dress ➡️ A woman in a black dress, followed closely by a dog
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Video ID: 0019_tennis_vest
Source prompt:
A tennis player in a navy tank top and white shorts is actively engaged in a tennis match on a clay court. The player is seen moving dynamically across the court, preparing to hit the ball with a tennis racket. The background features a blue banner and a chain-link fence, typical of a tennis court environment. The camera remains stationary, capturing the player's swift movements and actions.
Target prompt:
Edit1: A tennis player ➡️ An ultraman; Edit2: A tennis player ➡️ A robot;
Edit3: white shorts ➡️ black shorts; Edit4: A tennis player ➡️ A clay tennis player;
Edit5: A tennis player ➡️ A tennis player wearing a bright yellow fedora
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Video ID: 0035_helicopter
Source prompt:
A helicopter is hovering above a helipad at an airport, with a ground crew member in a reflective vest walking towards it. The camera remains stationary, capturing the scene with the helicopter's rotors spinning rapidly.
Target prompt:
Edit1: A helicopter ➡️ A drone; Edit2: A helicopter ➡️ A UFO;
Edit3: A helicopter ➡️ A red helicopter; Edit4: A helicopter ➡️ A wooden helicopter;
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Video ID: 0043_parkour
Source prompt:
A man is performing parkour by skillfully jumping over barriers and climbing over a railing in an urban environment. The camera remains stationary, capturing his agile movements as he navigates the concrete landscape.
Target prompt:
Edit1: A man ➡️ A Spiderman; Edit2: A man ➡️ A bear;
Edit3: A man ➡️ A man in purple; Edit4: A man ➡️ A holographic man;
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Video ID: 0062_elephant
Source prompt:
An elephant is walking slowly across a rocky enclosure in a zoo, with dust rising around its feet. The camera remains fixed, capturing the elephant's steady movement against the backdrop of trees and a building.
Target prompt:
Edit1: An elephant ➡️ A rhinoceros; Edit2: An elephant ➡️ A jeep;
Edit3: An elephant ➡️ A blue elephant; Edit4: An elephant ➡️ A stone elephant;
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Generated Videos
Video ID: 0065_twist_dance
Source prompt:
Two dancers in matching red skirts and black tops are performing a synchronized twist dance routine in a brightly lit dance studio with mirrored walls and a checkered backdrop. The camera remains stationary, capturing their energetic movements and coordination.
Target prompt:
Edit1: Two dancers ➡️ Two Spiderman; Edit2: Two dancers ➡️ Two robots;
Edit3: Two dancers in matching red skirts and black tops ➡️ Two dancers in matching white skirts and white tops;
Edit4: Two dancers ➡️ Two holographic dancers
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Video ID: 0083_A_boat
Source prompt:
A boat is sailing smoothly across a calm river, with trees and a small dock in the background. The camera remains fixed, capturing the boat's serene motion.
Target prompt:
Edit1: A boat ➡️ A kayak; Edit2: A boat ➡️ A swan;
Edit3: A boat ➡️ A magenta boat; Edit4: A boat ➡️ A bamboo boat
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Challenging Videos
Video ID: 0063_drift_turn
Source prompt:
A race car is skillfully drifting around a track, leaving a trail of smoke behind it as spectators watch from the stands. The camera follows the car's movement quickly, capturing the dynamic turn.
Target prompt:
Edit1: A race car ➡️ A race jeep; Edit2: A race car ➡️ A hovercraft;
Edit3: A race car ➡️ A black race car; Edit4: A race car ➡️ A carbon fiber race car;
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Video ID: 0068_planes_water
Source prompt:
A large passenger airplane is taxiing slowly on a wet runway at Schiphol Airport during dusk, with the sky overcast and lights illuminating the path. The camera remains fixed, capturing the airplane's gradual movement forward.
Target prompt:
Edit1: A large passenger airplane ➡️ A large cargo airplane;
Edit2: A large passenger airplane ➡️ A large spaceship;
Edit3: A large passenger airplane ➡️ A blue large passenger airplane;
Edit4: A large passenger airplane ➡️ A chrome large passenger airplane;
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit (Ours)
(h) Wan-Edit (Ours)
Video ID: 0088_cyclist
Source prompt:
A cyclist is pedaling vigorously along a tree-lined path in a city park. The camera follows the cyclist at a steady pace.
Target prompt:
Edit1: A cyclist ➡️ A skateboarder; Edit2: A cyclist ➡️ A rollerblader;
Edit3: A cyclist ➡️ A maroon cyclist; Edit4: A cyclist ➡️ A cardboard cyclist
(a) Source video
(b) TokenFlow
(c) DMT
(d) VidToMe
(e) AnyV2V
(f) VideoGrain
(g) Pyramid-Edit(Ours)
(h) Wan-Edit (Ours)
