Abstract
We introduce OmniCreator, a novel framework that can conduct text-prompted unified (image+video) generation as well as editing all in one place. OmniCreator acquires generative and universal editing capabilities in a self-supervised manner, taking original text-video pairs as conditions while utilizing the same video as a denoising target to learn the semantic correspondence between video and text. During inference, when presented with a text prompt and a video, OmniCreator is capable of generating a target that is faithful to both, achieving a universal editing effect that is unconstrained as opposed to existing editing work that primarily focuses on certain editing types or relies on additional controls (e.g., structural conditions, attention features, or DDIM inversion). On the other hand, when presented with a text prompt only, OmniCreator becomes generative, producing high-quality video as a result of the semantic correspondence learned. Importantly, we found that the same capabilities extend to images as is, making OmniCreator a truly unified framework. Further, due to the lack of existing generative video editing benchmarks, we introduce the OmniBench-99 dataset, designed to evaluate the performance of generative video editing models comprehensively. Extensive experiments demonstrate that OmniCreator exhibits substantial superiority over all other models.
Overview Video
Applications
Text-guided Video Editing = Reference Video + Editing Prompt
Editing Types
Foreground: "A volcano"
Edited Video
Background: "Dark sky"
Edited Video
Composite: "Dark sky, Game style mountain"
Edited Video
Overall/Style: "Oil painting style"
Edited Video
Editing Scenarios
Weather: "It's snowing"
Edited Video
Time: "At sunset"
Edited Video
Background: "Red background"
Edited Video
Addition: "Fishing boats"
Edited Video
Replacement: "Roses"
Edited Video
Removal: "With no ship"
Edited Video
Appearance: "Tony Stark"
Edited Video
Motion/Pose: "Standing"
Edited Video
Text-to-Video Generation = Target Prompt
16 Frames
"A balloon floating away into the sky on a windy day."
"A robot assembling a car in a futuristic factory."
"A city skyline with fireworks exploding in the sky."
"An eagle flying over the mountains at dawn."
"A man hiking through a dense forest during autumn."
"A time-lapse of a flower blooming in a garden."
"A panda eating bamboo in a serene forest."
"A thunderstorm rolling in over the ocean with lightning strikes."
"A sunset casting long shadows over rolling hills."
"A diver exploring a coral reef underwater."
"A serious forest fire."
"A calm coastline with choppy waves."
32 Frames
"A thunderstorm with lightning striking across the sky."
"A serene forest with a waterfall flowing into a pond."
"A farmer plowing a field with an old tractor."
"A spooky river in the middle of a dense jungle."
64 Frames
"A cat resting on a sunny windowsill."
"A shipwreck under water."
"A sunset over a calm ocean."
"Fireworks exploding in the air."
Text-guided Image Editing = Reference Image + Editing Prompt
Addition: "A canoe"
Edited Image
Replacement: "A red parrot"
Edited Image
Removal: "With no cup"
Edited Image
Background: "A desert"
Edited Image
Style: "Van Gogh style"
Edited Image
Texture: "Crystal texture"
Edited Image
Action: "Flying"
Edited Image
Text-to-Image Generation = Target Prompt
"A serene mountain landscape at dawn, with soft mist covering the peaks and golden sunlight casting long shadows."
"An underwater scene with a vibrant coral reef, colorful fish swimming around, and sunlight filtering from above."
"A small, cozy cabin in the middle of a snowy forest, warm light glowing from its windows under a star-filled sky."
"A sleek, modern kitchen with stainless steel appliances, white marble countertops in natural light."
"A fantasy castle floating above the clouds, surrounded by ethereal light and mysterious mist."
"A beach at sunset, with soft waves lapping against the shore and a palm tree casting a shadow over the sand."
"A young woman with curly hair and a warm smile, standing in a sunflower field under a bright blue sky."
"A vibrant underwater world where buildings bioluminesce and a variety of sea creatures swim through the streets."
"A dramatic canyon landscape with towering rock formations, a winding river below, and a clear blue sky."
"A detailed close-up of an owl perched on a branch in a snowy forest, its feathers fluffed and eyes intense."
"A mystical cave with glowing crystals and an underground lake."
"Desert scene with towering sand dunes and traces of traffic on the sand."
…
Comparisons
Editing Types
Foreground: "A volcano"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Background: "Dark sky"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Style: "Oil painting style"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Composite: "Dark sky, Game style mountain"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Editing Scenarios
Weather: "Stormy skies with rain and mist"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Time: "Warm golden tones at sunset"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Background: "Distant mountain visible"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Addition: "With multiple bees"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Removal: "With no airplane"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Replacement: "A cruise ship"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Appearance: "A woman wearing a blue coat"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
Motion: "Pecking"
OmniCreator (Ours)
CCEdit
ControlVideo
InsV2V
Pix2Video
Tune-A-Video
TokenFlow
Video-P2P
…
BibTeX
Coming soon...
Project page template is borrowed from DreamBooth.