Paper #11810: Frame Guidance: Training-Free Guidance for Frame-Level Control in Video Diffusion Model

1. Keyframe-guided Video Generation

* You can find each original videos in "./supple_videos/keyframe" *

Video 1-1. Comparison on other keyframe-guided video generation methods.

Video 1-2. Comparison on other keyframe-guided video generation methods.

Video 1-3. Comparison on other keyframe-guided video generation methods.

2. Stylized Video Generation

* You can find each original videos in "./supple_videos/stylized" *

Video 2-1. Comparison on other stylized video generation methods.

Video 2-2. Comparison on other stylized video generation methods.

Video 2-3. Other stylized video generation result. (Left) Style reference image. (Right) Generated video.

Prompt: A New York City street scene with a man and a woman walking down the street, a dog running after them, and a bicyclist passing by, in oil painting style.

Video 2-4. Other stylized video generation result. (Left) Style reference image. (Right) Generated video.

Prompt: A New York City street scene with a man and a woman walking down the street, a dog running after them, and a bicyclist passing by, in watercolor painting style.

Video 2-5. Other stylized video generation result. (Left) Style reference image. (Right) Generated video.

Prompt: A New York City street scene with a man and a woman walking down the street, a dog running after them, and a bicyclist passing by, in Ukiyo-e style.

Video 2-6. Other stylized video generation result. (Left) Style reference image. (Right) Generated video.

Prompt: A bustling Paris cafe in the morning, waiters serving coffee, people chattingat tables, and a dog lying under in Impasto oil painting style with vibrant colors.

Video 2-7. Other stylized video generation result. (Left) Style reference image. (Right) Generated video.

Prompt: A bustling Paris cafe in the morning, waiters serving coffee, people chattingat tables, and a dog lying under in Picasso-inspired cubist aesthetic style.

3. Loop Video Generation

* You can find each original videos in "./supple_videos/loop" *

Video 3-1. Loop video generation using Text-to-Video.

Prompt: A red monster toy jumping on a trampoline in slow motion, landing and bouncing back up endlessly, playful loop.

Video 3-2. Loop video generation using Text-to-Video.

Prompt: A red monster toy jumping on a trampoline in slow motion, landing and bouncing back up endlessly, playful loop.

Video 3-3. Loop video generation using Text-to-Video.

Prompt: A squirrel runs along a tree branch, carrying a nut in its mouth, it stops midway, sniffs the bark, then hides the nut in a small crevice, it turns back and runs the same path again toward another nut on the ground.

4. Other Video Results

* You can find each original videos in "./supple_videos/other" *

Video 4-1. Color Block guidance.

Prompt: A timelapse video of three green mountains gradually transforming into different seasons: the left mountain becomes lush and deep green as summer approaches, the center mountain turns white with snow as winter arrives, and the right mountain turns orange with autumn leaves and slowly gets covered with snow. Cinematic transition, smooth seasonal changes, wide landscape shot.

(Left) First frame (Right) Last frame.

Video 4-2. Color Block guidance.

Prompt: A close-up shot of two colorful liquids …As they meet in the glass, the yellow liquid, being denser, creating a distinct two-layer separation. The interface between the two liquids is sharp and clear.

(Left) First frame (Right) Last frame.