Skip to content

Commit

Permalink
Annotations
Browse files Browse the repository at this point in the history
  • Loading branch information
RyannDaGreat committed Jan 16, 2025
1 parent 4cdf632 commit 12dc6c5
Showing 1 changed file with 15 additions and 1 deletion.
16 changes: 15 additions & 1 deletion index.html
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,7 @@
window.location.href = 'https://vgenai-netflix-eyeline-research.github.io/Go-with-the-Flow/';
}
</script>

<title>Go with the Flow</title>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">

Expand Down Expand Up @@ -2109,6 +2109,7 @@ <h2 class="title is-3">Image-to-Image Based Video Generation Applications: Relig
the optical flow of the input video.
<br>Here, we show various noise warping and interpolation techniques side-by-side,
for both relighting using DiffRelight and super-resolution using DeepFloyd Stage II.
<br>What's cool about using noise warping on image-to-image translation: it's a training-free way to get better temporal consistency from image diffusion models that were never trained on videos!
</p>
</div>
</div>
Expand All @@ -2120,6 +2121,18 @@ <h2 class="title is-3">Image-to-Image Based Video Generation Applications: Relig
<div class="container" id="carousel">
<div id="slick">

Here's a comparison of DiffRelight, an image-to-image relighting algorithm that works in Stable Diffusion's latent space, where we compare using the same initial noise on every frame (fixed noise) versus our more temporally consistent warped noise. Note how there are less "sticking artifacts"!
<div class="slide">
<div class="card">
<video class="img-fluid" autoplay loop muted playsinline>
<source
src="assets/kevin_fixed_vs_warped.mp4"
type="video/mp4">
</video>
</div>
</div>

Here's another DiffRelight comparison, where we compare many noise strategies.
<div class="slide">
<div class="card">
<video class="img-fluid" autoplay loop muted playsinline>
Expand All @@ -2130,6 +2143,7 @@ <h2 class="title is-3">Image-to-Image Based Video Generation Applications: Relig
</div>
</div>

And here's an example of us warping noise with many strategies using DeepFloyd's Stage II superresolution model, which runs in RGB pixel-space as opposed to latent space!
<div class="slide">
<div class="card">
<video class="img-fluid" autoplay loop muted playsinline>
Expand Down

0 comments on commit 12dc6c5

Please sign in to comment.