File size: 30,645 Bytes
bee9822 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 | <!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
<meta name="description" content="ϕ-Noise: Training-Free Temporal Video Conditioning via Phase-Based Noise Manipulation — Reichman University Canvas-Lab"/>
<title>ϕ-Noise: Training-Free Temporal Video Conditioning</title>
<!-- Fonts -->
<link rel="preconnect" href="https://fonts.googleapis.com"/>
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin/>
<link href="https://fonts.googleapis.com/css2?family=EB+Garamond:ital,wght@0,400;0,500;0,600;1,400&family=DM+Mono:wght@400;500&family=Inter:wght@300;400;500&display=swap" rel="stylesheet"/>
<!-- Icons -->
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.5.0/css/all.min.css"/>
<!-- Project stylesheet -->
<link rel="stylesheet" href="static/css/style.css"/>
</head>
<body>
<!-- ════════════════════════════════════════
NAVIGATION
════════════════════════════════════════ -->
<nav id="navbar">
<a href="#abstract">Abstract</a>
<a href="#project-video">Video</a>
<a href="#applications">Applications</a>
<a href="#overview">Method</a>
<a href="#comparisons">Comparisons</a>
<a href="#bibtex">BibTeX</a>
</nav>
<!-- ════════════════════════════════════════
HERO
════════════════════════════════════════ -->
<section id="hero">
<div class="venue-badge">Preprint · 2026</div>
<h1 style="text-align: center;">
<span class="phi">ϕ</span>-Noise:<br>
Training-Free Temporal Video Conditioning<br>via Phase-Based Noise Manipulation
</h1>
<div class="authors">
<span class="author"><a href="https://ofir1080.github.io/personal-webpage/">Ofir Abramovich</a><sup>*</sup></span>
<span class="author"><a href="https://nadavc220.github.io/">Nadav Z. Cohen</a><sup>*</sup></span>
<span class="author"><a href="https://www.linkedin.com/in/adi-rosenthal-24a3291aa/">Adi Rosenthal</a><sup>*</sup></span>
<span class="author"><a href="https://faculty.runi.ac.il/arik/site/index.html">Ariel Shamir</a></span>
</div>
<div class="affiliations">
Canvas-Lab · Department of Computer Science, Reichman University
</div>
<div class="equal-contrib">* Equal Contribution</div>
<div class="ctas">
<a class="btn btn-outline" href="http://arxiv.org/abs/2605.24509" target="_blank" rel="noopener noreferrer">
<img style="width: 40px; height: 30px"; class="site-icon" src="static/icons/arxiv.png" alt="arXiv logo" />
arXiv
</a>
<a class="btn btn-outline" href="http://arxiv.org/pdf/2605.24509" target="_blank" rel="noopener noreferrer">
<img style="width: 20px; height: 20px"; class="site-icon" src="static/icons/pdf.png" alt="PDF logo" />
PDF
</a>
<a class="btn btn-outline" href="https://huggingface.co/ofirab/phi-noise" target="_blank" rel="noopener noreferrer">
<img style="width: 30px; height: 30px"; class="site-icon" src="static/icons/huggingface.png" alt="Hugging Face logo" />
Hugging Face
</a>
<a class="btn btn-outline" href="https://github.com/ofir1080/phi-noise" target="_blank" rel="noopener noreferrer">
<img style="width: 20px; height: 20px"; class="site-icon" src="static/icons/github.png" alt="GitHub logo" />
Code
</a>
<!-- <a class="btn btn-outline" href="#">
<i class="fas fa-database"></i> Dataset
</a> -->
</div>
</section><!-- /hero -->
<!-- ════════════════════════════════════════
TEASER VIDEO
════════════════════════════════════════ -->
<div id="teaser">
<p class="subtitle text-center muted-strong" style="text-align: center;">
Simple, zero-training video motion control through<br>frequency-domain phase injection into diffusion noise latents.
<br><br>
<div style="text-align: center;">
<img src="static/media/teaser.gif" type="image/gif" style="display: block; margin: 0 auto;">
</div>
<!-- <video autoplay muted loop playsinline class="responsive-video"> -->
<!-- </video> -->
</p>
</div><!-- /teaser -->
<!-- New Video Gallery -->
<section id="project-video">
<div class="container project-video-container">
<!-- <h2>Project Video</h2>
<p class="muted-small">A short project teaser demonstrating ϕ-Noise in action.</p> -->
<video controls playsinline preload="metadata" poster="static/media/repr_frame.png" class="project-video">
<source src="static/media/phi_noise_vid.mp4" type="video/mp4">
</video>
</div>
</section>
<!-- ════════════════════════════════════════
TEASER VIDEO
════════════════════════════════════════ -->
<section id="applications">
<div class="method-figure">
<h2>Applications</h2>
<div class="teaser-row">
<div class="teaser-item">
<h3>I2V Motion Transfer</h3>
<img src="static/media/results/i2v.gif" class="image/gif">
</div>
<div class="teaser-item">
<h3>Motion Transfer (+ Structure Conditioning)</h3>
<img src="static/media/results/t2v.gif" class="image/gif">
</div>
<div class="teaser-item">
<h3>Cut & Drag</h3>
<img src="static/media/results/cnd.gif" class="image/gif">
</div>
</div><!-- /teaser-row -->
</div><!-- /method-figure -->
</section>
<section id="abstract">
<div class="method-figure">
<!-- <article id="abstract" class="overview-card overview-abstract"> -->
<h2>Abstract</h2>
<p class="abstract-text">
Latent video diffusion models generate videos by progressively transforming Gaussian noise
into realistic samples conditioned on text or visual inputs. However, existing conditioning
methods often require additional training and computational overhead. Motivated by recent
findings on the importance of frequency components in generative models, we propose a
simple, training-free approach for motion-conditioned video generation by injecting
low-frequency phase information from a reference video directly into the diffusion noise latents.
</p>
<p class="abstract-text">
Our method transfers motion cues without modifying the model architecture or inference
pipeline. Using several applications, we demonstrate effective control over both appearance
and dynamics in generated videos, while achieving competitive or superior results compared
to more complex conditioning approaches.
</p>
<!-- </article> -->
</section>
<section id="method">
<div class="method-figure">
<h2>How Does It Work?</h2>
<p>
ϕ-Noise operates by decomposing both the reference video latent and the Gaussian noise
latent into the frequency domain via the Discrete Fourier Transform (DFT). The
low-frequency <em>phase</em> components of the noise — which encode coarse spatial
structure and dominant temporal dynamics — are replaced with those from the reference.
A Spectral-Temporal Energy Balancing Mask <strong>Φ</strong> is then applied to preserve
total signal energy and ensure stable denoising.
</p>
<img src="static/teaser_gif.gif" alt="teaser" class="figure-image" />
<p class="teaser-caption">
<strong>ϕ-Noise</strong> injects low-frequency phase information from a reference video
into Gaussian noise latents, enabling training-free motion and structural conditioning
across diverse video generation tasks — with no changes to the diffusion model.
</p>
</div>
</div>
</section>
<!-- ════════════════════════════════════════
METHOD
════════════════════════════════════════ -->
<!-- <section id="method">
<div class="method-figure">
<h2>Method</h2>
<p>
ϕ-Noise operates by decomposing both the reference video latent and the Gaussian noise
latent into the frequency domain via the Discrete Fourier Transform (DFT). The
low-frequency <em>phase</em> components of the noise — which encode coarse spatial
structure and dominant temporal dynamics — are replaced with those from the reference.
A Spectral-Temporal Energy Balancing Mask <strong>Φ</strong> is then applied to preserve
total signal energy and ensure stable denoising.
</p>
<div class="method-figure">
<img src="static/method_v1.png" alt="Method figure" class="figure-image"/>
<p class="fig-caption">
<strong>Figure 1.</strong> Method overview. DFT decomposes both reference and noise into
phase & magnitude. Low-frequency phases from the reference replace those in the noise;
a spectral energy mask Φ restores stability before the modified noise enters the diffusion model.
</p>
</div>
</div>
</section>/method -->
<!-- Method figure image placeholder -->
<section id="analysis">
<div class="method-figure">
<!-- <div class="container"> -->
<h2>Analysis</h2>
<h3 class="underlined">Phase Substitution & Energy Balancing</h3>
<img src="static/analysis.png"
alt="ϕ-Noise method overview figure"
class="figure-image"/>
<p class="fig-caption">
<strong>Figure 2.</strong> Phase and Energy Analysis — phase distributions, latent energy
evolution across denoising steps, and qualitative comparison with/without energy balancing.
</p>
<p>
Naïve phase injection disrupts the expected energy profile of the noise, causing saturation
artifacts and out-of-distribution denoising. ϕ-Noise applies a frequency-dependent scaling
mask <strong>Φ</strong> that scales down injected low-frequency magnitudes by
<code>1/γ</code> while compensating by scaling high frequencies by <code>β</code>,
guaranteeing exact energy preservation: <em>E(z̃<sub>k</sub> ⊙ Φ) = E(z̃)</em>.
</p>
<p>
Because the DFT and its inverse are computationally negligible relative to a single
diffusion step, ϕ-Noise introduces <strong>no additional runtime or memory overhead</strong>
and requires no changes to the model architecture or inference pipeline.
</p>
<!-- </div> -->
</div>
</section><!-- /method -->
<!-- ════════════════════════════════════════
COMPARISONS (tabbed)
════════════════════════════════════════ -->
<section id="comparisons">
<div class="method-figure">
<h2>Comparisons</h2>
<p>
ϕ-Noise supports three distinct video generation tasks within a single unified framework,
all using <strong>WAN (Wan2.2-14B)</strong> as the base model with no additional training.<br>
</p>
<span class="text-red">Choose a task from the tabs below.</span>
<!-- Tab bar -->
<div class="tab-bar">
<button class="tab-btn" data-tab="t2v">
<i class="fas fa-text-width"></i> T2V Motion Transfer
</button>
<button class="tab-btn" data-tab="i2v">
<i class="fas fa-image"></i> I2V Motion Transfer
</button>
<button class="tab-btn" data-tab="cnd">
<i class="fas fa-cut"></i> Cut & Drag
</button>
</div>
<!-- ── TAB: T2V ── -->
<div class="tab-panel" data-panel="t2v">
<p>
Given a reference video and a text prompt, generate a video matching the prompt
while preserving input motion dynamics. Spatial ϕ-Noise achieves strong alignment
with both textual content and motion patterns.
</p>
<!-- Example 1 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Reference</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/climbing/matched_original.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/climbing/matched_ours.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">DMT</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/climbing/matched_dtm.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">DiTFlow</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/climbing/matched_ditflow.webm"></video>
</div>
</div>
<p class="prompt-tag"><strong>Prompt:</strong> "A monkey is climbing a climbing wall."</p>
<!-- Example 2 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Reference</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/car/matched_original.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/car/matched_ours.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">DMT</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/car/matched_dtm.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">DiTFlow</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/car/matched_ditflow.webm"></video>
</div>
</div>
<p class="prompt-tag"><strong>Prompt:</strong> "A motorcycle is driving on the road."</p>
<!-- Example 4 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Reference</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/dog_jump/matched_original_dog_jump.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/dog_jump/matched_ours_dog_jump.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">DMT</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/dog_jump/matched_dmt_dog_jump.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">DiTFlow</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/motion_transfer/dog_jump/matched_ditflow_dog_jump.webm"></video>
</div>
</div>
<p class="prompt-tag"><strong>Prompt:</strong> "A dolphin is jumping in the air into the water."</p>
</div><!-- /tab t2v -->
<!-- ── TAB: I2V ── -->
<div class="tab-panel" data-panel="i2v">
<p>
Align with both a text prompt and a first-frame image condition while following the
motion of the reference video. ϕ-Noise successfully transfers motion across varying
subjects and handles complex dynamics while preserving identity coherence.
</p>
<!-- Example 1 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Reference</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/backflip/matched_original_backflip.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/backflip/matched_ours_backflip.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">MotionClone</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/backflip/matched_mc_backflip.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Wan</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/backflip/matched_wan_backflip.webm"></video>
</div>
</div>
<p class="prompt-tag"><strong>Prompt:</strong> "The person is performing a backflip."</p>
<!-- Example 2 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Reference</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/labubu/matched_original_labubu.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/labubu/matched_aa_ours_labubu.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">MotionClone</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/labubu/matched_mc_labubu.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Wan</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/labubu/matched_wan_labubu.webm"></video>
</div>
</div>
<p class="prompt-tag"><strong>Prompt:</strong> "A motorcycle is driving on the road."</p>
<!-- Example 3 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Reference</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/fly/matched_original.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/fly/matched_ours.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">MotionClone</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/fly/matched_mc.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Wan</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/fly/matched_wan.webm"></video>
</div>
</div>
<p class="prompt-tag"><strong>Prompt:</strong> "The person is sitting, and then suddenly flies magically."</p>
<!-- Example 4 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Reference</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/fish/matched_source_svd_Fish.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/fish/matched_ours_Fish.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">MotionClone</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/fish/matched_mc_Fish.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Wan</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/i2v_motion_transfer/fish/matched_wan_Fish.webm"></video>
</div>
</div>
<p class="prompt-tag"><strong>Prompt:</strong> "The shark is swimming in the ocean."</p>
</div><!-- /tab i2v -->
<!-- ── TAB: Cut & Drag ── -->
<div class="tab-panel" data-panel="cnd">
<p>
Users cut object patches from an image or add sprites, then animate them with rigid
drag paths. ϕ-Noise generates coherent videos following prescribed motion while
producing plausible non-rigid dynamics (e.g., fire breath, tentacle movement).
</p>
<!-- Example 1 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Input</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/humburger/matched_original_Humburger.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/humburger/matched_ours_Hamburger.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Time-to-Move</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/humburger/matched_ttm_Hamburger.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Go-with-the-Flow</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/humburger/matched_gwtf_Hamburger.webm"></video>
</div>
</div>
<!-- Example 2 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Input</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/monkey/matched_original_Monkey.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/monkey/matched_ours_Monkey.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Time-to-Move</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/monkey/matched_ttm_Monkey.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Go-with-the-Flow</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/monkey/matched_gwtf_Monkey.webm"></video>
</div>
</div>
<!-- Example 3 -->
<div class="comparison-row cols-4">
<div class="video-cell">
<span class="cell-label label-ref">Input</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/birds/matched_original_Birds.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label label-ours">Ours</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/birds/matched_ours_Birds.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Time-to-Move</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/birds/matched_ttm_Birds.webm"></video>
</div>
<div class="video-cell">
<span class="cell-label">Go-with-the-Flow</span>
<video muted loop playsinline preload="none" class="comparison-video"><source data-src="static/media/comparisons/cut_n_drag/birds/matched_gwtf_Birds.webm"></video>
</div>
</div>
</div><!-- /tab cnd -->
</div><!-- /container -->
</section><!-- /comparisons -->
<!-- ════════════════════════════════════════
RESULTS
════════════════════════════════════════ -->
<section id="results">
<div class="method-figure">
<h2>Quantitative Results</h2>
<p>
We evaluate on a diverse benchmark of <strong>60 high-quality videos</strong>: 20 from
the TTM dataset, 30 from LOVEU-TGVE-2023, and 10 in-the-wild videos for real-world
generalization. ϕ-Noise achieves competitive or state-of-the-art results at negligible
additional cost.
</p>
<div class="results-table-wrap">
<table>
<thead>
<tr>
<th>Model</th>
<th>CLIP-T ↑</th>
<th>Aes ↑</th>
<th>Img ↑</th>
<th>LPIPS-T ↓</th>
<th>Flow-E ↓</th>
<th>Subj-C ↑</th>
<th>Smooth ↑</th>
<th>Dyn-D ↑</th>
</tr>
</thead>
<tbody>
<tr>
<td>Wan-I2V</td>
<td>0.308</td><td>0.652</td><td>0.644</td>
<td>0.116</td><td>181.10</td><td>0.942</td><td>0.978</td><td>0.647</td>
</tr>
<tr>
<td>GWTF</td>
<td>0.314</td><td>0.620</td><td>0.637</td>
<td>0.097</td><td>152.81</td><td>0.942</td><td>0.981</td><td>0.647</td>
</tr>
<tr>
<td>TTM</td>
<td>0.311</td><td>0.647</td><td>0.653</td>
<td>0.110</td><td>102.39</td><td>0.948</td><td>0.978</td><td>0.705</td>
</tr>
<tr class="ours">
<td><strong>Ours (ϕ-Noise)</strong></td>
<td>0.313</td><td>0.637</td><td>0.627</td>
<td>0.171</td>
<td class="best">101.49</td>
<td>0.918</td><td>0.964</td>
<td class="best">0.764</td>
</tr>
</tbody>
</table>
</div>
<p class="muted-small">
ϕ-Noise achieves the best <strong>Flow-Error</strong> (motion fidelity) and
<strong>Dynamics-Degree</strong> scores — the two metrics most directly measuring motion
transfer quality — while requiring no additional training.
</p>
</div>
</section><!-- /results -->
<!-- ════════════════════════════════════════
BIBTEX
════════════════════════════════════════ -->
<section id="bibtex">
<div class="container">
<h2>BibTeX</h2>
<p>If you find this work useful, please cite:</p>
<div class="bibtex-block" id="bibtex-code">
<button class="copy-btn" onclick="copyBibtex()">
<i class="fas fa-copy"></i> Copy
</button>
<pre><span class="kw">@article</span>{abramovich2025phinoise,
<span class="kw">title</span> = {<span class="val">ϕ-Noise: Training-Free Temporal Video Conditioning
via Phase-Based Noise Manipulation</span>},
<span class="kw">author</span> = {<span class="val">Abramovich, Ofir and Cohen, Nadav Z. and
Rosenthal, Adi and Shamir, Ariel</span>},
<span class="kw">journal</span> = {<span class="val">arXiv preprint</span>},
<span class="kw">year</span> = {<span class="val">2025</span>},
}</pre>
</div>
</div>
</section><!-- /bibtex -->
<!-- ════════════════════════════════════════
FOOTER
════════════════════════════════════════ -->
<footer>
<p>ϕ-Noise · Canvas-Lab, Reichman University · 2025</p>
<p>Website template inspired by
<a href="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">
Academic Project Page Template
</a>
</p>
</footer>
<!-- Project JS -->
<script src="static/js/main.js"></script>
</body>
</html>
|