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 &nbsp;·&nbsp; 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 &amp; 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 &amp; 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>&nbsp; T2V Motion Transfer
      </button>
      <button class="tab-btn" data-tab="i2v">
        <i class="fas fa-image"></i>&nbsp; I2V Motion Transfer
      </button>
      <button class="tab-btn" data-tab="cnd">
        <i class="fas fa-cut"></i>&nbsp; Cut &amp; 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">&#x3d5;-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 &nbsp;·&nbsp; Canvas-Lab, Reichman University &nbsp;·&nbsp; 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>