-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
600 lines (572 loc) · 28.1 KB
/
index.html
File metadata and controls
600 lines (572 loc) · 28.1 KB
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
<!doctype html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>AVGen-Bench: A Task-Driven Benchmark for Multi-Granular Evaluation of Text-to-Audio-Video Generation</title>
<meta
name="description"
content="AVGen-Bench is a task-driven benchmark for multi-granular evaluation of Text-to-Audio-Video generation."
/>
<link rel="stylesheet" href="assets/site.css" />
<!-- 表格专属优化样式 -->
<style>
/* 表格外部容器，增加阴影和圆角以提升现代感 */
.results-table-wrap {
width: 100%;
overflow-x: auto;
margin: 32px 0 20px 0;
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.05), 0 2px 4px -1px rgba(0, 0, 0, 0.03);
border-radius: 8px;
background-color: #ffffff;
border: 1px solid #e5e7eb;
}
/* 表格基础设置 */
.results-table {
width: 100%;
border-collapse: collapse;
font-size: 15px;
white-space: nowrap; /* 防止内容折行 */
}
/* 单元格间距与下划线 */
.results-table th,
.results-table td {
padding: 14px 12px;
border-bottom: 1px solid #e5e7eb;
}
/* 去除最后一行的底边框 */
.results-table tbody tr:last-child td {
border-bottom: none;
}
/* 表头样式：粘性定位、浅灰背景 */
.results-table th {
font-weight: 700;
color: #111827;
background-color: #f8fafc;
position: sticky;
top: 0;
z-index: 10;
}
/* 排版优化：前两列（名称/组件）左对齐，后续指标居中并使用等宽数字对齐小数点 */
.results-table th:nth-child(1),
.results-table td:nth-child(1),
.results-table th:nth-child(2),
.results-table td:nth-child(2) {
text-align: left;
padding-left: 20px;
}
.results-table th:nth-child(n+3),
.results-table td:nth-child(n+3) {
text-align: center;
font-variant-numeric: tabular-nums;
}
/* 交互优化：鼠标悬停整行浅灰高亮，防止视线串行 */
.results-table tbody tr {
transition: background-color 0.15s ease;
}
.results-table tbody tr:hover {
background-color: #f1f5f9;
}
.results-table strong {
font-weight: 700;
color: #000;
}
.results-table em {
font-style: italic;
color: #374151;
}
/* --- 胶囊状标签 (Badges) 样式 --- */
.comp-tag {
display: inline-block;
padding: 2px 8px;
border-radius: 999px;
font-size: 12px;
font-weight: 600;
margin-right: 6px;
margin-bottom: 2px;
letter-spacing: 0.02em;
}
/* 专有组件（橙色系） */
.comp-tag.proprietary {
background-color: #ffedd5;
color: #c2410c;
border: 1px solid #fed7aa;
}
/* 开源组件（蓝色系） */
.comp-tag.open-source {
background-color: #e0f2fe;
color: #0369a1;
border: 1px solid #bae6fd;
}
/* 表格底部的注脚文字 */
p.note {
font-size: 14px;
color: #64748b;
margin-top: 12px;
margin-bottom: 8px;
line-height: 1.5;
}
</style>
</head>
<body>
<header class="hero">
<div class="container">
<h1>AVGen-Bench: A Task-Driven Benchmark for Multi-Granular Evaluation of Text-to-Audio-Video Generation</h1>
<p class="subtitle">
A task-driven benchmark for <strong>multi-granular</strong> evaluation of
<strong>Text-to-Audio-Video (T2AV)</strong> generation.
</p>
<div class="actions">
<a class="btn primary" href="https://github.com/microsoft/AVGen-Bench">
<svg viewBox="0 0 24 24" aria-hidden="true" fill="currentColor">
<path d="M12 .5C5.65.5.5 5.66.5 12.03c0 5.1 3.3 9.43 7.88 10.96.58.11.79-.25.79-.56 0-.28-.01-1.2-.02-2.17-3.2.7-3.88-1.36-3.88-1.36-.52-1.34-1.28-1.7-1.28-1.7-1.05-.72.08-.71.08-.71 1.16.08 1.78 1.2 1.78 1.2 1.03 1.77 2.7 1.26 3.36.97.1-.75.4-1.26.73-1.55-2.55-.29-5.23-1.28-5.23-5.7 0-1.26.45-2.28 1.19-3.09-.12-.29-.52-1.47.11-3.06 0 0 .97-.31 3.17 1.18a10.9 10.9 0 0 1 5.78 0c2.2-1.49 3.17-1.18 3.17-1.18.63 1.59.23 2.77.11 3.06.74.81 1.19 1.83 1.19 3.09 0 4.43-2.69 5.4-5.25 5.69.41.35.78 1.03.78 2.08 0 1.5-.01 2.7-.01 3.07 0 .31.21.68.8.56A11.54 11.54 0 0 0 23.5 12.03C23.5 5.66 18.35.5 12 .5Z"/>
</svg>
<span>Github</span>
</a>
<a class="btn" href="https://arxiv.org/abs/2604.08540">
<svg viewBox="0 0 24 24" aria-hidden="true" fill="currentColor">
<path d="M6 2.75A2.75 2.75 0 0 0 3.25 5.5v13A2.75 2.75 0 0 0 6 21.25h12a2.75 2.75 0 0 0 2.75-2.75V8.62c0-.73-.29-1.43-.8-1.94l-3.63-3.63a2.75 2.75 0 0 0-1.94-.8H6Zm0 1.5h7.75V8c0 .69.56 1.25 1.25 1.25h4.25v9.25c0 .69-.56 1.25-1.25 1.25H6c-.69 0-1.25-.56-1.25-1.25v-13c0-.69.56-1.25 1.25-1.25Zm9.25.31 3.69 3.69h-3.69V4.56ZM7.5 12.25a.75.75 0 0 1 .75-.75h7.5a.75.75 0 0 1 0 1.5h-7.5a.75.75 0 0 1-.75-.75Zm0 3a.75.75 0 0 1 .75-.75h7.5a.75.75 0 0 1 0 1.5h-7.5a.75.75 0 0 1-.75-.75Z"/>
</svg>
<span>Paper</span>
</a>
<a class="btn" href="https://huggingface.co/papers/2604.08540">
<span aria-hidden="true">🤗</span>
<span>HF Paper</span>
</a>
<a class="btn" href="https://huggingface.co/datasets/microsoft/AVGen-Bench">
<span aria-hidden="true">🎬</span>
<span>Hugging Face</span>
</a>
<a class="btn" href="#results">View Results</a>
<a class="btn" href="#failure-demos">Failure Demos</a>
</div>
</div>
</header>
<main>
<section class="section">
<div class="container">
<h2>Framework Overview</h2>
<p>
AVGen-Bench evaluates T2AV systems across three granularities: basic
uni-modal quality, cross-modal alignment, and fine-grained semantic
controllability.
</p>
<figure>
<img src="assets/avbench_outline.png" alt="AVGen-Bench framework overview" />
</figure>
</div>
</section>
<!-- tmp -->
<section class="section alt" id="comparison">
<div class="container">
<h2>Benchmark Comparison</h2>
<p>
Compared with prior benchmarks, AVGen-Bench emphasizes joint AV
evaluation, richer fine-grained metrics, and realistic complex prompts.
</p>
<figure>
<img
src="assets/benchmark_comparison.drawio.png"
alt="Comparison of AVGen-Bench with prior benchmarks"
loading="lazy"
/>
</figure>
</div>
</section>
<section class="section" id="results">
<div class="container">
<h2>Main Quantitative Results</h2>
<p>
AVGen-Bench follows the paper's reporting narrative across visual/audio
quality, synchronization, text/face/music/speech controllability, physical
plausibility, and holistic semantic alignment. The table reports `AV`/`Lip`
as complementary synchronization measurements, while `Lo-Phy` and `Hi-Phy`
are weighted as separate fine-grained dimensions.
</p>
<div class="results-table-wrap">
<table class="results-table">
<thead>
<tr>
<th>Model</th>
<th>Components</th>
<th>Vis</th>
<th>Aud (PQ)</th>
<th>AV</th>
<th>Lip</th>
<th>Text</th>
<th>Face</th>
<th>Music</th>
<th>Speech</th>
<th>Lo-Phy</th>
<th>Hi-Phy</th>
<th>Holistic</th>
<th>Total</th>
</tr>
</thead>
<tbody>
<tr>
<td>Seedance 2.0</td>
<td><span class="comp-tag proprietary">Seedance 2.0</span></td>
<td>0.945</td>
<td><em>7.15</em></td>
<td><strong>0.15</strong></td>
<td>4.14</td>
<td>74.83</td>
<td><strong>60.95</strong></td>
<td><strong>28.12</strong></td>
<td>94.09</td>
<td>3.89</td>
<td><strong>83.16</strong></td>
<td><strong>89.61</strong></td>
<td><strong>72.07</strong></td>
</tr>
<tr>
<td>Veo 3.1-fast</td>
<td><span class="comp-tag proprietary">Veo 3.1-fast</span></td>
<td><em>0.960</em></td>
<td>6.64</td>
<td><em>0.21</em></td>
<td>2.39</td>
<td>75.10</td>
<td>52.77</td>
<td>3.13</td>
<td><em>94.53</em></td>
<td>3.68</td>
<td>67.43</td>
<td>86.27</td>
<td><em>67.87</em></td>
</tr>
<tr>
<td>Veo 3.1-quality</td>
<td><span class="comp-tag proprietary">Veo 3.1-quality</span></td>
<td>0.954</td>
<td>6.77</td>
<td>0.24</td>
<td>3.59</td>
<td><em>76.53</em></td>
<td>52.90</td>
<td>5.00</td>
<td><strong>96.09</strong></td>
<td>3.74</td>
<td>68.53</td>
<td>84.10</td>
<td>66.28</td>
</tr>
<tr>
<td>Sora-2</td>
<td><span class="comp-tag proprietary">Sora-2</span></td>
<td>0.848</td>
<td>5.91</td>
<td>0.25</td>
<td>4.50</td>
<td>74.84</td>
<td>51.17</td>
<td>7.81</td>
<td>88.63</td>
<td><strong>4.05</strong></td>
<td><em>78.95</em></td>
<td><em>88.89</em></td>
<td>64.16</td>
</tr>
<tr>
<td>Wan2.6</td>
<td><span class="comp-tag proprietary">Wan2.6</span></td>
<td>0.959</td>
<td><em>7.15</em></td>
<td>0.30</td>
<td>4.32</td>
<td><strong>76.95</strong></td>
<td>49.27</td>
<td>1.75</td>
<td>89.33</td>
<td>3.69</td>
<td>66.92</td>
<td>80.98</td>
<td>62.97</td>
</tr>
<tr>
<td>Seedance-1.5 Pro</td>
<td><span class="comp-tag proprietary">Seedance-1.5 Pro</span></td>
<td><strong>0.970</strong></td>
<td><strong>7.48</strong></td>
<td>0.26</td>
<td>3.43</td>
<td>38.28</td>
<td>54.42</td>
<td>1.88</td>
<td>93.45</td>
<td>3.72</td>
<td>66.88</td>
<td>77.38</td>
<td>62.55</td>
</tr>
<tr>
<td>Kling-V2.6</td>
<td><span class="comp-tag proprietary">Kling-V2.6</span></td>
<td>0.906</td>
<td>6.93</td>
<td><em>0.21</em></td>
<td><em>2.30</em></td>
<td>14.52</td>
<td><em>57.33</em></td>
<td>5.00</td>
<td>89.62</td>
<td>3.84</td>
<td>63.92</td>
<td>76.74</td>
<td>61.82</td>
</tr>
<tr>
<td>LTX-2.3</td>
<td><span class="comp-tag open-source">LTX-2.3</span></td>
<td>0.858</td>
<td>7.11</td>
<td>0.36</td>
<td><strong>2.00</strong></td>
<td>54.17</td>
<td>45.06</td>
<td>1.38</td>
<td>86.66</td>
<td><em>3.99</em></td>
<td>64.31</td>
<td>65.22</td>
<td>59.97</td>
</tr>
<tr>
<td>NanoBanana2 + MOVA</td>
<td>
<span class="comp-tag proprietary">NanoBanana2</span>
<span class="comp-tag open-source">MOVA</span>
</td>
<td>0.890</td>
<td>6.71</td>
<td>0.44</td>
<td>2.70</td>
<td>68.26</td>
<td>41.33</td>
<td>0.59</td>
<td>82.45</td>
<td>3.91</td>
<td>60.95</td>
<td>72.48</td>
<td>58.10</td>
</tr>
<tr>
<td>LTX-2</td>
<td><span class="comp-tag open-source">LTX-2</span></td>
<td>0.828</td>
<td>6.84</td>
<td>0.23</td>
<td>4.76</td>
<td>24.76</td>
<td>48.53</td>
<td>5.75</td>
<td>87.07</td>
<td><strong>4.05</strong></td>
<td>60.20</td>
<td>66.59</td>
<td>56.62</td>
</tr>
<tr>
<td>Emu3.5 + MOVA</td>
<td>
<span class="comp-tag open-source">Emu3.5</span>
<span class="comp-tag open-source">MOVA</span>
</td>
<td>0.911</td>
<td>6.80</td>
<td>0.38</td>
<td>4.83</td>
<td>64.72</td>
<td>48.44</td>
<td>0.62</td>
<td>81.74</td>
<td>3.89</td>
<td>55.85</td>
<td>66.55</td>
<td>56.12</td>
</tr>
<tr>
<td>Wan2.2 + HunyuanVideo-Foley</td>
<td>
<span class="comp-tag open-source">Wan2.2</span>
<span class="comp-tag open-source">HunyuanVideo-Foley</span>
</td>
<td>0.936</td>
<td>6.60</td>
<td>0.23</td>
<td>5.38</td>
<td>48.46</td>
<td>36.23</td>
<td>3.44</td>
<td>53.40</td>
<td>3.90</td>
<td>54.11</td>
<td>60.63</td>
<td>53.29</td>
</tr>
<tr>
<td>Ovi</td>
<td><span class="comp-tag open-source">Ovi</span></td>
<td>0.839</td>
<td>6.31</td>
<td>0.37</td>
<td>5.40</td>
<td>41.36</td>
<td>49.05</td>
<td><em>11.25</em></td>
<td>76.49</td>
<td>3.93</td>
<td>52.92</td>
<td>57.45</td>
<td>52.02</td>
</tr>
</tbody>
</table>
</div>
<p class="note">
Metric direction: higher is better for Vis, Aud (PQ), Text, Face, Music,
Speech, Lo-Phy, Hi-Phy, and Holistic; lower is better for AV and Lip.
</p>
<p class="note">
Models are sorted by Total in descending order. Bold marks the best score,
and italics mark the second-best score in each metric. Orange tags indicate
proprietary components, while blue tags indicate open-source components.
</p>
</div>
</section>
<section class="section alt">
<div class="container">
<h2>Fine-grained Evaluation Cases</h2>
<div class="grid two">
<figure>
<img
src="assets/finegrained_evaluation_modules.drawio.png"
alt="Fine-grained evaluation modules"
loading="lazy"
/>
<figcaption>Detailed workflow of six fine-grained evaluation modules.</figcaption>
</figure>
<figure>
<img
src="assets/finegrained_example.drawio.png"
alt="Failure examples across fine-grained dimensions"
loading="lazy"
/>
<figcaption>Representative failure modes revealed by AVGen-Bench.</figcaption>
</figure>
</div>
</div>
</section>
<section class="section" id="failure-demos">
<div class="container">
<h2>Failure Demo Videos</h2>
<p>
Multi-model qualitative failures from Appendix A. Each case shows the
original prompt and side-by-side outputs from Veo 3.1 Fast, Ovi, LTX-2,
and Kling 2.6.
</p>
<article class="failure-case" id="case-text-customers">
<h3>Case 1: Prompted Text Rendering ("Your customers are talking")</h3>
<details class="prompt-details">
<summary>Original Prompt</summary>
<p>A single wind-up chattering teeth toy clacks continuously against a solid teal background. The scene cuts to a blue screen displaying the white text "Your customers are talking," abruptly followed by rows of multi-colored chattering teeth toys all moving at once, creating a loud chaotic mechanical clatter. A green screen appears with the text "Are you listening?" before cutting to a generic product logo and a "Try it free" button on a white background as the noise ceases.</p>
</details>
<div class="model-grid">
<article class="demo-card"><h4>Veo 3.1 Fast</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case1/veo3_1_fast.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Ovi</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case1/ovi.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>LTX-2</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case1/ltx2.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Kling 2.6</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case1/kling26.mp4" type="video/mp4" /></video></article>
</div>
</article>
<article class="failure-case" id="case-text-eighty-seven-seconds">
<h3>Case 2: Trailer Title Rendering ("EIGHTY-SEVEN SECONDS")</h3>
<details class="prompt-details">
<summary>Original Prompt</summary>
<p>Four-shot high-tempo teaser with clean sync hits. Shot 1: Inside a bank vault, fluorescent hum and distant alarms; a timer on a device beeps faster as a thief whispers, "Eighty-seven seconds, move." Shot 2: Close-up of a glass cutter scoring a pane with a sharp scratch, then a suction cup pops as the circle lifts free, landing on a bass hit. Shot 3: Smash cut to a getaway car; engine revs, tires chirp, and the car fishtails out of a tight alley with gravel spraying and rattling off the chassis. Shot 4: A final slow-motion shot of a duffel bag hitting the pavement with a heavy thud as sirens surge; the title EIGHTY-SEVEN SECONDS slams onto black with a metallic logo sting.</p>
</details>
<div class="model-grid">
<article class="demo-card"><h4>Veo 3.1 Fast</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case2/veo3_1_fast.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Ovi</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case2/ovi.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>LTX-2</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case2/ltx2.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Kling 2.6</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case2/kling26.mp4" type="video/mp4" /></video></article>
</div>
</article>
<article class="failure-case" id="case-physics-chladni">
<h3>Case 3: Physical Plausibility (Chladni Plate)</h3>
<details class="prompt-details">
<summary>Original Prompt</summary>
<p>A top-down view of a black square metal plate sprinkled evenly with fine white sand as a tone generator plays a pure sine wave that sweeps upward in pitch. As the plate begins to vibrate, the rising tone makes the sand suddenly jitter and chatter across the metal, then fall quiet as grains slide into crisp geometric nodal lines that sharpen and rearrange each time the pitch crosses a new resonance.</p>
</details>
<div class="model-grid">
<article class="demo-card"><h4>Veo 3.1 Fast</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case3/veo3_1_fast.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Ovi</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case3/ovi.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>LTX-2</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case3/ltx2.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Kling 2.6</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case3/kling26.mp4" type="video/mp4" /></video></article>
</div>
</article>
<article class="failure-case" id="case-physics-briggs">
<h3>Case 4: Physical Plausibility (Briggs-Rauscher)</h3>
<details class="prompt-details">
<summary>Original Prompt</summary>
<p>A high-speed time-lapse shows a beaker on a magnetic stirrer, the stir plate motor making a steady whir as a stir bar spins. The beaker contains a Briggs-Rauscher mixture (hydrogen peroxide, potassium iodate, malonic acid, and a metal-ion catalyst with starch indicator). While the vortex turns, the liquid repeatedly cycles through several distinct visible states in a rhythmic pattern, switching abruptly and then returning again and again as the stirring continues.</p>
</details>
<div class="model-grid">
<article class="demo-card"><h4>Veo 3.1 Fast</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case4/veo3_1_fast.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Ovi</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case4/ovi.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>LTX-2</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case4/ltx2.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Kling 2.6</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case4/kling26.mp4" type="video/mp4" /></video></article>
</div>
</article>
<article class="failure-case" id="case-semantic-vacation">
<h3>Case 5: Semantic Misalignment (Vacation Ad)</h3>
<details class="prompt-details">
<summary>Original Prompt</summary>
<p>A young boy hits a beach ball as a group of children runs past him and jumps into a swimming pool with loud splashes, while a voiceover states, "We went on vacation with a toe dipper." The camera follows the kids underwater as bubbles roar and feet kick past the lens, and the voiceover finishes, "and left with a cannonballer." Finally, the view resurfaces to show a laughing girl in the water as on-screen text reads "Book your family home now."</p>
</details>
<div class="model-grid">
<article class="demo-card"><h4>Veo 3.1 Fast</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case5/veo3_1_fast.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Ovi</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case5/ovi.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>LTX-2</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case5/ltx2.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Kling 2.6</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case5/kling26.mp4" type="video/mp4" /></video></article>
</div>
</article>
<article class="failure-case" id="case-music-a4">
<h3>Case 6: Music Pitch Accuracy (Single Note A4)</h3>
<details class="prompt-details">
<summary>Original Prompt</summary>
<p>A zoomed-in tutorial shot of a clean-tone electric guitar fretboard and picking hand. The player frets a single note A4 and plucks it four times with even timing, letting each note ring briefly. The pitch stays stable (no bend, no vibrato), and no other strings ring.</p>
</details>
<div class="model-grid">
<article class="demo-card"><h4>Veo 3.1 Fast</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case6/veo3_1_fast.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Ovi</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case6/ovi.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>LTX-2</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case6/ltx2.mp4" type="video/mp4" /></video></article>
<article class="demo-card"><h4>Kling 2.6</h4><video controls muted playsinline preload="metadata"><source src="assets/failure_demos/videos/case6/kling26.mp4" type="video/mp4" /></video></article>
</div>
</article>
</div>
</section>
<section class="section alt" id="citation">
<div class="container">
<h2>Citation</h2>
<p>If you find AVGen-Bench useful, please cite:</p>
<pre><code>@misc{zhou2026avgenbenchtaskdrivenbenchmarkmultigranular,
title={AVGen-Bench: A Task-Driven Benchmark for Multi-Granular Evaluation of Text-to-Audio-Video Generation},
author={Ziwei Zhou and Zeyuan Lai and Rui Wang and Yifan Yang and Zhen Xing and Yuqing Yang and Qi Dai and Lili Qiu and Chong Luo},
year={2026},
eprint={2604.08540},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2604.08540},
}</code></pre>
</div>
</section>
</main>
<footer>
<div class="container footer-inner">
<span>AVGen-Bench</span>
<a href="https://github.com/microsoft/AVGen-Bench/blob/main/README.md">Documentation</a>
</div>
</footer>
</body>
</html>