Qwen3.5-Omni: Scaling Up, Toward Native Omni-Modal AGI

QWEN CHAT Hugging Face Offline Demo Hugging Face Realtime Demo ModelScope Offline Demo ModelScope Realtime Demo

Qwen3.5-Omni is Qwen’s latest generation of fully omnimodal LLM, supporting the understanding of text, images, audio, and audio-visual content. Both the Thinker and Talker in Qwen3.5-Omni adopt the Hybrid-Attention MoE. Qwen3.5-Omni series includes Instruct versions in three sizes: Plus, Flash, and Light, with support for 256k long-context input. The model can process more than 10 hours of audio input and over 400 seconds of 720P audio-visual input at 1 FPS. It is natively pretrained in an omnimodal manner on massive amounts of text, visual data, and more than 100 million hours of audio-visual data, demonstrating outstanding full-modality perception and generation capabilities. Compared with Qwen3-Omni, Qwen3.5-Omni offers significantly enhanced multilingual capabilities, supporting speech recognition in 113 languages/dialects and speech generation in 36 languages/dialects. It is currently available via the Offline API and Realtime API.

Offline

Qwen3.5-Omni-Plus has achieved SOTA results on 215 audio and audio-visual understanding, reasoning, and interaction subtasks/benchmarks, covering 3 audio-visual benchmarks, 5 audio benchmarks, 8 ASR benchmarks, 156 language-specific S2TT tasks, and 43 language-specific ASR tasks. In particular, it surpasses Gemini-3.1 Pro across general audio understanding, reasoning, recognition, translation, and dialogue, while its overall audio-visual understanding reaches the level of Gemini-3.1 Pro. Meanwhile, its visual and text capabilities match those of Qwen3.5 models of the same size. One of Qwen3.5-Omni-Plus’s standout features is its audio and audio-visual captioning capability, which can generate controllable, detailed, and structured captions, as well as screenplay-level fine-grained descriptions, including automatic segmentation, timestamp annotation, and detailed descriptions of characters and their relationship to audio. In addition, through native multimodal scaling, we observed the emergence of a new capability in omnimodal models: directly performing coding based on audio-visual instructions, which we call Audio-Visual Vibe Coding; all of the above features are available through the Offline API. We strongly encourage users to read the Demo Section.

Realtime

Beyond its strong base capabilities, we further focused on enhancing the interactive abilities of Qwen3.5-Omni. First, we support semantic interruption by developing native turn-taking intent recognition based on Omni, which avoids interruptions caused by backchanneling and meaningless background noise; this capability is already natively supported in the API. Second, we natively support WebSearch and complex FunctionCall capabilities, enabling the model to autonomously decide whether to invoke WebSearch to respond to users’ real-time questions. Third, we support end-to-end voice control and dialogue, allowing the model to follow instructions like a human and freely control aspects such as speaking volume, speed, and emotion. Fourth, Qwen3.5-Omni supports voice cloning, allowing users to upload a voice to customize the AI Assistant’s voice; all of the above features are available through the Realtime API. Users can also modify the system prompt to change the model’s behavior, such as its conversational style or identity. Fifth, to address speech instability in streaming voice interaction caused by differences in text and speech token encoding efficiency—such as omissions, misreadings, or unclear pronunciation of numbers—we propose ARIA (Adaptive Rate Interleave Alignment), a technique that dynamically aligns text and speech units. While preserving real-time performance, ARIA significantly improves the naturalness and robustness of speech synthesis. We strongly encourage users to read the Demo Section to experience the model’s latest capabilities.

Performance#

Audio-Visual Next

Audio Next

Visual Next

Text Next

Speech Generation Next

Below we present the comprehensive evaluation of our models against frontier models in a wide range of evaluation tasks, covering different tasks and modalities.

Audio-Visual#

* VideoMME (with audio): We evaluate our model with use_audio_in_video=True.
* OmniGAIA：We evaluate our model without a thinking prompt and without <answer> formatting. All results are evaluated using deepseek-v3.2-thinking.

Audio#

* SongFormBench: We use a unified prompt defining an SRT-like output timestamp format and a closed vocabulary for evaluation. The vocabulary follows the 'SongForm-HX-8Class' specified in the official codebase.
* URO-Bench-Pro: We use the pro track of URO-Bench and denote the three evaluation dimensions as follows: U for Understanding, R for Reasoning, and O for Oral Conversation. We use GenStyle-en, GenStyle-zh, Multilingual tasks for oral dimension.
* ASR performance is evaluated using WER/CER, where lower values indicate better performance.
* Fleurs: The top59 languages are English, Chinese, Cantonese, Korean, Japanese, Vietnamese, Thai, Malay, German, Russian, Italian, French, Spanish, Portuguese, Dutch, Indonesian, Turkish, Arabic, Polish, Hindi, Urdu, Filipino, Persian, Czech, Greek, Swedish, Hebrew, Danish, Finnish, Norwegian, Icelandic, Bengali, Punjabi, Javanese, Marathi, Swahili, Ukrainian, Gujarati, Kannada, Azerbaijani, Malayalam, Cebuano, Kazakh, Romanian, Hungarian, Bulgarian, Belarusian, Catalan, Tamil, Croatian, Bosnian, Slovak, Galician, Kyrgyz, Macedonian, Slovenian, Latvian, Estonian, and Asturian; compared with the top60 list, Afrikaans is excluded because the Fleurs S2TT test set does not cover this language.
* MIR-1K: Transcription is converted into Simplified Chinese.

Visual#

Text#

* Qwen3.5-Plus-NoThinking: we use Qwen3.5-Plus-NoThinking as the primary baseline, since all models here are evaluated in the same no-thinking setting.
* TAU2-Bench: we follow the official setup except for the airline domain, where all models are evaluated by applying the fixes proposed in the Claude Opus 4.5 system card.

Speech-Generation#

* Stability is measured by Word Error Rate (WER, ↓).
* "Public-Multilingual-avg" refers to the average performance on the public TTS-Multilingual-Test-Set, covering 20 languages.
* "Inhouse-Multilingual-avg" refers to the average performance on an internal multilingual test set built upon Fleurs, covering 9 languages.
* The performance is tested with the following APIs: ElevenLabs-Multilingual-V2 (9YHcvj6GT2YYXdXww), Gemini-2.5 Pro-Preview-TTS (Achernar), GPT-Audio-2025-08-28 (Alloy) and Minimax-Speech-2.8-HD (English_expressive_narrator) in March 2026.

* Stability is measured by Word Error Rate (WER, ↓) and similarity is measured by Cosine Similarity (SIM, ↑).
* "Public-Multilingual-avg" refers to the average performance on the public TTS-Multilingual-Test-Set, covering 20 languages.
* "Inhouse-Multilingual-avg" refers to the average performance on an internal multilingual test set built upon Fleurs, covering 9 languages.
* Empty cells (-) indicate scores not yet available or not applicable.

Architecture#

Qwen3.5-Omni continues to adopt the Thinker-Talker architecture. The Thinker receives visual and audio signals through the Vision Encoder and AuT, while audio-visual signals are interleaved and encoded with positional information using TMRoPE. The Thinker is responsible for processing omnimodal signals and outputting text, while the Talker receives multimodal inputs and text outputs from the Thinker to perform contextual speech generation. Speech representations are encoded with the RVQ method proposed in Qwen3-Omni, replacing the computationally heavy DiT operations. Thanks to the chunk-wise streaming input design and the streaming Talker design, the entire model supports realtime interaction. Unlike the dual-track Talker input in the previous generation Qwen3-Omni, the Talker adopts ARIA (Adaptive Rate Interleave Alignment) in its input organization to dynamically align text and speech units and then interleave them, thereby avoiding speech instability caused by differences in text and speech token encoding efficiency, such as omissions, misreadings, or unclear pronunciation of numbers.

Qwen3.5-Omni vs Qwen3-Omni#

Demo#

General Audio-Visual Understanding#

With audio-visual input, Qwen3.5-Omni-Plus can follow instructions to generate accurate, fine-grained, structured, and timestamped captions for scenarios such as audio-video analysis, shot breakdown, and content moderation.

Documentary: Complex Scenes + Animals + Sound Effects Analysis - Video Studio#

Storyline#

00:00.000 – 00:02.500
The screen is black, then a deep, low-frequency whoosh swells as the camera drifts in from space toward Earth. The planet’s night-side hemisphere glitters with city lights while the sun crests over its limb, bathing the atmosphere in a brilliant blue halo. A faint orchestral pad begins to build beneath the whoosh.

00:02.500 – 00:36.800
A rapid-fire montage of wildlife and natural scenes unfolds, each lasting roughly one second, accompanied by swelling cinematic strings, brass accents, and assorted animal sound effects that punctuate every cut:
• High-altitude aerial of cloud-wreathed mountains; wind rush layered under the score.
• Lush rainforest canopy shrouded in mist; distant bird calls echo.
• Hummingbird hovers at orange blossoms; high-pitched wing buzz audible.
• Flock of seagulls wheels across a clear sky; sharp cries pierce the music.
• Underwater vortex of schooling fish; muffled aquatic whooshes.
• Surfer rides a turquoise wave; splash and surf hiss.
• Hammerhead shark cruises above sandy seabed; bubbling ambience.
• Brown pelicans glide over muddy water; soft wing beats.
• Hippopotamus surfaces, jaws agape; guttural snort.
• Brown bear wades through a river; water splashes.
• Bald eagle lands on a stump; talon scrape and caw.
• Bison herd trots across snowy grassland; heavy hoof thuds.
• Iguana flicks forked tongue; faint rustle.
• Wildebeest thunder across rolling plains; pounding hooves.
• Elephant calf splashes in a watering hole; trumpeting call.
• Pronghorn antelope sprint through desert scrub; wind rush.
• Cheetah cub pads forward; soft paw taps.
• Red fox pounces into tall grass; muted thump.
• Lioness yawns widely; resonant growl.
• Polar bear roars amid snow; icy wind.
• Tiger snarls; fierce roar reverberates.
Throughout, the orchestra rises toward a triumphant climax.

00:36.800 – 00:41.000
The view returns to Earth from orbit. White, all-caps text fades in over the globe: “LIFE IN THE ANIMAL KINGDOM.” The music resolves into a sustained chord, then gently subsides.

00:41.000 – 00:44.000
Against the rotating Earth, the single word “Royalty” appears in white serif type at left, lingering briefly before dissolving. Ambient synth tones replace the earlier orchestral swell.

00:44.000 – 00:49.000
Screen cuts to black. Large block letters spelling “LION” materialize; each glyph is filled with moving close-ups of lion fur, eyes, and muzzle. A crystalline chime rings out, followed by a deep bass note. As the letters fade, an extreme close-up of a male lion’s face fills the frame. Flies crawl near its nose while it blinks languidly. In the lower-right corner, small white text reads “Panthera leo.” Night insects chirp softly beneath a subdued musical bed.

00:49.000 – 01:04.000
Narration begins in a calm, mid-range male voice with a neutral British accent: “Lions, majestic creatures known for their regal appearance and powerful presence, have long captivated our imagination.” Visuals alternate between the resting male lion on green grass—shaking its mane—and a lioness weaving through dense foliage. Gentle strings and light percussion underscore the narration; cicadas hum in the background.

01:04.000 – 01:24.000
As the narrator explains lions’ adaptability to savannas, grasslands, woodlands, and semi-deserts, footage shows a lioness striding through tall grass at dusk beneath a pink-tinged sky, then a tight shot of another lioness panting lightly. Music grows more rhythmic; distant lion roars blend with the score.

01:24.000 – 01:40.000
While the narrator notes lions’ distribution across sub-Saharan Africa and India’s Gir Forest, the image cuts to a CGI Earth rotating to center on Africa. Dozens of glowing red dots bloom across the continent, marking populations. Orchestral strings surge, then taper.

01:40.000 – 01:54.000
Back on the ground, an extreme close-up captures a male lion’s amber eye blinking slowly. The narrator describes the species’ robust body, broad head, prominent male mane, and tawny coat. Cut to a male lion lying in dry grass, meticulously licking its forepaw; subtle licking sounds mix with soft ambient music.

01:54.000 – 02:11.000
A lioness stands half-hidden in golden savanna grass, scanning the horizon where a distant herd grazes. The narrator remarks that the fur’s coloration provides effective camouflage. Wind rustles through the grass; the score remains gentle and observational.

02:11.000 – 02:32.000
Golden-hour light bathes a male lion with a dark, full mane as he walks purposefully through mixed green and dry brush. The narrator explains that manes vary in color and size, signify maturity, and aid in attracting mates. Music introduces brighter melodic phrases; occasional bird calls are heard.

02:32.000 – 02:40.000
On a reddish dirt track flanked by sparse vegetation, a majestic male lion strides toward camera while several small birds flutter around his paws. Warm sunset hues dominate the palette; the orchestral bed maintains a steady, dignified rhythm.

02:40.000 – 02:51.000
A lioness leads a procession of at least eight playful cubs through lush grass dotted with trees. The narrator states that lions are social animals living in prides of related females and offspring. Cubs scamper, tumble, and glance curiously at the lens. Light, uplifting strings accompany the scene; faint cub mews are audible.

02:51.000 – 03:00.000
Final tableau: two adult male lions recline side by side on sandy ground amid dry shrubs under a pale sky. One gazes ahead; the other rests its chin on its paws, occasionally blinking. The narrator has finished; only soft ambient music and distant insect chirps remain. The image holds, then gently fades to silence and black.

Visible Text#

00:37.000 – 00:40.000
“LIFE IN THE ANIMAL KINGDOM”: white, clean sans-serif capitals; centered horizontally, slightly above mid-frame; superimposed over the orbital Earth; fades in and out smoothly.

00:41.000 – 00:43.000
“Royalty”: white serif word; positioned left-center over the rotating Earth; static during its brief appearance, then fades.

00:44.000 – 00:48.000
“LION”: very large bold sans-serif capitals filling most of the frame against black; each letter contains animated close-ups of lion facial features; no outline; appears via quick fade-in, holds, then dissolves.

00:52.000 – 00:56.000
“Panthera leo”: small white sans-serif text; lower-right corner of an extreme close-up of a male lion’s face; fades in and out without movement.

(No additional textual elements appear outside these intervals.)

Speakers and Transcript#

Speaker profiles:
Narrator – Adult male, neutral British English accent, warm baritone timbre, measured pacing, informative and documentary tone.

00:56.796 – 01:04.716
Speaker: Narrator
State: Calm, authoritative; medium volume over soft ambient music and insect ambience.
Content: “Lions, majestic creatures known for their regal appearance and powerful presence, have long captivated our imagination.”

01:14.316 – 01:23.916
Speaker: Narrator
State: Steady, explanatory; slight emphasis on key terms; orchestral underscore rising gently.
Content: “They are highly adaptable animals and can be found in a variety of habitats, ranging from savannas and grasslands to woodlands and semi-desert areas.”

01:31.724 – 01:38.444
Speaker: Narrator
State: Informative, neutral; music momentarily subdued to foreground speech.
Content: “They are spread across Sub-Saharan Africa, and a small population lives in the Gir Forest of India.”

01:43.404 – 01:53.804
Speaker: Narrator
State: Descriptive, slightly emphatic on physical traits; strings swell behind voice.
Content: “Lions are characterized by their distinctive appearance, including a robust body, a broad head with a prominent mane in males, and a sleek tawny coat.”

02:05.644 – 02:10.604
Speaker: Narrator
State: Matter-of-fact; softer musical bed, light wind noise underneath.
Content: “The coloration of their fur serves as effective camouflage in their natural environment.”

02:18.508 – 02:24.188
Speaker: Narrator
State: Engaged, mildly enthusiastic; brighter musical motif begins.
Content: “Male lions are easily recognized by their impressive manes, which vary in color and size.”

02:27.788 – 02:32.108
Speaker: Narrator
State: Continues seamlessly; slight crescendo in score.
Content: “The mane is a sign of maturity and plays a role in attracting mates.”

02:42.764 – 02:50.444
Speaker: Narrator
State: Concluding, warm; uplifting strings accompany visuals of cubs.
Content: “They are social animals and are often found in groups known as prides, typically consisting of related females and their offspring.”

(There are no other human speakers; all remaining audible elements are music, animal sounds, or environmental ambience.)

Additional Notes on Audio Design#

1. Music: Predominantly orchestral with cinematic scope—strings, brass, and percussion—modulating in intensity to match visual pacing. It begins with a dramatic swell during the montage, recedes to ambient textures under narration, and ends with a gentle, reflective cadence.
2. Sound Effects: Carefully synchronized animal calls (roars, chirps, splashes) punctuate corresponding visuals, enhancing realism without overpowering narration.
3. Mixing: Narration is consistently foregrounded; music ducks subtly whenever speech occurs, ensuring clarity. Environmental sounds are balanced to add depth yet remain secondary.

Thematic and Cultural Context#

The piece functions as a concise nature-documentary vignette introducing lions as emblematic “royalty” of the animal kingdom. By juxtaposing a global montage of diverse species with focused lion imagery and authoritative narration, it situates lions within broader ecological and cultural narratives of majesty, adaptation, and social structure. The use of orbital Earth shots and scientific nomenclature (“Panthera leo”) underscores a modern, educational intent, while the sweeping score evokes awe and reverence typical of contemporary wildlife filmmaking.

Film: Multi-Character, Multi-Shot Audio-Visual Analysis with Complex Sound Effects#

Storyline#

00:00.000 – 00:04.796
A flat, dark-green screen fills the frame. Centered white capital letters present the Motion Picture Association of America’s standard preview-approval notice. Two smaller white website addresses sit along the lower edge. No music is heard yet; the soundtrack is silent, giving the text full attention.

00:04.796 – 00:06.465
The green card cuts to black. A single, deep orchestral hit with metallic overtones blooms, establishing a tense, cinematic atmosphere.

00:06.465 – 00:10.219
Nighttime aerial footage of a sprawling metropolis glides past. Skyscrapers glitter with gold, blue, and red lights; the camera slowly dollies toward one illuminated tower. A gravelly male voice, close-miked and intimate, murmurs, “Come in close.” The orchestral bed swells beneath his words.

00:10.219 – 00:17.768
The scene snaps to the interior of a packed theatre. From a high angle, four performers—three men and one woman—stand on a glossy black circular stage ringed by white neon. A spotlight picks out a man in a dark suit and fedora who tips his hat toward the crowd. The camera cuts to the audience: an elderly Black man in a fedora watches intently. Back on stage, the female performer in a short black dress stands inside a glowing yellow rectangular frame while the fedora-wearing man gestures theatrically. The narrator continues, “Because the more you think you see… the easier it’ll be to fool you.” A sharp percussive sting punctuates the last phrase. A close-up shows a young man with tousled hair flipping a playing card; the Ace of Hearts materialises.

00:17.768 – 00:21.188
A blue lens flare sweeps across black, revealing the silver Summit Entertainment logo with its stylised mountain peak. The orchestral score surges. The view then cuts to a sweeping night shot of the Las Vegas Strip, dominated by the gold-lit Paris Las Vegas Eiffel Tower and neon signs for “BALLY’S,” “MIRAGE,” and “THE LINQ.”

00:21.188 – 00:30.113
Inside the theatre again, the four magicians—now in coordinated dark suits—stride onto the stage. The audience roars. A blue tarp is whipped away to unveil a transparent, steel-framed teleportation booth. A digital clock on the booth’s side reads “0:00.” The lead magician in a pale suit announces, “Ladies and gentlemen, for our final trick, we are going to rob a bank. On the count of three, you will be teleported through space and time to your bank in Paris.” The crowd gasps. He counts, “One, two, three!” A bassy whoosh and crackling energy sound accompany the booth’s blue flash.

00:30.113 – 00:37.955
Cut to Paris: the façade of the opulent “CRÉDIT RÉPUBLICAIN” bank at night. Inside its vault, the four magicians—now in formal evening wear—stand before a massive circular door. Back in the theatre, the female magician addresses the audience: “Everyone in this room was a victim of hard times. Some of you lost your homes, your cars. And so tonight, we’re gonna return some of that money back to you.” A torrent of euro banknotes erupts from the stage, fluttering over cheering spectators who reach up to catch the cash.

00:37.955 – 00:46.255
The magicians bow amid the falling money. The lead man proclaims, “Thank you, everyone. We are the Four Horsemen. Good night!” The audience erupts in applause. The scene shifts to a marble-floored bank interior where stern men in suits, led by an older white-haired gentleman, descend a grand staircase. He declares, “Your bank was the distraction while they set up the real trick. I was a $140 million distraction.” His voice drips with smug satisfaction.

00:46.255 – 00:55.514
A dim bar: the older Black man in a fedora confers with the white-haired banker, exchanging knowing glances. In a bright, sterile vault corridor, a technician in a white jumpsuit wheels carts of cash. The fedora-wearing elder muses, “Who doesn’t love a good magic trick?” A metallic crash and a shout of “FBI!” cut to a modern office where an agent yells, “Hands where I can see ’em!”

00:55.514 – 01:03.397
On the Las Vegas Strip, a suited man with a phone asks incredulously, “Did you say magicians robbed a bank?” A split-screen montage shows each Horseman in separate interrogation rooms, coolly toying with cards or handcuffs. One smirks, “You have what we in the business like to call ‘nothing up your sleeve.’” The camera lingers on his confident grin.

01:03.397 – 01:12.614
The interrogation intensifies. The same magician leans forward, voice low but taunting: “Because if you did, it means that you and the FBI and your friends at Interpol actually believe in magic.” A sudden slam of his cuffed hands on the table startles the interrogator. He concludes, “First rule of magic: always be the smartest guy in the room.” A quick montage flashes: a fedora tips, the Horsemen bow on stage, and the older Black man whispers, “Wanna know how they did it? Say the magic word.”

01:12.614 – 01:24.001
A female agent in a car remarks, “A year ago, these guys were a bunch of street magicians.” Cut to the Horsemen performing atop a skyscraper against a glittering cityscape. She continues, “Now they’re pulling off amazing robberies and not keeping a single cent for themselves.” The older Black man, now in a tuxedo, intones, “You do realize this is a game played out on a global scale.”

01:24.001 – 01:34.469
Scenes intercut rapidly: a white-jumpsuited trio approaches a vault; an armored truck in a parking garage explodes, spilling money; the Horsemen study a holographic blueprint of a vault; the female agent warns, “We are dealing with something far bigger than us.” The white-haired banker growls, “Expose them now and destroy them.”

01:34.469 – 01:45.605
Action escalates: a sleek private jet soars above clouds; a red sports car rockets across the Eiffel Tower’s iron lattice; a man in a dark room brandishes a card marked “THE TOWER / LA MAISON DIEU.” The fedora-wearing elder states, “Vegas was just a start. This trick was designed a long time ago.”

01:45.605 – 02:00.203
A woman with reddish-brown hair pleads, “We’re all here for the same reason.” The elder Black man counters grimly, “We cannot quit now.” A suited gunman levels a pistol in a sparse room. On stage, one Horseman is yanked upward into a blinding spotlight, vanishing. The orchestral score reaches a thunderous peak.

02:00.203 – 02:11.006
The elder Black man’s voice overlays frenetic images: “Whatever is about to follow, whatever this grand trick is… it’s really going to amaze.” A red car bursts through the roof of a graffiti-covered building, showering the night sky with cash. The audience, including the fedora-wearing elder and a blonde woman, stare upward, mouths agape.

02:11.006 – 02:17.554
The narrator delivers the final caution: “Look closely, because the closer you think you are, the less you’ll actually see.” The screen cuts to black, then a brilliant blue lens flare reveals the metallic title “NOW YOU SEE ME,” its letters gleaming with chrome reflections.

02:17.554 – 02:24.603
Against a dark backdrop, bold white text announces “MAY 31,” followed by the Facebook logo and “NowYouSeeMeMovie,” the hashtag “#NowYouSeeMe,” and a small copyright line. The music resolves with a last resonant chord, then silence as the image fades to black.

Visible Text#

00:00.000 – 00:04.796
“THE FOLLOWING PREVIEW HAS BEEN APPROVED FOR” – white uppercase sans-serif, centered on dark-green background
“APPROPRIATE AUDIENCES” – larger, bold white uppercase, centered
“BY THE MOTION PICTURE ASSOCIATION OF AMERICA, INC.” – white uppercase, centered
“www.filmratings.com” – small white lowercase, bottom left
“www.mpaa.org” – small white lowercase, bottom right

00:17.768 – 00:19.500
“SUMMIT ENTERTAINMENT” – silver uppercase serif beneath stylised mountain logo, centre screen
“A LIONSGATE COMPANY” – smaller silver uppercase, centred below main logo

00:19.500 – 00:21.188
“BALLY’S” – large red neon letters on hotel façade, upper left of frame
“MIRAGE” – white neon letters on adjacent building, mid-left
“THE LINQ” – white neon letters on right-side building

00:21.188 – 00:30.113
“0:00” – red seven-segment digits on small black display affixed to teleportation booth, stage right

00:28.000 – 00:30.000
“CRÉDIT RÉPUBLICAIN” – gold serif capitals on stone bank façade, Paris night scene

01:40.000 – 01:41.500
“THE TOWER” – black uppercase at top of tarot-style card
“LA MAISON DIEU” – black uppercase at bottom of same card

02:17.554 – 02:19.000
“NOW YOU SEE ME” – large metallic silver 3-D letters with blue lens-flare glow, centred on black

02:20.000 – 02:22.000
“MAY 31” – large white uppercase, centred
Facebook “f” logo followed by “NowYouSeeMeMovie” – white, centred below date
“#NowYouSeeMe” – white, centred below social line
“© 2013 SUMMIT ENTERTAINMENT, LLC. ALL RIGHTS RESERVED.” – very small white uppercase, bottom centre
Lionsgate stylised “L” logo – small white, bottom right

Speakers and Transcript#

Speaker profiles:
Narrator – older male, deep gravelly American voice, measured, ominous tone
Lead Magician – mid-30s male, clear mid-range American accent, showman confidence, energetic delivery
Female Magician – late-20s female, bright American accent, persuasive, enthusiastic tone
Older Banker – late-60s male, refined American accent, authoritative, smug
Fedora Elder – late-60s Black male, resonant baritone, calm, philosophical
Interrogator – 40s male, firm American accent, forceful, impatient
Street-Magician Interrogated – early-30s male, relaxed American accent, sardonic, playful
Female Agent – 30s female, steady American accent, analytical, concerned

00:08.364 – 00:09.004
Speaker: Narrator
State: low volume, intimate, ominous
Content: “Come in close.”

00:10.364 – 00:17.724
Speaker: Narrator
State: slow, cautionary, gravelly
Content: “Because the more you think you see, the easier it’ll be to fool you.”

00:19.644 – 00:23.004
Speaker: Lead Magician
State: projected, theatrical excitement
Content: “Ladies and gentlemen, for our final trick, we are going to rob a bank.”

00:23.884 – 00:29.484
Speaker: Lead Magician
State: ringing announcement, rising cadence
Content: “On the count of three, you will be teleported through space and time to your bank in Paris.”

00:29.724 – 00:31.244
Speaker: Lead Magician
State: loud, rhythmic countdown
Content: “One, two, three.”

00:31.244 – 00:36.444
Speaker: Female Magician
State: earnest, compassionate
Content: “Everyone in this room was a victim of hard times.”

00:36.604 – 00:40.764
Speaker: Female Magician
State: rallying, hopeful
Content: “Some of you lost your homes, your cars, and so tonight, we’re gonna return some of that money back to you.”

00:44.044 – 00:44.684
Speaker: Lead Magician
State: grateful, upbeat
Content: “Thank you, everyone.”

00:44.844 – 00:46.284
Speaker: Lead Magician
State: triumphant proclamation
Content: “We are the Four Horsemen.”

00:46.284 – 00:47.004
Speaker: Lead Magician
State: cheerful sign-off
Content: “Good night.”

00:48.780 – 00:52.860
Speaker: Older Banker
State: controlled, explanatory
Content: “Your bank was the distraction while they set up the real trick.”

00:53.100 – 00:57.260
Speaker: Older Banker
State: boastful, self-satisfied
Content: “I was a hundred and forty million dollar distraction.”

00:57.260 – 01:00.140
Speaker: Fedora Elder
State: amused, reflective
Content: “Who doesn’t love a good magic trick?”

01:01.100 – 01:02.540
Speaker: Interrogator
State: loud command, tense
Content: “FBI! Hands where I can see ’em.”

01:02.940 – 01:04.140
Speaker: Street-Magician Interrogated
State: sarcastic disbelief
Content: “I don’t think I heard you correctly.”

01:04.220 – 01:05.900
Speaker: Street-Magician Interrogated
State: incredulous, taunting
Content: “Did you say magicians robbed a bank?”

01:06.060 – 01:07.260
Speaker: Street-Magician Interrogated
State: smug, challenging
Content: “You are going to be played.”

01:07.260 – 01:10.540
Speaker: Street-Magician Interrogated
State: lecturing, playful
Content: “You have what we in the business like to call nothing up your sleeve.”

01:10.700 – 01:15.340
Speaker: Street-Magician Interrogated
State: mocking, confident
Content: “Because if you did, it means that you and the FBI and your friends at Interpol actually believe in magic.”

01:18.380 – 01:19.660
Speaker: Street-Magician Interrogated
State: didactic, crisp
Content: “First rule of magic, always be the smartest guy in the room.”

01:23.660 – 01:24.540
Speaker: Fedora Elder
State: teasing, conspiratorial
Content: “Wanna know how they did it?”

01:24.780 – 01:25.740
Speaker: Fedora Elder
State: inviting, playful
Content: “Say the magic word.”

01:26.700 – 01:28.940
Speaker: Female Agent
State: analytical, matter-of-fact
Content: “A year ago, these guys were a bunch of street magicians.”

01:31.180 – 01:34.860
Speaker: Female Agent
State: impressed, wary
Content: “Now they’re pulling off amazing robberies and not keeping a single cent for themselves.”

01:35.980 – 01:40.300
Speaker: Fedora Elder
State: grave, explanatory
Content: “You do realize this is a game played out on a global scale.”

01:40.780 – 01:41.900
Speaker: Fedora Elder
State: ominous, foreboding
Content: “Vegas was just a start.”

01:42.460 – 01:45.260
Speaker: Fedora Elder
State: reflective, portentous
Content: “This trick was designed a long time ago.”

01:46.220 – 01:48.460
Speaker: Female Agent
State: concerned, urgent
Content: “We are dealing with something far bigger than us.”

01:49.100 – 01:50.540
Speaker: Red-haired Woman
State: resolute, earnest
Content: “We’re all here for the same reason.”

01:50.620 – 01:51.980
Speaker: Fedora Elder
State: determined, forceful
Content: “We cannot quit now.”

01:53.660 – 01:56.060
Speaker: Older Banker
State: cold, commanding
Content: “Expose them now and destroy them.”

02:01.036 – 02:09.196
Speaker: Fedora Elder
State: anticipatory, grandiose
Content: “Whatever is about to follow, whatever this grand trick is, is really going to amaze.”

02:11.116 – 02:17.356
Speaker: Narrator
State: slow, cautionary, resonant
Content: “Look closely, because the closer you think you are, the less you’ll actually see.”

Additional Notes on Cinematic Style and Themes#

1. Visual Aesthetics: The trailer juxtaposes sleek, high-contrast stage performances bathed in blue and gold lighting with gritty urban nightscapes and sterile government interiors, reinforcing the duality of spectacle versus clandestine operations.
2. Motifs: Recurrent images of playing cards, vault doors, falling money, and iconic landmarks (Eiffel Tower, Las Vegas Strip) underscore themes of illusion, wealth redistribution, and global stakes.
3. Sound Design: A hybrid orchestral-electronic score drives momentum, punctuated by metallic impacts, whooshes, and crowd roars that synchronize tightly with visual reveals, enhancing the sense of grand illusion.
4. Narrative Arc: The trailer establishes the Four Horsemen as charismatic magician-thieves who expose corruption by stealing from the wealthy and gifting money to the public, while law enforcement and powerful financiers scramble to unmask them, setting up a cat-and-mouse game on an international scale.

Game: Determine Whether Violent Scenes Are Present Based on Audio-Visual Content#

Demo

User

Please provide a detailed description of the video. It must explicitly include two specific analytical sections to identify content that is inappropriate for minors.

### Section 1: Compliance Alert (Summary)
Provide a table summarizing all flagged segments for quick review:
| Time Range | Category | Risk Level | Justification |
| :--- | :--- | :--- | :--- |
| xx:xx - xx:xx | e.g., Violence | High | e.g., Realistic physical assault observed. |

### Section 2: Summary of Safety Findings
Provide a final assessment on whether the video is suitable for minors, citing the most critical timestamps and explaining the overall risk profile. Apply a zero-tolerance policy; if there is ambiguity regarding whether a scene is harmful, err on the side of caution.

---

## Output Format:

```markdown
## Compliance Alert (Summary)

| Time Range | Category | Risk Level | Justification |
| :--- | :--- | :--- | :--- |

## Summary of Safety Findings

<Paragraphs detailing the safety assessment.>

Qwen3.5-Omni-Plus

This video is a compilation of gameplay from Mortal Kombat 11 Ultimate, focusing on the game’s special finishing moves called Fatalities. It begins by showing the main menu, where the player selects the Fatalities mode. They then choose a character from the roster, such as Shang Tsung, Shao Kahn, or The Joker, and enter a one-on-one fight. After winning the match, the player performs a unique, graphic finishing move on their defeated opponent. Each Fatality is a cinematic sequence of extreme violence, showing characters being torn apart, decapitated, frozen and shattered, or disintegrated. The video cycles through several different characters, each demonstrating their own distinct and brutal method of killing their opponent. The presentation is stylized with dark, dramatic lighting and slow-motion effects to highlight the gore. The video concludes after showcasing a series of these Fatalities.

Section 1: Compliance Alert (Summary)#

Section 2: Summary of Safety Findings#

This video is not suitable for minors. The entire video consists of gameplay from Mortal Kombat 11 Ultimate, a game rated for mature audiences due to its extreme content. The video showcases multiple Fatalities, which are graphic and sadistic finishing moves. These scenes depict realistic and brutal violence, including dismemberment, decapitation, evisceration, and characters being set on fire or shattered. The high level of blood and gore, combined with the detailed and cinematic nature of the violence, makes the content inappropriate for anyone under the age of 18.

Short Video: Daily Life Short Video Analysis#

Storyline#

00:00.000 – 00:04.300
The clip opens with a hand-held, wide-angle selfie shot. A Black man in his twenties, sporting short curly hair and a bright magenta-yellow-red striped long-sleeve top, fringed magenta skirt, matching leg warmers, and large fluffy magenta ankle attachments, fills most of the frame. Behind him stretches a sun-bleached, sandy clearing under a blue sky mottled with white clouds. Dozens of onlookers—men, women, and children in casual clothes—form a loose semicircle, many holding up smartphones. Farther back, several drummers stand beside tall, cylindrical drums. The performer shouts excitedly toward the lens, “I’m here in Ivory Coast and I’m about to do the hardest dance in the world. Let’s go!” His voice is loud, breathy, and exuberant. As he finishes, a dense wall of polyrhythmic hand-drumming erupts, accompanied by scattered cheers and whoops from the crowd.

00:04.300 – 00:15.000
The camera swings outward into a medium-wide view that now frames two principal dancers. On the left is the striped-shirt host; on the right stands a masked dancer whose face is painted solid green beneath an ornate headdress topped with red-green-white feathers. This second figure wears a flowing cape and skirt in vertical bands of green, white, and orange, plus striped leg wraps and shaggy brown ankle rattles that jingle with every step. Both men launch into vigorous footwork—rapid stamping, hopping, and spinning—sending puffs of dust into the air. The drum ensemble behind them pounds out layered rhythms while the audience claps and yells encouragement. Around 00:09 a male spectator near the camera calls out, “Why?” in playful disbelief. At roughly 00:12 the masked dancer drops into a low squat, nearly touching the ground, then springs back up, prompting louder applause.

00:15.000 – 00:18.000
The host abruptly pivots the phone back to himself for another close-up. Breathing hard, sweat glistening on his forehead, he gasps, “It is very hard. It’s even hard for me.” His grin mixes exhaustion with exhilaration. Drumbeats continue underneath, slightly muffled by his proximity to the microphone.

00:18.000 – 00:26.000
The viewpoint widens again. The two lead dancers re-engage, circling each other. The masked performer brandishes a short wooden stick, slashing it through the air in time with the drums, while the host mirrors some of the steps but struggles to keep pace, occasionally stumbling and laughing at himself. Dust swirls around their feet; sunlight flashes off metallic ornaments on the costume. The crowd’s energy rises—shouts, whistles, and rhythmic clapping blend with the relentless percussion.

00:26.000 – 00:31.000
In a climactic flourish, the masked dancer spins rapidly, cape flaring like a multicolored wheel. He closes the distance and enfolds the host in a brief celebratory hug. The host throws his head back, mouth open in a joyous shout, eyes squeezed shut. Camera shake intensifies as both men laugh and sway together amid thunderous drumming and roaring approval from spectators.

00:31.000 – 00:32.000
Still locked in the embrace, the host leans toward the lens and bellows, “W’s in the chat! Like and sub!” His tone is triumphant and teasing, aimed at an online audience. Immediately after the final word, the image freezes for a split second and the video cuts to black, ending the clip.

Visible Text#

No visible text, captions, graphics, or on-screen typography appear at any point in the video.

Speakers and Transcript#

Speaker profiles:
Host – Male, mid-20s, Black, African-American English accent; voice loud, energetic, occasionally breathless; primary on-camera performer in striped outfit.
Spectator – Unseen male voice from crowd; casual tone, brief interjection.

00:00.000 – 00:04.240
Speaker: Host
State: Shouting, highly enthusiastic, projecting over ambient noise
Content: “I’m here in Ivory Coast and I’m about to do the hardest dance in the world. Let’s go!”

00:09.000 – 00:10.000
Speaker: Spectator
State: Playful exclamation, raised voice
Content: “Why?”

00:15.200 – 00:17.600
Speaker: Host
State: Breathless, strained yet amused
Content: “It is very hard. It’s even hard for me.”

00:31.000 – 00:32.000
Speaker: Host
State: Triumphant shout, promotional tone
Content: “W’s in the chat! Like and sub!”

Additional Notes on Audio-Visual Style#

• Music: Continuous live West African drumming featuring multiple djembe-style drums producing interlocking high, mid, and low tones; tempo fast and steady, no melodic instruments detected.
• Ambient sound: Frequent crowd cheers, claps, whistles; occasional individual shouts; no discernible wind or traffic noise, indicating an outdoor but relatively sheltered setting.
• Cinematography: Entirely handheld smartphone footage; frequent rapid pans between selfie close-ups and wider shots; slight fisheye distortion suggests a wide-angle lens attachment.
• Lighting: Bright natural daylight with strong overhead sun casting short shadows; colors appear saturated, enhancing the vividness of costumes and surroundings.
• Cultural context: Costumes, masks, and drumming strongly evoke traditional Ivorian ceremonial dance forms such as those associated with the Goli or Zaouli traditions, though the presence of social-media slang (“W’s in the chat,” “Like and sub”) indicates a contemporary, internet-savvy performance intended for online sharing.

Recommended Usage#

Fixed Structured-Description Prompt#

Provide a detailed description of the video.

It should explicitly include three sections: 

1. A structured chronological storyline of **every noticeable audio and visual details**
2. A structured list of all visible text. For each text element, include start timestamp, end timestamp, the exact text content, the appearance characteristics. If no text appears, explicitly state so.
3. A structured speech-to-text transcription, include speaker（Corresponding to the character or voice‑over in Section 1, including their accent and tone）, exact spoken content, start timestamp, end timestamp, and speaking state (prosody, emotion, and style). If no speech appears, explicitly state so.

Aside from these three required sections, you are free to organize any additional content in any way you find helpful. This additional content can include global information about the entire video or localized information about specific moments. You may choose the topic of this extra content freely.

Output Format:

```
## Storyline

<xx:xx.xxx> - <xx:xx.xxx>
<an unstructured long paragraph in natural language describing what happened during this period, blending both audio and video details.>

<xx:xx.xxx> - <xx:xx.xxx>
<an unstructured long paragraph in natural language describing what happened during this period, blending both audio and video details.>

<xx:xx.xxx> - <xx:xx.xxx>
<an unstructured long paragraph in natural language describing what happened during this period, blending both audio and video details.>

...

## Visible Text

<xx:xx.xxx> - <xx:xx.xxx>
“<element>”: <appearance>
“<element>”: <appearance>

<xx:xx.xxx> - <xx:xx.xxx>
“<element>”: <appearance>
“<element>”: <appearance>
“<element>”: <appearance>

<xx:xx.xxx> - <xx:xx.xxx>
“<element>”: <appearance>

...

## Speakers and Transcript

Speaker profiles:
<speaker> - <profile>
<speaker> - <profile>
<speaker> - <profile>
...

<xx:xx.xxx> - <xx:xx.xxx>
Speaker: <speaker>
State: <description>
Content: “<content>”

<xx:xx.xxx> - <xx:xx.xxx>
Speaker: <speaker>
State: <description>
Content: “<content>”

<xx:xx.xxx> - <xx:xx.xxx>
Speaker: <speaker>
State: <description>
Content: “<content>”

...

## <another section>

<paragraphs>

## <another section>

<paragraphs>

...
```

Audio-Visual Vibe Coding#

Snake Game#

Demo

User

Qwen3.5-Omni-Plus

User

Qwen3.5-Omni-Plus

Demo2Code#

Demo

User

Qwen3.5-Omni-Plus

User

Qwen3.5-Omni-Plus

Audio-Visual Conversation#

Travel Planning – Weather/Hotel Lookup, WebSearch/Tool#

Demo

Voice Clone#

Demo

Multi-Turn Dialogue and Intelligent Interruption#

Demo

Voice Style, Emotion, and Volume Control in Conversation#

Demo

General Audio Understanding#

Multi-Language Song Understanding#

Demo

User

Please analyze this multilingual song segment by segment along the timeline, and output the start and end times, complete original lyrics, singer's gender, vocal characteristics, language type, emotional tone, and performance style (solo/chorus) for each segment.

Qwen3.5-Omni-Plus

[00:00.000 – 00:05.440]

Instrumental introduction; no lyrics present. The orchestra establishes a solemn, anticipatory atmosphere with sustained strings and distant percussion. No vocals are heard yet, so the mood is tense and expectant.

[00:05.440 – 00:16.320]

Solo male tenor (clear, bright, slightly nasal timbre) sings in English with a firm, declamatory delivery. Emotional tone is urgent and rallying, evoking defiance and hope. Style is solo, front-and-center over restrained orchestral support.

Lyrics: “Do you hear the people sing? Singing the song of angry men? It is the music of a people who will not be slaves again.”

[00:16.320 – 00:28.080]

Same solo tenor continues in English, voice gaining intensity and forward projection. Emotional tone shifts to passionate resolve. Solo performance remains dominant, with strings swelling beneath.

Lyrics: “When the beating of your heart echoes the beating of the drum, there is a life about to start when tomorrow comes.”

[00:28.080 – 00:33.600]

Baritone male voice enters in French, resonant and weighty, projecting solidarity. Emotional tone is dignified and communal. Solo delivery, supported by low brass undercurrents.

Lyrics: “À la volonté du peuple et à la santé du progrès.”

[00:33.600 – 00:39.280]

Tenor male voice in German, ringing and forceful, sustaining long vowels for emphasis. Tone is fervent and insistent. Solo style, with orchestral punctuation.

Lyrics: “Das ist die Symphonie von Menschen, die nicht länger Sklaven sind.”

[00:39.280 – 00:50.800]

Baritone male voice in Japanese, warm and measured, delivering lines with rhythmic clarity. Emotional tone is determined yet reflective. Solo performance, strings providing gentle momentum.

Lyrics: “新たに熱い命が始まる。明日が来た時、そうさ、明日。”

[00:50.800 – 00:56.720]

Tenor male voice in Hungarian, agile and impassioned, slight vibrato adding urgency. Tone is exhortative. Solo style, light woodwind coloration behind.

Lyrics: “El sem ellik, hogyha kell, kiállsz e értünk, harcunkért.”

[00:56.720 – 01:07.360]

Baritone male voice in Swedish, rich and rounded, projecting calm strength. Emotional tone is resolute and inclusive. Solo delivery, brass subtly reinforcing cadence.

Lyrics: “Från vår barrikad kan man se ett framtidens land. Så kom med oss, låt oss kämpa om du kan.”

[01:07.360 – 01:12.880]

Tenor male voice in Polish, bright and piercing, conveying indignation. Tone is sharp and motivating. Solo performance, percussion accents underline key words.

Lyrics: “Pytasz, słuchasz, śpiewasz lud, co nie chce żyć w niewoli znów.”

[01:12.880 – 01:18.800]

Baritone male voice in Dutch, sonorous and steady, imparting gravitas. Emotional tone is earnest and collective. Solo style, lower strings provide foundation.

Lyrics: “Al die mensen die verdommen om nog langer slaaf te zijn.”

[01:18.800 – 01:29.920]

Return to English solo tenor, now fuller and more expansive, voice soaring above thicker orchestration. Tone is triumphant anticipation. Solo performance leading toward ensemble build-up.

Lyrics: “When the beating of your heart echoes the beating of the drums, there is a life about to start when tomorrow comes.”

[01:29.920 – 01:35.440]

Tenor male voice in German, powerful and ringing, emphasizing finality. Emotional tone is climactic and commanding. Solo delivery, cymbal crashes accent phrases.

Lyrics: “Wenn du kämpfst mit ganzer Kraft, hat bald ein Ende alle Not.”

[01:35.440 – 01:40.480]

English solo tenor resumes, voice urgent and questioning, slight edge of challenge. Tone is provocative, urging action. Solo style, dynamic rise in accompaniment.

Lyrics: “Some will fall and some will live. Will you stand up and take your chance?”

[01:40.480 – 01:45.840]

English solo tenor continues, voice imbued with patriotic fervor, broad phrasing. Emotional tone is inspirational. Solo performance, brass fanfare hints at coming chorus.

Lyrics: “The blood of the martyrs will water the meadows of France.”

[01:45.840 – 01:51.440]

Tenor male voice in Danish, clear and lyrical, gentle vibrato. Tone is hopeful and forward-looking. Solo delivery, harp arpeggios shimmer beneath.

Lyrics: “Kan du høre folkesangen? Det er håp om morgendagen.”

[01:51.440 – 01:57.120]

Tenor male voice in Czech, robust and declamatory, strong consonants. Emotional tone is defiant. Solo style, timpani rolls add tension.

Lyrics: “Nechť ani šíp nespalí víru, kdo zpívá píseň svobody.”

[01:57.120 – 02:02.560]

Tenor male voice in Finnish, resonant and solemn, measured pacing. Tone is steadfast and noble. Solo performance, low strings sustain harmonic bed.

Lyrics: “Jotka kuuntelevat uuslaulun sävelten jälkeen.”

[02:02.560 – 02:08.080]

Tenor male voice in Danish, lyrical and uplifting, smooth legato. Emotional tone is optimistic. Solo delivery, flutes echo melodic fragments.

Lyrics: “Der skal de leve og leve i den nye dag.”

[02:08.080 – 02:13.680]

English solo tenor returns, voice expansive and inviting, slight crescendo toward phrase end. Tone is inclusive and stirring. Solo style, full orchestra begins to swell.

Lyrics: “Will you join in our crusade? Who will be strong and stand with me?”

[02:13.680 – 02:19.280]

Tenor male voice in Danish/Norwegian, bright and encouraging, forward placement. Emotional tone is motivational. Solo performance, snare drum march rhythm emerges.

Lyrics: “Kom dig en fremtid, vær dig selv og fri, som aldrig vil jeg mere.”

[02:19.280 – 02:30.640]

English solo tenor, voice at peak intensity, sustained high notes, rich vibrato. Tone is exultant and prophetic. Solo lead, orchestra builds to forte.

Lyrics: “Do you hear the people sing? Say, do you hear the distant drums? It is the future that we bring when tomorrow comes.”

[02:30.640 – 02:43.520]

Transition from solo to mixed choir; multiple male and female voices blend in homophonic texture. Language shifts among previously heard tongues, creating a polyglot tapestry. Emotional tone is ecstatic unity. Chorus style, full orchestral tutti, cymbals and brass fanfares.

Lyrics: “One day more!”

[02:43.520 – 02:49.120]

Choir sustains “One day more!” in layered harmonies, dynamics climbing. Tone is jubilant and anthemic. Chorus performance, tempo accelerates slightly.

Lyrics: “One day more!”

[02:49.120 – 02:54.640]

Choir repeats refrain with heightened volume; soprano line soars above. Emotional tone is triumphant. Chorus style, percussion drives momentum.

Lyrics: “One day more!”

[02:54.640 – 03:00.240]

Choir continues, voices interweaving in counterpoint, building to climax. Tone is euphoric resolve. Full chorus, brass and timpani reinforce downbeats.

Lyrics: “One day more!”

[03:00.240 – 03:05.760]

Choir reaches fortissimo peak, holding final chord with radiant resonance. Emotional tone is victorious culmination. Chorus performance, cymbal crash and organ pedal point underline grandeur.

Lyrics: “One day more!”

[03:05.760 – 03:41.027]

Extended choral coda; repeated iterations of “One day more!” in overlapping entries, gradually decrescendoing into a luminous final chord. Emotional tone transitions from ecstatic to serene resolution. Full chorus with orchestra tapering to silence, leaving a lingering sense of hope.

Lyrics: “One day more! One day more! One day more! One day more!”

BBC Audio Understanding#

Storyline#

00:00.000 – 00:06.800
Inside a reverberant television-studio control room, two young production assistants—one woman, one man—argue in agitated, fast voices about having waited “over twenty minutes” to go on-air. The woman’s tone is high-pitched and panicked; the man’s is clipped and frustrated. A low, steady HVAC hum underlies the exchange. A brief paper rustle and a single cough from somewhere off-mic punctuate the tension.

00:06.800 – 00:11.500
The argument escalates: the man calls the situation “incredibly unprofessional,” the woman fires back “So is your mom!” and immediately apologizes, admitting she is “very stressed out.” Their voices overlap slightly. A sharp, comic “whoosh” stinger and a burst of canned laughter announce that the feed is about to go live.

00:11.500 – 00:18.600
Still backstage, the woman snaps, “Oh, okay, they’re coming back to us. Get through the whole broadcast in five minutes. Speed it up!” The man protests, “How’s that even possible?” She groans, “I’m gonna throw up,” and, after a metallic clank, he mutters, “No, there’s no time.” A triumphant orchestral news-theme sting swells, blending with audience laughter as the program officially begins.

00:18.600 – 00:25.800
The music segues into the main theme. A resonant male announcer booms, “Live, this is Channel 8 News at Ten with John Harper and Sandra Burbank.” The theme fades. In the studio, Anchor John Harper greets viewers in a polished baritone: “Good evening and welcome to World News at Ten. I’m John Harper.” Anchor Sandra Burbank follows crisply: “And I’m Sandra Burbank.”

00:25.800 – 00:36.300
Sandra introduces the top story about “four and eight in the Middle East,” then tosses to correspondent Philip Castle “live on the scene.” Philip, speaking through a slightly compressed remote line, tersely confirms the tension. Sandra presses, “Any chance for a quick resolution?” Philip answers “No.” Sandra instantly thanks him. Audience laughter erupts, then subsides.

00:36.300 – 00:45.000
John pivots to a tease about “clean energy,” promising more later, and throws to weather. A perky male weathercaster flatly states, “Clouds.” John immediately yanks the broadcast back: “We now go back to our story on clean energy.” Laughter and a quick stinger accent the gag.

00:45.000 – 00:59.900
John introduces “local expert Dr. Simon Gunther.” After a brief “My pleasure,” John rapid-fires questions about solar power and fossil-fuel dependence. The doctor’s hesitant “Well… yeah, actually…” is cut off mid-sentence. John snaps, “Thank you, doctor,” triggering another wave of audience laughter.

00:59.900 – 01:06.100
Sandra delivers a morbid one-liner: “And now for a story about a non-profit animal shelter that caught fire. Everything died. Hmm.” The studio audience roars; a short news sting follows.

01:06.100 – 01:15.100
John introduces Diane Dubanowski’s “exhaustive story on immigration.” A calm female voice begins, “The borders between countries are—” but John cuts in with “Food for thought.” Fresh laughter and a brisk stinger.

01:15.100 – 01:25.000
Sandra presents the “weekly segment on the state of the nation” and asks political analyst Craig Jones about America’s biggest challenge. Craig starts, “Obama is—” only to be silenced by Sandra’s “Thank you, Craig.” The audience responds with its loudest laugh yet, followed by light applause.

01:25.000 – 01:39.400
A field reporter, Joseph Jensen, crackles in via satellite: “The aliens are about to make their first contact.” Sandra, unfazed, replies, “Fascinating, Joseph, thank you.” Laughter and a short stinger.

01:39.400 – 01:48.900
John hands off to “Steve with sports.” Steve, sounding bewildered, utters a single “What?” John thanks him and the audience, invites them to “stay tuned for Fallon,” and reminds them to watch the morning news at six. Sandra adds, “Have a good night, everyone.” The closing theme swells; applause and cheers fill the room.

01:48.900 – 01:54.900
The theme reaches full orchestral grandeur, then fades under enthusiastic clapping. As the music dies, the ambience shifts back to the control room’s drier acoustics.

01:54.900 – 01:59.998
The female production assistant, now calmer but still hurried, tells her colleague, “So that was too fast. You have four more minutes, so just stretch it out.” A final audience chuckle and a faint exhale close the recording.

Speakers and Transcript#

Speaker profiles:
Female PA – Young adult woman, standard American accent; high-pitched, anxious, rapid speech.
Male PA – Young adult man, standard American accent; tense, clipped delivery.
Announcer – Middle-aged man, deep “news voice,” authoritative.
John Harper – Male anchor, mid-40s, resonant baritone, polished broadcast cadence.
Sandra Burbank – Female anchor, early-40s, bright mezzo, precise diction.
Philip Castle – Male correspondent, 30s, slightly nasal, remote-line compression.
Weathercaster – Male, 30s, casual tone, very brief.
Dr. Simon Gunther – Male, 50s, hesitant academic tone.
Diane Dubanowski – Female reporter, 30s, measured, professional.
Craig Jones – Male analyst, 40s, confident, political pundit style.
Joseph Jensen – Male field reporter, 30s, excited, satellite line.
Steve – Male sports anchor, 30s, puzzled tone.
Audience – Mixed-gender studio crowd; laughter, applause, cheers.

00:00.160 – 00:00.960
Speaker: Female PA
State: strained, urgent, rising pitch
Content: “How much longer?”

00:00.960 – 00:02.880
Speaker: Male PA
State: frustrated, quick tempo
Content: “We’ve been waiting to go on for over twenty minutes.”

00:03.120 – 00:04.800
Speaker: Female PA
State: exasperated, fast
Content: “I know, the game went into triple overtime.”

00:04.800 – 00:06.160
Speaker: Female PA
State: annoyed, clipped
Content: “They’re eating up our time slot.”

00:06.240 – 00:07.680
Speaker: Male PA
State: indignant, emphatic
Content: “Well, this is incredibly unprofessional.”

00:07.680 – 00:08.320
Speaker: Female PA
State: mocking, sharp
Content: “So is your mom.”

00:08.320 – 00:08.800
Speaker: Male PA
State: shocked, abrupt
Content: “What?”

00:08.800 – 00:10.400
Speaker: Female PA
State: flustered, apologetic
Content: “I’m sorry, I’m very stressed out.”

00:11.200 – 00:12.640
Speaker: Female PA
State: suddenly focused, brisk
Content: “Oh, okay, they’re coming back to us.”

00:12.800 – 00:14.320
Speaker: Female PA
State: commanding, hurried
Content: “Get through the whole broadcast in five minutes.”

00:14.400 – 00:14.880
Speaker: Female PA
State: urgent, loud
Content: “Speed it up.”

00:14.880 – 00:15.200
Speaker: Male PA
State: incredulous, raised voice
Content: “How is that even possible?”

00:15.200 – 00:16.800
Speaker: Female PA
State: nauseated, groaning
Content: “I’m gonna throw up.”

00:16.800 – 00:17.680
Speaker: Male PA
State: tense, clipped
Content: “No, there’s no time.”

00:18.572 – 00:22.972
Speaker: Announcer
State: booming, ceremonious
Content: “Live, this is Channel 8 News at 10 with John Harper and Sandra Burbank.”

00:23.932 – 00:25.292
Speaker: John Harper
State: warm, formal
Content: “Good evening and welcome to World News at 10.”

00:25.292 – 00:25.772
Speaker: John Harper
State: steady, confident
Content: “I’m John Harper.”

00:25.852 – 00:26.652
Speaker: Sandra Burbank
State: bright, professional
Content: “And I’m Sandra Burbank.”

00:26.652 – 00:29.292
Speaker: Sandra Burbank
State: explanatory, brisk
Content: “Our top story tonight, four and eight in the Middle East has some people up in arms.”

00:29.292 – 00:31.212
Speaker: Sandra Burbank
State: transitional, clear
Content: “We go now to our correspondent Philip Castle, who is live on the scene.”

00:31.212 – 00:34.092
Speaker: Sandra Burbank
State: inquisitive, formal
Content: “Philip, things seem to be rather tense regarding this issue.”

00:34.252 – 00:34.492
Speaker: Philip Castle
State: terse, factual
Content: “Yes.”

00:34.652 – 00:35.772
Speaker: Sandra Burbank
State: probing, quick
Content: “Any chance for a quick resolution?”

00:36.092 – 00:36.332
Speaker: Philip Castle
State: firm, abrupt
Content: “No.”

00:36.492 – 00:36.812
Speaker: Sandra Burbank
State: clipped, polite
Content: “Thank you, Philip.”

00:36.812 – 00:42.092
Speaker: John Harper
State: smooth, announcer-like
Content: “Recent developments in clean energy have some people asking questions, but more on that later in the program.”

00:42.092 – 00:44.252
Speaker: John Harper
State: upbeat, segueing
Content: “Let’s first check in with the weather.”

00:44.492 – 00:44.812
Speaker: Weathercaster
State: flat, deadpan
Content: “Clouds.”

00:45.052 – 00:46.732
Speaker: John Harper
State: brisk, redirecting
Content: “We now go back to our story on clean energy.”

00:47.532 – 00:49.212
Speaker: John Harper
State: cordial, introducing
Content: “We go now to local expert Dr. Simon Gunther.”

00:49.212 – 00:50.092
Speaker: John Harper
State: polite
Content: “Thank you for joining us, doctor.”

00:50.572 – 00:50.892
Speaker: Dr. Simon Gunther
State: courteous
Content: “My pleasure.”

00:50.892 – 00:51.692
Speaker: John Harper
State: eager, rapid
Content: “Let’s get right to it, shall we?”

00:51.692 – 00:54.332
Speaker: John Harper
State: enthusiastic, quick
Content: “Solar power can run cars, homes, even entire cities.”

00:54.332 – 00:54.412
Speaker: John Harper
State: confirming
Content: “Is that correct?”

00:54.732 – 00:56.812
Speaker: John Harper
State: pressing, fast
Content: “And I understand this technology has been around for some time?”

00:57.052 – 00:57.452
Speaker: Dr. Simon Gunther
State: tentative
Content: “Yeah, actually—”

00:57.452 – 00:58.892
Speaker: John Harper
State: challenging, rapid
Content: “So why are we still dependent on gas and oil?”

00:59.292 – 00:59.852
Speaker: John Harper
State: dismissive
Content: “Thank you, doctor.”

01:01.692 – 01:04.332
Speaker: Sandra Burbank
State: somber, matter-of-fact
Content: “And now for a story about a non-profit animal shelter that caught fire.”

01:04.332 – 01:04.972
Speaker: Sandra Burbank
State: flat, grim
Content: “Everything died.”

01:05.292 – 01:05.692
Speaker: Sandra Burbank
State: pensive hum
Content: “Hmm.”

01:06.172 – 01:12.172
Speaker: John Harper
State: formal, introducing
Content: “We now go to a special report by Diane Dubanowski, who has spent the last six months preparing an exhaustive story on immigration.”

01:12.492 – 01:14.252
Speaker: Diane Dubanowski
State: measured, serious
Content: “The borders between countries are—”

01:14.412 – 01:14.972
Speaker: John Harper
State: cutting in, wry
Content: “Food for thought.”

01:17.068 – 01:21.388
Speaker: Sandra Burbank
State: upbeat, introducing
Content: “And now for our weekly segment on the state of the nation.”

01:21.388 – 01:23.788
Speaker: Sandra Burbank
State: cordial
Content: “We have, as always, political analyst Craig Jones.”

01:24.028 – 01:24.108
Speaker: Sandra Burbank
State: inquisitive
Content: “Craig,”

01:24.108 – 01:24.748
Speaker: Sandra Burbank
State: direct
Content: “what is the biggest challenge facing America today?”

01:24.108 – 01:24.908
Speaker: Craig Jones
State: declarative, cut short
Content: “Obama is—”

01:24.908 – 01:25.228
Speaker: Sandra Burbank
State: abrupt, polite
Content: “Thank you, Craig.”

01:32.748 – 01:38.348
Speaker: Joseph Jensen
State: excited, remote-line buzz
Content: “The aliens are about to make their first contact.”

01:38.348 – 01:39.388
Speaker: Sandra Burbank
State: dry, dismissive
Content: “Fascinating, Joseph, thank you.”

01:40.268 – 01:41.708
Speaker: John Harper
State: brisk, segueing
Content: “Let’s go now to Steve with sports.”

01:42.188 – 01:42.508
Speaker: Steve
State: confused
Content: “What?”

01:42.828 – 01:44.988
Speaker: John Harper
State: cheerful, closing
Content: “Thank you, Steve, and thank you all for joining us here tonight.”

01:45.388 – 01:48.268
Speaker: John Harper
State: promotional, upbeat
Content: “Stay tuned for Fallon and be sure to check out the morning news team tomorrow morning at six.”

01:48.348 – 01:49.148
Speaker: Sandra Burbank
State: warm sign-off
Content: “Have a good night, everyone.”

01:54.988 – 01:56.108
Speaker: Female PA
State: relieved, conversational
Content: “So that was too fast.”

01:56.748 – 01:59.948
Speaker: Female PA
State: instructive, calming
Content: “You have four more minutes, so just stretch it out.”

Global Acoustic & Production Notes#

1. Studio acoustics shift noticeably between the dry control-room segments and the reverberant main studio, underscoring the narrative’s backstage-versus-on-air structure.
2. Non-speech elements—orchestral news themes, comic stingers, whooshes, paper rustles, and audience reactions—are tightly synchronized with dialogue to heighten comedic timing.
3. The entire piece is a parody of live news broadcasts, lampooning time constraints, sensationalism, and abrupt cut-offs. Rapid pacing, overlapping lines, and sudden topic shifts create a satirical, farcical rhythm.
4. Audience responses (laughter, applause, cheers) function as a rhythmic punctuation, often triggered immediately after punchlines or abrupt dismissals, reinforcing the sketch-comedy format.
5. Dynamic range is moderate: dialogue sits prominently above a consistent low-level HVAC hum; musical stings and theme swell briefly to near-equal loudness before ducking under speech.
6. No music beds persist beneath dialogue except during the opening and closing themes; otherwise, silence or faint room tone separates segments, allowing jokes to land cleanly.

Thematic & Cultural Context#

The audio parodies American local-news conventions: hyperbritical anchors, token experts, trivial weather updates, and sensational headlines. References to “Fallon” suggest a late-night talk-show lead-in, situating the piece within U.S. broadcast culture. The absurdity of compressing an entire newscast into minutes, coupled with cut-off interviews and darkly comic one-liners (e.g., “Everything died”), satirizes the rush for brevity and the superficiality of modern news cycles.

Lyric Subtitle Generation#

Demo

User

Please identify the lyrics of this song and provide the timestamps line by line. The format example is as follows:
[00:00:15,020 --> 00:00:28,085] When you walk through a storm, hold your head up high.
[00:00:28,085 --> 00:00:40,200] And don't be afraid of the dark.
[00:00:40,200 --> 00:01:02:580] At the end of the storm, there's a golden sky.
...

Do not output any other content except for the above mentioned.

Qwen3.5-Omni-Plus

[00:00:20,686 –> 00:00:24,996] Clink of crystal, smoke in the air

[00:00:24,996 –> 00:00:29,506] Smiles like porcelain, nobody’s there

[00:00:29,506 –> 00:00:34,266] Tie knotted tight, shoes polished black

[00:00:34,266 –> 00:00:38,436] Laugh on cue, never look back

[00:00:38,436 –> 00:00:43,266] Champagne flutes rise like hollow toasts

[00:00:43,266 –> 00:00:47,796] To dreams we sold for designer ghosts

[00:00:47,796 –> 00:00:52,906] The band plays sweet, but I hear it wrong

[00:00:52,906 –> 00:00:56,806] A syncopated sense of gone

[00:00:56,806 –> 00:01:01,436] I’m dancing offbeat in a perfect show

[00:01:01,436 –> 00:01:06,226] Smiling wide while the silence grows

[00:01:06,226 –> 00:01:11,126] They call it charm, I call it strange

[00:01:11,126 –> 00:01:15,066] This masquerade where nothing changes

[00:01:15,066 –> 00:01:23,106] Yeah, I play along, but I don’t belong

[00:01:23,106 –> 00:01:28,966] Her perfume lingers, words like lace

[00:01:28,966 –> 00:01:33,066] But every glance hides an empty space

[00:01:33,066 –> 00:01:37,166] We quote philosophers over ice

[00:01:37,166 –> 00:01:42,086] While truth dissolves in cocktail vice

[00:01:42,086 –> 00:01:47,106] The sax hums low, but I feel it scream

[00:01:47,106 –> 00:01:51,586] A dissonant thread in this velvet dream

[00:01:51,586 –> 00:01:56,266] I’m dancing offbeat in a perfect show

[00:01:56,266 –> 00:02:01,066] Smiling wide while the silence grows

[00:02:01,066 –> 00:02:05,926] They call it charm, I call it strange

[00:02:05,926 –> 00:02:09,866] This masquerade where nothing changes

[00:02:09,866 –> 00:02:18,866] Yeah, I play along, but I don’t belong

[00:02:18,866 –> 00:02:29,986] What if I drop the glass, let it shatter on marble lights

[00:02:29,986 –> 00:02:46,666] Would they notice the crack in me, or just pour another round and sigh

[00:02:46,666 –> 00:03:00,546] Lights still glitter, music sways

[00:03:00,546 –> 00:03:05,146] I toast the void behind my face

[00:03:05,146 –> 00:03:10,466] Another night, another role

[00:03:10,466 –> 00:03:15,006] Jazz hands hiding a broken soul

[00:03:15,006 –> 00:03:23,000] (Instrumental fade out)

Recommended Usage#

Fixed Structured-Description Prompt#

Provide a detailed description of the audio.

It should explicitly include two sections: 

1. A structured chronological storyline of **every noticeable audio details**
2. A structured speech-to-text transcription, include speaker（Corresponding to the character or voice‑over in Section 1, including their accent and tone）, exact spoken content, start timestamp, end timestamp, and speaking state (prosody, emotion, and style). If no speech appears, explicitly state so.

Aside from these two required components, you are free to organize any additional content in any way you find helpful. This additional content can include global information about the entire audio or localized information about specific moments. You may choose the topic of this extra content freely.

Output Format:

```
## Storyline

<xx:xx.xxx> - <xx:xx.xxx>
<an unstructured long paragraph in natural language describing what happened during this period, blending both audio details.>

<xx:xx.xxx> - <xx:xx.xxx>
<an unstructured long paragraph in natural language describing what happened during this period, blending both audio details.>

<xx:xx.xxx> - <xx:xx.xxx>
<an unstructured long paragraph in natural language describing what happened during this period, blending both audio details.>

...

...

## Speakers and Transcript

Speaker profiles:
<speaker> - <profile>
<speaker> - <profile>
<speaker> - <profile>
...

<xx:xx.xxx> - <xx:xx.xxx>
Speaker: <speaker>
State: <description>
Content: “<content>”

<xx:xx.xxx> - <xx:xx.xxx>
Speaker: <speaker>
State: <description>
Content: “<content>”

<xx:xx.xxx> - <xx:xx.xxx>
Speaker: <speaker>
State: <description>
Content: “<content>”

...

## <another section>

<paragraphs>

## <another section>

<paragraphs>

...
```

Play with Qwen3.5-Omni#

Chat with Qwen3.5-Omni#

Feel free to use Qwen3.5 on Qwen Chat.

ModelStudio#

Users can also access our flagship model, Qwen3.5-Plus-Omni, through Alibaba Cloud Model Studio. Two invocation modes are currently supported: Offline Invocation and Realtime Invocation.

To enable web search, simply pass the following parameter:

• enable_searchEnables web search functionality

Offline Invocation#

Example code is provided below:

# Preparation before running:
# Run the following command to install third-party dependencies
# pip install numpy soundfile openai

import os
import base64
import soundfile as sf
import numpy as np
from openai import OpenAI

client = OpenAI(
    api_key=os.getenv("DASHSCOPE_API_KEY"),  # Confirm the environment variable is set
    base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
)

# Audio and video analysis
completion = client.chat.completions.create(
    model="qwen3.5-omni-plus",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "video_url",
                    "video_url": {
                        "url": "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241115/cqqkru/1.mp4"
                    },
                },
                {"type": "text", "text": "What is the content of the video?"},
            ],
        },
    ],
    # Set the output modality. In audio/video analysis scenarios, it is recommended to return text results directly
    modalities=["text"],
    # stream must be set to True, otherwise an error will occur
    stream=True,
    stream_options={"include_usage": True},
)

print("Model response:")
for chunk in completion:
    # Process the text part
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

# Web search
try:
    completion = client.chat.completions.create(
        model="qwen3.5-omni-plus",
        messages=[{
            "role": "user", 
            "content": "Please check today's date and day of the week, and tell me what important holidays there are today"
        }],
        stream=True,
        stream_options={"include_usage": True},
        # Enable web search
        extra_body={
            "enable_search": True,
            "search_options": {
                # Web search strategy, only supports configuring as agent
                "search_strategy": "agent"
            }
        }
    )
    
    print("\n\n\n\nWeb search model response (including real-time information):")
    for chunk in completion:
        if chunk.choices and chunk.choices[0].delta.content:
            print(chunk.choices[0].delta.content, end="")
    print()
    
except Exception as e:
    print(f"Request failed: {e}")

Realtime Invocation#

Example code is provided below:

# Dependencies: dashscope >= 1.23.9, pyaudio
import os
import base64
import time
import pyaudio
from dashscope.audio.qwen_omni import MultiModality, AudioFormat,OmniRealtimeCallback,OmniRealtimeConversation
import dashscope

# Configuration: endpoint, API key, voice, model, and system instructions
# Specify the region: set to cn for mainland China (Beijing), or intl for international (Singapore)
region = 'cn'
base_domain = 'dashscope.aliyuncs.com' if region == 'cn' else 'dashscope-intl.aliyuncs.com'
url = f'wss://{base_domain}/api-ws/v1/realtime'
# Configure the API key. If no environment variable is set, replace the line below with: dashscope.api_key = "sk-xxx"
dashscope.api_key = os.getenv('DASHSCOPE_API_KEY')
# Specify the voice
voice = 'Tina'
# Specify the model
model = 'qwen3.5-omni-plus-realtime'
# Specify the system instructions
instructions = "You are a personal assistant named Xiaoyun. Please answer the user's questions in a humorous and witty way."
class SimpleCallback(OmniRealtimeCallback):
    def __init__(self, pya):
        self.pya = pya
        self.out = None
    def on_open(self):
        # Initialize audio output stream
        self.out = self.pya.open(
            format=pyaudio.paInt16,
            channels=1,
            rate=24000,
            output=True
        )
    def on_event(self, response):
        if response['type'] == 'response.audio.delta':
            # Play audio
            self.out.write(base64.b64decode(response['delta']))
        elif response['type'] == 'conversation.item.input_audio_transcription.completed':
            # Print transcription text
            print(f"[User] {response['transcript']}")
        elif response['type'] == 'response.audio_transcript.done':
            # Print assistant response text
            print(f"[LLM] {response['transcript']}")

# 1. Initialize audio device
pya = pyaudio.PyAudio()
# 2. Create callback and conversation session
callback = SimpleCallback(pya)
conv = OmniRealtimeConversation(model=model, callback=callback, url=url)
# 3. Connect and configure the session
conv.connect()
conv.update_session(output_modalities=[MultiModality.AUDIO, MultiModality.TEXT], voice=voice, instructions=instructions, enable_search=True)
# 4. Initialize audio input stream
mic = pya.open(format=pyaudio.paInt16, channels=1, rate=16000, input=True)
# 5. Main loop for processing audio input
print("Conversation started. Speak into the microphone (Ctrl+C to exit)...")
try:
    while True:
        audio_data = mic.read(3200, exception_on_overflow=False)
        conv.append_audio(base64.b64encode(audio_data).decode())
        time.sleep(0.01)
except KeyboardInterrupt:
    # Clean up resources
    conv.close()
    mic.close()
    callback.out.close()
    pya.terminate()
    print("\nConversation ended")

Qwen3.5-Omni Voice List#

Chinese and English Custom Voice (5 Speakers)

Chinese and English Scenario Voices (19 Speakers)

Chinese Dialect Voices (8 Speakers)

Multilingual Voice (23 Speakers)

Citation#

Feel free to cite the following article if you find Qwen3.5-Omni helpful:

@misc{qwen35omniblog,
    title = {Qwen3.5-Omni: Scaling Up, Toward Native Omni-Modal AGI},
    url = {https://qwen.ai/blog?id=qwen3.5-omni},
    author = {Qwen Team},
    month = {March},
    year = {2026}
}