VibeVoice Speech Recognition Promo
By @HaileyAILabs. Made with Claude Opus 5.5. Watch the original post.
Prompt
VibeVoice | Making a 40-second, high-quality creative hype video Using the storyboard I uploaded as a reference, make a 40-second, high-quality creative hype video for Microsoft's open-source project VibeVoice. Project URL: https://github.com/microsoft/VibeVoice 1. Creative goals What I want is not an ordinary tech product introduction, and not a GitHub README turned into a PowerPoint animation. It should be a product hype video with outstanding visual taste, creative camera work, smooth animation and a strong musical rhythm, one that also lets viewers intuitively understand VibeVoice's core capabilities. The overall look and feel should be on par with brand promo work from a professional Motion Design studio. You have full creative freedom. There are no restrictions on technical approach, animation style, asset sources or production tools. You can use Three.js, Remotion, Blender, Canvas, SVG, WebGL, GSAP, FFmpeg, AI image and video generation, and any other suitable open-source projects or tools. You can mix 2D, 3D, motion graphics, character animation, kinetic typography, audio visualization and real product demos. It doesn't have to be pure code, and it doesn't have to strictly replicate the storyboard. The storyboard is there to convey the creative direction; it is not a template that limits the final visuals. Based on how things actually turn out in production, proactively improve the composition, camera moves, scenes, animation and transitions. In the end, all that matters is the quality of the finished video. 2. Video specs Duration: 40 seconds Aspect ratio: 16:9 Resolution: 1920×1080 or higher Frame rate: 60fps preferred Language: English Style: a premium, modern, design-driven creative tech animation Audio: rhythmic background music, necessary sound effects and real speech demos Final format: MP4 Avoid a cheap neon tech look, PowerPoint-style transitions, meaningless particle effects and piles of overly complex UI. The visuals need strong, memorable moments; don't rely on pretty UI alone. 3. The full 40-second story Act 1 | Voices out of control (0–6s) A 60-minute multi-speaker recording starts playing. Different people's voices blend together, forming complex, interwoven waveforms. As the music speeds up, the voices gradually spiral out of control and the frame fills with chaos. Suddenly, all the sound stops. On screen: "An hour-long multi-speaker recording. Who said what?" Requirement: grab attention fast in the first 3 seconds, and make viewers curious about what changes next. Act 2 | Voices get identities (6–13s) VibeVoice enters. The chaotic voices begin to separate, and the identities of the four speakers gradually become clear. Each person has their own color, waveform and visual marker. This can be shown with people, abstract characters, animated avatars or other, more creative approaches. The four voices gradually go from chaos to order, ending up as clear speaker labels. Caption: "First, figure out who's speaking." The point is to let viewers directly see speaker identification at work, not just four lines in different colors. Act 3 | Timeline magic (13–23s) The camera goes inside the world of sound. A 60-minute timeline gradually unfolds. Speech segments from the different speakers are labeled automatically, generating the matching text and timestamps. Whenever someone speaks, their voice track and text appear at the same time. The camera can weave between sound, people, text and the timeline, creating continuous, rhythmic visual changes. Captions: "Who said it, when they said it, what they said." "All sorted out in one go." This act is the core climax of the whole video and needs the most powerful visuals. Act 4 | Chaos becomes order (23–33s) The camera pulls back, and the complex world of sound we just saw gradually organizes itself into a clear, orderly speech transcript interface. What the four speakers said is laid out in chronological order. Viewers can clearly see how the speakers, timestamps and recognized text correspond. Through dynamic camera work and an interactive demo, help viewers understand the real value of structured recognition for long audio. Caption: "Even 60 minutes of audio can be transcribed into structured text." Note: 60 minutes refers to the officially supported audio length; it doesn't mean recognition finishes instantly. Act 5 | Twist ending (33–40s) All the audio tracks, speaker markers and text gradually converge. The camera pulls back, and the complex world of sound from earlier finally turns into a clean VibeVoice logo. Appears: VibeVoice Microsoft · Open Source http://github.com/microsoft/VibeVoice Just when viewers think the video is over, a tiny waveform suddenly starts bouncing again. On screen: "Wait, I've got one more line!" The video ends with a playful sound effect. The ending has to be fun, not just an ordinary logo animation. 4. Visual and animation requirements You can redesign every scene, but the whole piece must share one unified visual language. Focus on: Visual creativity: wherever possible, express the abstract tech through fun visual metaphors, and make sound the true star of the visuals. Shot design: shots should connect to one another; avoid making every act a separate, static frame. Motion quality: movement has natural acceleration and deceleration, rhythm, a sense of weight, and layering. Composition: establish a clear visual focal point, and control negative space and information density. Kinetic typography: text must be clear and easy to read, and also be part of the visual design. Audio-visual sync: sound changes, captions, character reactions and music cues line up precisely. Emotional arc: from chaos and curiosity to surprise and clarity, finally wrapping up with humor. You may adjust shot durations, animation structure, transitions and visual treatment to improve the final video. But don't change the core product capabilities, and don't add unverified features. 5. Music and sound design Music is an important part of this video. Design a complete music and sound-effects structure that follows the story. Build tension and chaos at the start; let the rhythm gradually become clearer as the speakers separate; hit the most powerful section when the timeline unfolds; wrap up naturally at the end and leave a fun sound twist. You can use original electronic music, funk, a rhythm-driven score or any other style that suits the visuals. Prefer real VibeVoice demo audio. Voice-over isn't required. If the music, character voices, captions and animation can tell the story clearly, don't add unnecessary narration. Don't add voice-over just for the sake of it. 6. Product accuracy Before you start, read the official VibeVoice repo and related docs to make sure the product information is accurate. This video focuses on showcasing VibeVoice-ASR's: Long-form speech recognition Multi-speaker identification Timestamp generation Structured text output Note: distinguishing speakers is not the same as actually separating a mixed recording into multiple independent audio files. If you can't actually run VibeVoice, you can showcase it using the official demo results, and clearly distinguish real output from concept animation. Do not fake model performance, recognition accuracy or processing speed. 7. Production and quality improvement Decide for yourself on the best combination of tech and production workflow; you don't need to stick to any one framework. You can look for high-quality open-source animation libraries, visual references, ready-made components or production tools to help. But don't sacrifice aesthetics just to pile on tech. Follow these principles during production: First, define the art style, camera language and visual rhythm for the whole video. Complete every scene in that unified style. Make sure the main visuals, captions, sound and camera movement work together. Actually render and check the key frames and video clips. Proactively fix problems such as inconsistent characters, stiff motion, abrupt shots, monotonous visuals, caption errors and audio-video sync issues. If an approach doesn't work well in practice, switch to a different implementation; don't stick with the wrong direction. Prioritize the overall polish of the finished piece over showy effects in individual shots. You may revise and redesign as many times as needed until it becomes a complete, unified piece that people will want to share. 8. Final deliverables Please complete and deliver: The complete 40-second MP4 video Editable project source files The assets and audio used Any necessary render or export instructions If the current environment doesn't support a technology or tool, choose a workable alternative on your own; don't pretend you've completed steps you haven't actually carried out. Don't output only a production plan, code or a static storyboard. Final goal: viewers are first drawn in by the creativity and visuals, then genuinely understand what VibeVoice can do, and want to go to GitHub to learn more or try the project. Start production right away and complete all the work you can on your own.