ByteDance's first audio-visual joint generation model with native synchronized audio, phoneme-level lip sync across 8+ languages, and cinematic camera control.
Seedance 1.5 Pro: Native Audio-Visual Generation & Lip Sync
Developer: ByteDance Seed Lab · Released: December 16, 2025 · Technical paper: arXiv:2512.13507 · Official page: Seedance 1.5 Pro · Model ID: seedance-v1.5-pro
Seedance 1.5 Pro is ByteDance's first audio-visual joint generation model, sitting between Seedance 1.0 and 2.0 in the family. For a complete overview of all versions and which fits your workflow, see the Seedance family guide.
What Is Seedance 1.5 Pro?
Seedance 1.5 Pro is ByteDance's first audio-visual joint generation model, released on December 16, 2025. It is the first model in the Seedance family to generate synchronized video and audio in a single unified pass, rather than treating audio as a post-generation addition.
Where Seedance 1.0 focused on raising the "performance floor" by improving motion stability and structural coherence, Seedance 1.5 Pro aims higher: it raises the "performance ceiling" with stronger visual impact, cinematic camera control, and native audio-visual synchronization that includes dialogue, sound effects, ambient sound, and background music.
The technical paper accompanying the release, "Seedance 1.5 pro: A Native Audio-Visual Joint Generation Foundation Model," formally documents the model's dual-branch Diffusion Transformer architecture, which processes audio and video simultaneously in a shared latent space. This architectural shift ensures that audio and visuals are synchronized from generation, not stitched together afterward, a distinction that matters for lip sync accuracy, foley coherence, and audio-driven pacing.
ByteDance describes Seedance 1.5 Pro as "a robust engine for professional-grade content creation," positioning it as the model for workflows where native audio, multilingual lip sync, and elevated visual quality are non-negotiable requirements.
The Core Innovation: Native Audio-Visual Joint Generation
The defining feature of Seedance 1.5 Pro is its dual-branch architecture. Unlike earlier models that generate video first and add audio as a separate pass, Seedance 1.5 Pro generates both simultaneously in the same latent space through a cross-modal joint module.
In practical terms, this means:
Lip sync is phoneme-accurate, not word-approximate. When a character speaks, their mouth shapes align with individual phonemes—the smallest units of sound—rather than approximate syllable boundaries. The result reads as a performance, not as an animation track applied to a face.
Foley and ambient sound respond to visual motion. A character's footsteps on gravel, a door closing, a glass being set on a table—these sound effects generate in sync with the visual action because the model reasons about both simultaneously. The audio isn't guessing what happened in the video; it's being generated as part of the same decision process.
Music and pacing align naturally. When a reference audio track is provided, the model adjusts visual rhythm, cut timing, and motion tempo to match the beat. The synchronization isn't post-processed; it's native to the generation architecture.
Multilingual and dialect support. Seedance 1.5 Pro supports phoneme-level lip sync across Mandarin Chinese, English, Japanese, Korean, Spanish, Indonesian, and regional Chinese dialects including Sichuanese and Cantonese. This language coverage, particularly the dialect support, remains unique among production-scale video models as of mid-2026.
The technical paper describes this as "native, joint audio-video generation," a system where the audio generation and video generation are not independent processes synchronized after the fact, but a unified denoising process that treats both as correlated expressions of the same creative decision.
Architecture and Technical Foundation
Seedance 1.5 Pro uses a dual-branch Diffusion Transformer architecture with a specialized cross-modal joint module. The model processes video and audio in parallel branches that share information through the cross-modal module, allowing the two modalities to influence each other during generation rather than being composed afterward.
Key Technical Components
Dual-branch Diffusion Transformer: The video branch and audio branch denoise in parallel within a shared latent space. This allows the model to reason about audio-visual synchronization, such as lip sync and foley timing, as part of the generation process rather than as a post-generation alignment task.
Cross-modal joint module: This component passes information between the video and audio branches during denoising, ensuring that visual motion and audio events remain synchronized at the frame level. A character speaking, a door closing, or a footstep generates audio and visual motion that are temporally aligned from the start.
Multi-stage data pipeline: The model is trained on a curated dataset with precise audio-visual annotations, enabling it to learn correlations between specific visual events (mouth shapes, object interactions, environmental conditions) and their corresponding audio signatures (phonemes, impact sounds, ambient characteristics).
Post-training optimization: After pre-training, the model undergoes supervised fine-tuning on high-quality audio-visual datasets, followed by reinforcement learning from human feedback (RLHF) with multi-dimensional reward models that assess lip sync accuracy, audio-visual temporal alignment, dialogue clarity, and overall generation quality independently.
Inference acceleration: Through distillation and system-level optimizations, Seedance 1.5 Pro achieves over 10× inference speedup compared to baseline audio-visual generation approaches, making it practical for iterative creative workflows despite the increased computational complexity of joint generation.
Generation Capabilities
Text-to-Video with Audio (T2V)
Seedance 1.5 Pro generates video and audio from text prompts. The prompt can describe both visual elements (scene, action, camera, lighting) and audio elements (dialogue, sound effects, ambient sound, music tone).
Recommended prompt structure: Scene description → Character and action → Dialogue or audio direction → Camera specification → Lighting and atmosphere
Example: "A quiet coffee shop in the morning, warm natural light from large windows. A woman sits at a corner table reading a book. Soft ambient chatter and the hiss of an espresso machine in the background. Steady medium shot, 50mm lens, shallow depth of field."
For dialogue-driven content:
"A man in his forties standing on a busy street corner, looking at his phone. He speaks in English: 'I'll be there in ten minutes.' Traffic noise and pedestrian chatter in the background. Handheld medium shot, slight camera drift."
The model generates the character speaking the line with phoneme-accurate lip sync, synchronized ambient street noise, and the specified camera movement in a single pass.
Image-to-Video with Audio (I2V)
Seedance 1.5 Pro I2V uses the source image as the first frame anchor and generates motion and synchronized audio from that starting point. This mode is particularly effective for:
- Character dialogue from portrait photography: Animate a still portrait with synchronized speech and facial expressions
- Product scenes with ambient sound: Add motion and environmental audio to product photography
- Architectural walkthroughs with spatial audio: Generate camera movement through a space with appropriate acoustic characteristics
The first-frame lock ensures visual identity is preserved exactly while audio is generated to match the visual context of the scene.
Reference Audio-to-Video
When an audio reference file is provided, Seedance 1.5 Pro adjusts visual pacing, motion rhythm, and shot timing to match the audio's tempo and beat structure. This is particularly useful for:
- Music video generation: Visual content that moves in sync with a provided music track
- Dialogue animation: Matching character animation to pre-recorded voice performance
- Audio-driven pacing: Generating visual sequences where the timing is dictated by an existing audio edit
The model treats the reference audio as a strong conditioning signal, ensuring that visual motion and cuts align with the audio's structure rather than being independently timed.
Output Specifications
| Specification | Details |
|---|---|
| Clip duration | 2–12 seconds per generation |
| Native resolution | 1080p |
| Aspect ratios | 16:9 · 9:16 · 4:3 · 1:1 |
| Frame rate | 24 FPS |
| Audio output | Dual-channel stereo, native |
| Audio components | Dialogue · SFX · Ambient · BGM |
| Lip sync languages | Mandarin · English · Japanese · Korean · Spanish · Indonesian |
| Dialect support | Sichuanese · Cantonese |
| Lip sync method | Phoneme-level (not word-level) |
| Generation modes | Text-to-Video · Image-to-Video · Reference Audio-to-Video |
| Model ID | seedance-v1.5-pro |
| Architecture | Dual-branch Diffusion Transformer with Cross-modal Joint Module |
Source: arXiv:2512.13507 · Official Seedance 1.5 Pro page
What "Native Audio" Actually Means
The distinction between native audio generation and post-processed audio addition is often understated but matters significantly for output quality.
Post-processed audio models generate silent video first, then add audio as a separate layer using a secondary model or synthesis pipeline. The audio is synchronized to the video through alignment algorithms, but it's not informed by the video's generative process. Lip sync is achieved by matching phoneme timings to mouth positions after the fact. Foley and ambient sound are added based on what the video appears to contain, not what the model intended to generate.
Native audio generation in Seedance 1.5 Pro means the audio and video are part of the same generation process. The model doesn't generate a silent video and then guess what sounds should accompany it. It generates audio and video together, reasoning about both simultaneously. A character's mouth moves in sync with their speech because the speech generation and mouth movement generation are causally linked during denoising, not matched afterward.
This architectural difference produces:
- Tighter lip sync with no visible drift
- Foley that responds to visual motion rather than approximating it
- Ambient sound that matches the acoustic properties of the scene
- Music or dialogue pacing that drives visual rhythm naturally
The difference is most visible in dialogue scenes, where post-processed lip sync often shows small timing errors or mouth shapes that don't quite match the phoneme, and in scenes with precise foley, where the sound of an object hitting a surface needs to align with the exact frame of contact.
Performance and Use Case Positioning
Seedance 1.5 Pro was not benchmarked on the Artificial Analysis leaderboard at launch, so direct Elo comparisons with Seedance 1.0 and 2.0 are not available. ByteDance's positioning of the model emphasizes its role in raising the "performance ceiling" rather than competing for the top leaderboard spot.
From the technical paper and independent testing, Seedance 1.5 Pro shows particularly strong performance on:
- Phoneme-level lip sync accuracy in Mandarin and English (highest scores in documented comparisons)
- Audio-visual temporal alignment across dialogue, foley, and ambient sound categories
- Multilingual and dialect lip sync support (broader language coverage than competing models)
- Cinematic camera control with dynamic movement responding to narrative pacing
- Reference audio-driven visual pacing where the model adjusts visual rhythm to match a provided audio track
The model's practical value is most apparent in workflows where audio and video need to be generated together rather than assembled in post, particularly for multilingual content, dialogue-driven scenes, and audio-synced visual pacing.
When Seedance 1.5 Pro Is the Right Choice
Seedance 1.5 Pro serves specific production needs that earlier and later models do not address as directly:
Multilingual localized content. If your workflow requires generating the same visual content with dialogue in multiple languages, Seedance 1.5 Pro's phoneme-level lip sync across 8+ languages and regional dialects removes the need for separate dubbing pipelines. Generate once per language with native lip sync rather than animating mouth regions in post.
Dialogue-driven narrative. For short films, brand stories, explainer videos, or any content where character speech is central to the narrative, Seedance 1.5 Pro's tight lip sync and natural audio-visual pacing make it the most direct path from script to synchronized output.
Audio-driven visual pacing. When the visual sequence needs to move in sync with a specific music track, voiceover performance, or audio edit, Seedance 1.5 Pro's reference audio-to-video mode lets the audio drive the visual rhythm rather than forcing both to be manually synchronized in post.
Elevated visual ceiling without multimodal reference complexity. If your project needs stronger visual impact and cinematic camera control than Seedance 1.0 but doesn't require the multimodal reference input surface of Seedance 2.0 (multiple images, video clips, and audio files as inputs), Seedance 1.5 Pro sits in that gap.
How Seedance 1.5 Pro Compares to Other Versions
vs. Seedance 1.0: Seedance 1.0 offers faster inference and silent generation at 1080p with strong motion stability. Seedance 1.5 Pro adds native audio, phoneme-level lip sync, elevated visual quality, and cinematic camera control. Use 1.0 when audio isn't required and speed matters; use 1.5 Pro when audio and video must be generated together.
vs. Seedance 2.0: Seedance 2.0 offers unified multimodal architecture with support for up to 9 reference images, 3 video clips, and 3 audio files as inputs. Seedance 1.5 Pro offers tighter focus on audio-visual joint generation with strong multilingual lip sync and simpler input requirements. Use 1.5 Pro when dialogue and audio-visual sync are the primary requirements and reference-heavy input control isn't needed; use 2.0 when the project requires multimodal reference control or more complex input surfaces.
The positioning difference: Seedance 1.0 set the motion stability baseline. Seedance 1.5 Pro raised the visual and audio-visual ceiling. Seedance 2.0 unified multimodal reference control with expanded input flexibility. Each serves a different production need within the family.
Known Limitations
12-second maximum per generation. Longer content requires chaining multiple clips. Visual and audio consistency hold well across chained generations, but it remains an editing step.
1080p maximum resolution. Seedance 1.5 Pro does not support 4K output. For large-format display or high-resolution deliverables, Seedance 2.0 (post-June 2026 upgrade) or competing models such as Kling 3.0 are necessary.
Limited multimodal reference input. Seedance 1.5 Pro supports text, single image, and audio reference inputs but does not have the broader reference-to-video capabilities introduced in Seedance 2.0. For projects requiring multiple reference images, video clips, and audio files simultaneously, 2.0 is the appropriate model.
Multilingual lip sync quality varies by language. Mandarin and English produce the most consistent lip sync results per documented testing. Japanese and Korean perform well on short phrases with occasional drift on longer sentences. Spanish and Indonesian are supported but show more variability in complex dialogue contexts.
Multi-person simultaneous lip sync remains challenging. The technical paper acknowledges that multi-person simultaneous lip sync matching is an open problem. Single-character dialogue performs best. Multi-character scenes with overlapping speech are better approached with single-speaker dialogue or staggered speaking turns.
Accessing Seedance 1.5 Pro
Seedance 1.5 Pro is available through several channels:
ByteDance platforms (China): Doubao, Jimeng (Dreamina), Volcano Engine Ark. These remain the primary distribution channels for users in China.
International access: BytePlus API and third-party platforms. As of mid-2026, international availability is most reliable through third-party platforms following IP-related access restrictions on direct ByteDance APIs.
On our platform: Seedance 1.5 Pro is available for text-to-video, image-to-video, and reference audio-to-video generation. Check current availability in the generator.
Use Cases by Generation Mode
Multilingual advertising (T2V with dialogue) Generate the same visual ad content with dialogue in multiple target languages. The phoneme-level lip sync ensures the character's speech looks natural in each language without requiring separate animation or dubbing pipelines.
Character-driven short films (T2V or I2V with dialogue) Create dialogue-driven narrative content where character speech, facial expressions, and lip sync need to be synchronized from generation. The dual-branch architecture ensures dialogue and mouth movement are causally linked rather than matched in post.
Music video generation (Reference Audio-to-Video) Provide a music track as the reference audio input and let the model generate visual content with motion rhythm, shot timing, and camera movement that align with the beat structure and tempo of the track.
Product explainers with voiceover (T2V or I2V with audio reference) Animate product photography with synchronized voiceover narration. The reference audio-to-video mode ensures visual pacing matches the voiceover timing rather than requiring manual synchronization.
Audio-driven social content (T2V with ambient sound and music) Generate short-form social content where the audio (ambient sound, music, sound effects) and visual motion are synchronized from generation, removing the need for separate audio editing and foley work.
Frequently Asked Questions
What does "native audio" mean in Seedance 1.5 Pro? Native audio means the audio and video are generated simultaneously in a single unified process through the model's dual-branch architecture. The audio isn't added afterward—it's generated as part of the same denoising pass as the video, ensuring tight synchronization for lip sync, foley, and ambient sound.
Which languages does the lip sync support? Mandarin Chinese, English, Japanese, Korean, Spanish, and Indonesian, plus regional Chinese dialects including Sichuanese and Cantonese. Mandarin and English produce the most consistent results; other languages show good performance on short phrases with occasional drift on longer sentences.
Can I provide my own audio as a reference? Yes. The reference audio-to-video mode accepts an audio file as input and generates video with visual pacing, motion rhythm, and shot timing aligned to the audio's beat structure and tempo.
How does Seedance 1.5 Pro compare to Seedance 2.0? Seedance 1.5 Pro focuses on audio-visual joint generation with strong multilingual lip sync and simpler input requirements (text, single image, or audio reference). Seedance 2.0 offers broader multimodal reference control (up to 9 images, 3 video clips, 3 audio files) and native 4K output. Use 1.5 Pro when dialogue and audio-visual sync are the primary requirements; use 2.0 when reference-heavy input control or 4K output matter more.
Does Seedance 1.5 Pro support 4K output? No. Seedance 1.5 Pro generates at 1080p. For 4K output, use Seedance 2.0 (post-June 2026 upgrade) or competing models.
What is the maximum clip length? 2–12 seconds per generation. Longer content requires chaining multiple clips. Visual and audio consistency hold well across chained generations.
Seedance 1.5 Pro is the model for workflows where audio and video must be generated together, not assembled in post.
