Seedance 1.5 Pro on Novita AI: Complete Developer Guide

Seedance 1.5 Pro on Novita AI brings ByteDance’s advanced audio-visual AI to developers at scale. This 4.5-billion-parameter model delivers phoneme-level lip-sync accuracy across 8 languages, native cinematic controls, and synchronized spatial audio—capabilities that previously required expensive post-production teams.

For developers building dialogue-driven video applications, Novita AI’s serverless deployment with configurable resolution (480p/720p) and aspect ratios. Below we break down why this matters for production workflows.

What Makes Seedance 1.5 Pro Different

Native Joint Audio-Visual Generation

Unlike sequential video-then-audio pipelines, Seedance 1.5 Pro uses a dual-branch diffusion transformer that generates synchronized video frames and audio waveforms simultaneously. The cross-modal joint module maintains millisecond-level alignment between visuals and sound, solving the lip-sync drift issues plaguing earlier models.

This architecture delivers three critical advantages: phoneme-accurate lip movements (mapping individual speech sounds to correct mouth shapes), spatial audio positioning (footsteps echo correctly based on room acoustics), and emotional coherence (music intensity matches visual pacing). For dialogue-heavy applications, this eliminates the need for manual audio cleanup.

Multilingual Dialect Support

The model handles 8 languages including regional Chinese dialects—Sichuanese, Taiwanese Mandarin, Cantonese, Shanghainese—plus English, Japanese, Korean, Spanish, Portuguese, Indonesian, and Hindi. Each dialect retains authentic pronunciation patterns while maintaining lip-sync accuracy, critical for localized content campaigns.

Cinematic Control Vocabulary

Developers can specify camera movements in natural language: “dolly zoom on subject’s emotional peak,” “tracking shot following car chase,” “whip pan transition between speakers.” The model translates these directives into smooth camera motion with correct physics—no manual keyframing required.

Technical Specifications of Seedance 1.5 Pro

Specification	Details	Developer Impact
Model Architecture	4.5B parameter dual-branch diffusion transformer	Low VRAM for inference (~16GB), fast generation
Native Resolution	720p (480p optional)	Requires external upscaling for 4K workflows
Duration Range	4-12 seconds per clip	Best for short-form content, not long narratives
Audio Features	Spatial positioning, environmental effects, emotion-synced music	Significantly reduces post-production audio work

Visual Quality Assessment

Independent reviewers score Seedance 1.5 Pro at 7-8/10 vs live-action, noting improved skin textures and reduced banding compared to Kling 1.6 or Runway Gen-3. However, native 720p output limits fine detail—expect soft edges on text overlays and slight exposure inconsistencies across cuts.

The model handles complex physics well: snow particles, high-speed motion blur, water simulations render convincingly. Occasional hypersharpening artifacts appear in hair and foliage, addressable with “natural lighting” prompts.

Using Seedance 1.5 Pro on Novita AI

API Integration Setup

Novita AI exposes Seedance 1.5 Pro (their naming for 1.5 Pro) via two REST endpoints: Text-to-Video (T2V) and Image-to-Video (I2V). Both follow OpenAI-compatible request/response patterns with asynchronous task polling.

Text-to-Video Example

curl --location --request POST 'https://api.novita.ai/v3/async/seedance-v1.5-pro-t2v' \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${API_KEY}" \
--data-raw '{
  "fps": 24,
  "seed": 42,
  "ratio": "16:9",
  "prompt": "A colossal sci-fi mecha stands in the rain-soaked city nightscape, neon lights reflecting off its metallic armor. Slow motion captures every raindrop bouncing off the mecha's shoulder as it raises its arm cannon. Cinematic depth of field blurs the glowing skyscrapers behind. Anime style, dramatic lighting, 4K quality.",
  "duration": 8,
  "watermark": false,
  "resolution": "720p",
  "camera_fixed": false,
  "service_tier": "default",
  "generate_audio": true,
  "execution_expires_after": 172800
}'

Image-to-Video for Controlled Output

I2V mode accepts starting and ending keyframes, useful for precise character design consistency:

curl --location --request POST 'https://api.novita.ai/v3/async/seedance-v1.5-pro-i2v' \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${API_KEY}" \
--data-raw '{
  "fps": 24,
  "seed": 42,
  "image": "https://pub-32c83cde150f4d468bd19f0a5e372c23.r2.dev/multimodal-assets/2026-02/1771500580027-43159b2510134742.jpg",
  "ratio": "adaptive",
  "prompt": "A young woman dances energetically on a city street with graffiti walls and neon lights. The camera follows her fluid movements as she spins and grooves to the rhythm. Shot scale changes from medium to close-up, capturing her confident natural expression. Detail enhancement on her facial features and clothing textures. Smooth stabilization throughout the dance sequence with consistent neon lighting reflections.",
  "duration": 4,
  "watermark": false,
  "resolution": "720p",
  "camera_fixed": false,
  "service_tier": "default",
  "generate_audio": true,
  "execution_expires_after": 172800
}'

Cost of Seedance 1.5 Pro on Novita AI

Novita AI charges per generation task, not per token. 

Seedance 1.5 Pro · Text to Video (T2V)

Resolution	Audio	Online ($/s)	Batch ($/s)
480P	Silent	$0.012	$0.006
480P	Audio	$0.024	$0.012
720P	Silent	$0.026	$0.013
720P	Audio	$0.052	$0.026

Seedance 1.5 Pro · Image to Video (I2V)

Resolution	Audio	Online ($/s)	Batch ($/s)
480P	Silent	$0.012	$0.006
480P	Audio	$0.024	$0.012
720P	Silent	$0.026	$0.013
720P	Audio	$0.052	$0.026

Cost-Saving Tip: 

• Start with 480p for prototyping (fastest generation), then regenerate final versions at 720p.
• Use fixed camera (`camera_fixed: true`) to reduce processing time by ~30% when static shots are acceptable.
• Online jobs are processed in real time and return results immediately, while Batch jobs are executed asynchronously for large-scale generation at a lower cost.

Prompt Engineering Best Practices of Seedance 1.5 Pro

Structure for Optimal Results

Seedance 1.5 Pro performs best with explicit, layered prompts that separate visual action, audio cues, and camera directives:

[CHARACTER ACTION] + [DIALOGUE WITH LANGUAGE] + [AUDIO ENVIRONMENT] + [CAMERA MOVEMENT] + [LIGHTING/STYLE]

Example:
"Elderly woman laughs heartily while kneading dough in rustic kitchen. 
Says 'This is my grandmother's recipe!' in Sichuanese dialect with warm smile.
Background sounds: bubbling pot, wooden spoon clinking, soft folk music.
Slow dolly zoom focusing on hands, then face.
Warm afternoon sunlight through window, shallow depth of field."

Dialect and Emotion Keywords

For multilingual projects, specify dialect explicitly to trigger correct phoneme models:

• Chinese dialects: “in Cantonese dialect,” “using Taiwan Mandarin,” “with Shanghainese accent”
• Emotional intensity: “yelling angrily,” “whispering nervously,” “speaking confidently”
• Non-verbal audio: “footsteps echoing on marble,” “glass shattering off-screen,” “distant traffic noise”

What to Avoid

Reviewers note struggles with very complex action sequences—keep to 1-2 characters and limit simultaneous movements. Avoid prompts like:

• “Five characters having a group discussion” (model handles max 2-3 speakers well)
• “Character runs, jumps, then fights” (too many sequential actions for 10s)
• “Epic battle scene with explosions” (not optimized for action, better suited for dialogue/drama)

Common Gotchas and Solutions of Seedance 1.5 Pro

Issue: Exposure Shifts Between Cuts

Cause: Native 720p generation sometimes produces brightness inconsistencies across scene transitions.

Fix: Add “consistent lighting throughout scene” to prompt, or normalize exposure in post using Lumetri Color/Color Wheels.

Issue: Soft Text Overlays

Cause: 720p native resolution doesn’t retain sharp text edges.

Fix: Generate video without on-screen text, then add titles/graphics in post at higher resolution using After Effects or Motion.

Issue: Audio Drift in Multi-Speaker Scenes

Cause: Complex overlapping dialogue can occasionally desync by 100-200ms.

Fix: Limit to 2 speakers per clip. For group conversations, generate separate shot/reverse-shot clips and edit together.

Issue: Limited Camera Customization

Cause: Model interprets camera directives but doesn’t accept precise focal length/f-stop values.

Fix: Use descriptive terms like “shallow depth of field” or “wide-angle perspective” instead of technical specs.

Seedance 1.5 Pro on Novita AI delivers production-ready audio-visual generation for dialogue-focused short-form content. Its phoneme-level lip-sync accuracy and OpenAI-compatible REST API make it a fast path from script to rendered video for developers building localized ads, micro-dramas, and music video prototypes.

Frequently Asked Questions

Novita AI is an AI & agent cloud platform helping developers and startups build, deploy, and scale models and agentic applications with high performance, reliability, and cost efficiency.