Seedance 2.5

Positioning Overview

Seedance 2.5 is a joint audio-video generation model launched by ByteDance's Seed team. This endpoint handles pure text-to-video generation: developers write a prompt and receive a finished video with synchronized voice, sound effects, and ambient audio. It supports durations from 4 to 30 seconds, 720p and 1080p resolutions, seven aspect ratios, and mp4/mov output, and can optionally enable web search to supplement time-sensitive information in the prompt. It is suited to development teams that need to rapidly produce ads with sound, storyboard previsualizations, and short social media videos, and integrate them at scale through the unified HiAPI.ai interface.

Core Highlights

• Text generation with synchronized audio — Audio generation is enabled by default; visuals, voice, sound effects, and background audio are produced in one pass, eliminating post-production dubbing and audio alignment.
• Duration and aspect ratio specified per shot — Integer durations from 4 to 30 seconds can be freely set, with seven aspect ratios from 16:9 to 21:9, and adaptive letting the prompt decide.
• Prompt-triggered web search — With web_search enabled, the model can retrieve time-sensitive information referenced in the prompt for content requiring recent facts.
• Output formats suited to post-production workflows — mp4 by default for direct distribution, mov for professional editing and post-production.

Capabilities

One-pass audio-video generation: A single generation produces both visuals and synchronized audio, including dialogue, sound effects, and background sound; the output is a ready-to-use video file, reducing post-production compositing costs.

Controllable duration and resolution: Billed by duration, you can specify integer durations from 4 to 30 seconds and 720p or 1080p resolution, making it easy to control final video specifications per channel requirements.

Multi-aspect-ratio adaptation: Offers 16:9, 4:3, 1:1, 3:4, 9:16, 21:9, and adaptive aspect ratios, so the same prompt can cover landscape, portrait, and square placements.

Multilingual prompts: Supports prompts written in Chinese, English, Japanese, Korean, Spanish, Portuguese, Indonesian, Malay, Thai, Arabic, and Vietnamese.

Retrieval augmentation and packaging options: Can enable web search to introduce time-sensitive information, and choose whether to attach an AI-generated watermark and output mp4 or mov.

Use Cases

• Advertising and brand teams: one-sentence selling point descriptions — Generate 10-to-30-second product clips with voiceover and music, ready for feed advertising or social media teasers.
• Short-video creators: vertical narrative scripts — Use the 9:16 aspect ratio to generate complete narrative segments, producing footage ready to upload.
• Film pre-production teams: storyboard text descriptions — Turn storyboard text into dynamic previsualizations with sound, used to confirm shot pacing and duration before filming.
• Platform developers: batch template-based video production — Use the unified interface to batch-generate versions with different aspect ratios and durations from the same prompt template, integrating into their own content production pipelines.

Limitations and Selection Advice

This endpoint is pure text-to-video and does not accept image or reference material inputs. When you need to lock a character, product appearance, or make timestamp-level modifications, use the reference-generation or image-to-video endpoints in the same family instead. A single clip is limited to 30 seconds with no continuation parameter, so longer narratives require generating shots separately and stitching them together. Physical plausibility of complex actions and stability of multi-subject interactions may still be insufficient. For lightweight, short-duration needs, simpler models can also be considered to control costs.