# Seedance 2.5 now available on Vercel AI Gateway

Source: [Vercel](https://vercel.com/changelog/seedance-2-5-now-available-on-vercel-ai-gateway)  
Feed7 permalink: https://feed7.dev/p/seedance-2-5-now-available-on-vercel-ai-gateway-1vhkxhw  
Published: 2026-08-06T00:00:00.000Z  
Trust: Official Source (official_source)

## Why Included

Seedance 2.5 adds multimodal video generation and local edits to AI Gateway, with clips up to 30 seconds and separate image and video references.

## Source Summary

**Seedance 2.5** is available through AI Gateway as bytedance/seedance-2.5. It generates clips up to **30 seconds**, accepts text, video, image, and audio together, and supports prompts in **more than ten languages**.

## Practical Implication

Builders can use referenceImages for appearance and referenceVideos for motion, camera work, or pacing. Existing footage can also be edited locally to swap elements while retaining its framing, camera, and timing.

## Agent-Ready Context

**Seedance 2.5** is available through AI Gateway as bytedance/seedance-2.5. It generates clips up to **30 seconds**, accepts text, video, image, and audio together, and supports prompts in **more than ten languages**.

Builders can use referenceImages for appearance and referenceVideos for motion, camera work, or pacing. Existing footage can also be edited locally to swap elements while retaining its framing, camera, and timing.

References guide a new generation, while editing modifies a finished clip. Combining those roles ambiguously in one prompt produces less stable results, so each source asset needs an explicit purpose.

## Connected Context

Feed7 judgment across 419 accumulated Signals:

This broadens AI Gateway’s video routes with clips up to 30 seconds, multilingual prompts, combined multimodal inputs, and two distinct workflows: references guide new generation, while editing changes existing footage. The explicit asset roles make prompt and pipeline design consequential; availability alone still provides no comparative evidence on quality, latency, cost, or consistency.

- [MiniMax H3 now available on AI Gateway](https://feed7.dev/p/minimax-h3-now-available-on-vercel-ai-gateway-0c8g0d0) — Both expose reference-conditioned video through AI Gateway, but Seedance adds finished-clip editing and requires each asset’s generation or editing role to be explicit, whereas MiniMax separates mutually exclusive conditioning modes.
- [calesthio/OpenMontage](https://feed7.dev/p/openmontage-0bqxlgb) — OpenMontage’s provider-selection and approval pipeline could operationalize Seedance’s required distinction between reference-guided generation and editing existing footage.
- [Comfy-Org/ComfyUI](https://feed7.dev/p/comfyui-1515z33) — ComfyUI offers an inspectable, reusable alternative for encoding multimodal asset roles, contrasting with Seedance’s hosted gateway route and its simpler SDK access.
- [How Searchable ships customer-requested features in 30 minutes on Vercel](https://feed7.dev/p/how-searchable-ships-customer-requested-features-in-30-minutes-on-vercel-1woiy7v) — Searchable’s account supports the implementation benefit of adding this route through an existing Gateway integration, though it does not establish Seedance’s media performance.

## Context Map

- Layer: tools
- Domains: video, audio
- Topics: generative-media, agent-sdks, gateways

## Uncertainty

- References guide a new generation, while editing modifies a finished clip. Combining those roles ambiguously in one prompt produces less stable results, so each source asset needs an explicit purpose.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.