Agnes-Video-2.5

Agnes-Video-2.5 is a video generation model developed by Sapiens AI as part of its multimodal platform ecosystem. Designed to handle high-fidelity visual tasks, the model integrates text-to-video synthesis with advanced structural controls and multimodal references. It supports flexible integration via an asynchronous OpenAI-compatible video generation API, catering to diverse production workflows ranging from promotional media to detailed scene composition.

The model features robust first and last frame control, allowing creators to dictate precise keyframe constraints, transitions, and composition boundaries. In addition to standard text prompts, Agnes-Video-2.5 supports multi-source reference inputs including images, audio tracks, and reference videos to guide motion style, timing, and audio-visual synchronization. Output resolutions range dynamically across multiple aspect ratios and tiers, including 720P, 1080P, 1K, and 2K options.

Integration follows an asynchronous polling pattern where generation tasks are submitted to retrieve a unique task identifier for progress tracking and result retrieval. It accommodates varied scaling through companion versions like Agnes Video 2.5 Flash for lighter tasks, while providing full reference capabilities across multiple resolutions for more complex video editing and generation requirements.