en.Wedoany.com Reported - On July 3, Ocean Engine officially released the self-developed advertising governance large model Mamoda version 2.5. The model's capabilities have expanded from early single-point text recognition to full-form content recognition of images, short videos, and videos, providing technical support for advertising review, risk identification, and content governance scenarios.
The application scenarios of the advertising governance large model focus on advertising material review, violation content identification, false advertising judgment, low-quality content filtering, and platform content security management. The forms of internet advertising materials are becoming increasingly complex, making it difficult to rely solely on text rules or keyword recognition to cover risks arising from combinations of images, videos, subtitles, audio, visual actions, and context. Mamoda 1.0 started with single-point text recognition and gradually expanded to images and short videos. This 2.5 version further extends to full-form video, indicating that Ocean Engine is advancing advertising governance capabilities from single information recognition to multimodal comprehensive judgment. For advertising platforms, such capabilities can improve review efficiency, reduce repetitive manual judgments, and identify potential compliance risks earlier before ad placement.
The difficulty of video advertising governance is significantly higher than that of pure text and static images. A video material may simultaneously include voiceovers, subtitles, visuals, product displays, character actions, background text, and editing rhythm, forming complete semantics only after combining different elements.
After Mamoda 2.5 covers full-form video, it needs to handle not only visual recognition but also speech transcription, subtitle recognition, scene understanding, product information matching, and cross-segment correlation. Violations in advertising content sometimes do not appear directly in titles or copy but are hidden in voiceovers, visual hints, exaggerated demonstrations, or spliced multi-segment materials. Using large models for advertising governance can integrate text, images, video frames, and audio information into the same judgment logic, improving the ability to identify complex materials. For platform merchants and advertisers, a more stable governance model also helps reduce misjudgments, omissions, and fluctuations in review cycles, making the ad placement process more controllable.
The release of Mamoda 2.5 by Ocean Engine also reflects that advertising technology platforms are applying large models to internal governance infrastructure. In the past, advertising systems emphasized traffic distribution, placement efficiency, and conversion effectiveness, but now content compliance, brand safety, and platform responsibility have also become core capabilities. With the increase in short video ads, live-stream ads, and AI-generated ad materials, advertising governance models need to continuously identify new forms of expression and new types of risk content. The release of Mamoda 2.5 means that Ocean Engine continues to strengthen its self-developed model capabilities in advertising review and content security.
Currently, public information focuses on the release of Mamoda 2.5 and the expansion of its capability boundaries, without disclosing specific model parameters, test metrics, deployed clients, or review efficiency improvement data. What is certain is that Mamoda has expanded from text recognition to full-form video governance, and subsequent applications will increasingly focus on advertising material review, risk identification, platform governance, merchant placement compliance, and multimodal content security management.
