AI video services

Gcore Video Streaming offers advanced video processing features for both video on demand (VOD) and live streaming:

• Transcription and translation of captions. Our AI Automated Speech Recognition (AI ASR) feature generates and translates subtitles to a wide range of supported languages, making your videos accessible to users around the globe.
• Content moderation. We analyze videos for nudity, offensive language, or violence to ensure that your audience receives safe and appropriate content.

Our suite of AI-powered video services is comprehensive and fully managed, ensuring seamless integration into your video streaming workflow.

​

AI video services infrastructure

We use our own Gcore infrastructure to execute AI models. In this way, we ensure that all data is processed within our company on EU and US servers, minimizing the risk of data leaks.

Using a unified infrastructure facilitates the integration of your project with other Gcore AI-powered solutions, such as the AI GPU Cloud for accelerating machine learning and FastEdge for low-latency edge computing for serverless app deployment.

​

Activate AI video services

AI services are available to all customers of Video Streaming in their personal account and API. If for some reason there is no access to AI, please contact the support the Gcore support. While the general usage of the AI video service is free, certain features, like translation, may incur charges. For more details, refer to our pricing page.

​

Available features

Upon activation, you can use API methods for subtitle generation and translation. However, these features are only available for videos.

​

Content moderation

You can check our API documentation for advice on integrating the different types.

If you want to do one-off checks on single videos, you can use it via the Customer Portal; these docs will explain how.

We offer six AI content moderation tasks—four related to nudity, one for sports, and one for weapons.

​

Content unsafe for work

The nsfw task is the strictest nudity detection task, as it identifies and filters videos containing inappropriately clothed people (i.e., they don’t have to be fully or partially nude to get flagged). Our algorithm flags content as NSFW if it is potentially unsuitable for viewing at work or in public places. If NSFW content is detected, the AI model provides its confidence level (in percentage).

​

Soft nudity

The soft_nudity task is less strict than the nsfw task, as it only flags videos with full or partial nudity. It detects explicit and partial nudity, including videos of male and female faces and other uncovered body parts. Use it to check whether you can publish a video to all users or block them due to offensive and inappropriate content. The check returns information about the object identified, the video frame number where it was found, and the probability of the detected object. The response only includes objects with at least a 30% confidence level.

This task allows you to identify faces and other body parts and detect more objects than hard nudity detection, so analyzing a video for soft nudity takes more time.

​

Hard nudity

The hard_nudity task is less strict than the soft_nudity task, as it only detects exposed primary and secondary sex organs in a video. It can determine whether publishing a video to all users is appropriate or should be blocked due to offensive or inappropriate content. If hard nudity content is detected, the AI model will provide its confidence level (in percentage).

This task detects fewer objects than soft_nudity, so it works faster and better when you only need to detect exposed body parts.

​

Child sexual abuse materials

The child_pornography task is the least strict nudity detector, as it will only detect child sexual abuse materials (CSAM).

We first run the soft_nudity and hard_nudity tasks. If they indicate the presence of obscene content with the involvement of children (e.g., the face of a child) in a frame, then such a video is marked as obscene. Frames are designated by the age category of identified children.

The check returns information such as the frames containing the child’s face and the child’s age. The response includes objects with a probability of at least 30%.

​

Sport activities

The sport task identifies various sporting activities in a video. This can include detecting specific types, actions, events, and moments. It lets you determine whether particular sports were filmed and even locate crucial scenes, such as goal shots in football or dunks in a basketball game.

The check returns information about the object identified, the video frame number where it was found, and the probability of the detected object. Objects with a probability of at least 30% are included in the response.

​

Weapons

The weapon task identifies the presence of weapons in a video. The check returns information about the object identified, the number of video frames where it was found, and the probability of the detected object. Objects with a probability of at least 30% are included in the response.