Changelog

​

Version 1.3.1

​

Features

• Added with_smart_analysis option in API used for image classification and API used for image classification from URL: the flag enables the new Smart Full Analysis scope, but you must be enabled in your pricing plan to use it, otherwise the parameter is ignored. When enabled, a proprietary scoring algorithm selects the areas of the image with the highest informative content; each selected crop is classified by the most advanced model of the Revelio family and the single results are combined into an overall verdict. Being an advanced, smart version of both, Smart Full Analysis replaces the standard full analysis and the heatmap analysis when enabled: for this reason it cannot be combined with with_heatmap, and a request with both options set to true returns a 422 error. See Smart Full Analysis for more details.
• Added the same with_smart_analysis option in API used for video classification and API used for video classification from URL: when enabled, the Smart Full Analysis is applied to each analyzed frame. See Smart Full Analysis on video.

​

Version 1.3.0

​

Features

Introduced the specific moments analysis: instead of letting the system choose which parts of the media to analyze, you can list the exact moments of interest of a video and/or of its audio track, or of a speech. Moments are passed as a comma-separated list of HH:MM:SS.mmm timecodes, measured from the beginning of the file (e.g. 00:00:05.000,00:01:12.500).

• Video frames: added frames_timestamps option in API used for video classification and API used for video classification from URL. One frame is extracted at each listed timecode and analyzed; the option takes precedence over the time-spaced, key frames and faces extraction strategies (frames, key_frames, key_frames_method and ensure_face_per_frame are ignored). See Classify a video.
• Video audio track: added audio_timestamps option in the same video APIs, used together with with_audio. The audio track is analyzed in short chunks centred on the listed timecodes instead of over its whole duration. It can be combined with frames_timestamps or used on its own.
• Speech: added audio_timestamps option in API used for speech classification and API used for speech classification from URL, with the same behaviour. See Classify a speech.
• Credits and limits: one frame / audio chunk is charged for each distinct timecode; timecodes close enough to fall within the same audio chunk are analyzed together and the chunks not produced are refunded. Up to 30 timecodes are accepted if your pricing plan lets you choose the number of frames / audio chunks, otherwise up to the default ones (10 frames, 5 audio chunks). Malformed timecodes or too many timecodes are rejected with a 400 error.
• Availability: specific moments are a pricing plan feature, enabled separately for video frames and for audio; if your plan does not include them, frames_timestamps and audio_timestamps are ignored.

​

Changes

• Documented ensure_face_per_frame as the faces frame extraction strategy (see Classify a video): it is a pricing plan feature (ignored if your plan does not include it), it is an alternative to key frames (key_frames and key_frames_method are ignored when it is enabled) and it is ignored when frames_timestamps is provided. Frames that are not analyzed because not enough faces were found are refunded.

​

Version 1.2.0

​

Features

• Added the API used for media validation: it checks whether an image, video or audio file can be analyzed with the given options (file type and size, integrity, minimum resolution required by the selected models) without submitting it for classification and without consuming credits.

​

Version 1.1.3

​

Features

• Added a details field inside image/video submission response payload to provide more information about single detections extracted and analyzed from the media (faces found in image/frame when with_morphing option is enabled).

​

Version 1.1.2

​

Features

• Added a details field inside audio submission response payload to provide more information about single clips extracted from the audio. This field is available with a proper pricing.
• Added ensure_face_per_frame option in API used for video classification and API used for video classification from URL: the flag ensures that the video frames used for classification contain a face, improving the accuracy of the analysis for videos where faces are present when used in combination with with_morphing.
• Added with_audio option in API used for video classification and API used for video classification from URL: the flag enables the analysis of the audio track of the video, but you must be enabled in your pricing plan to use it. When enabled, the system will analyze the audio track of the video for deepfake detection, providing a more comprehensive analysis of the video content.

​

Version 1.1.1

​

Features

• When a webhook_url is provided during submission and the send fails (e.g., due to network issues or invalid URL), the system will now retry sending the webhook notification up to 3 times with a delay of 10s between attempts. This enhancement improves the reliability of webhook notifications for clients relying on them for real-time updates.
• Fixed api_version value in the response payload of the APIs used for polling results to correctly reflect the current API version when the analysis is in new status.

​

Version 1.1.0

​

Features

• The following fields have been added to the response payloads of the APIs used for polling results:
  • Status: indicates the status of the analysis (in progress, completed, or failed)
  • Errors: in case of a failed analysis, indicates the reason for the failure
  • API Version: reports the API version at the time of the analysis, allowing comparisons between analyses produced by different backend versions
• Revelio Onboarding II model internal artificial/human thresholds tuning: the internal thresholds used to classify images as artificial or human-generated have been adjusted to improve detection accuracy based on recent evaluation data.

​

Version 1.0.0

​

Initial Release

Welcome to the first version of the identifAI API. This release brings powerful AI detection capabilities to your applications.

​

Features

• Image Classification: Detect AI-generated images using our multiple model ensemble (Revelio-IV, etc.).
• Video & Audio Analysis: specialized endpoints for video and speech deepfake detection.
• Asynchronous Processing: Webhook-ready workflow with submission and result polling.
• Partner API: Endpoints for user management and credit handling.