Skip to main content
This page lists changes to the Reality Defender Platform API and SDKs, newest first. Use the filters to show only breaking changes, improvements, fixes, or other changes that may affect an existing integration.
BreakingModel updatesFixed

Breaking changes

  • Audio uploads are blocked on the Free API plan. POST /api/files/aws-presigned now returns HTTP 403 with the code free-tier-restriction and the message “Forbidden: Video, audio, and text uploads require a paid plan” when an organization on the Free API plan requests an upload URL for an audio file. See Free Plan Restriction for details. If your integration uploads audio on the Free API plan, upgrade to a paid plan or contact Sales. After an upgrade, it can take a few minutes for access to take effect.

Model updates

  • Audio analysis covers the full file. Speech in some files could previously be mistaken for silence and left out of the analysis. This speech is now analyzed, so results reflect more of each file. The response format is unchanged. Scores for the same file can differ from earlier results.

Fixed

  • YouTube downloads resolve more reliably. YouTube links submitted to POST /api/files/social that point to non-MP4 files are now downloaded and analyzed instead of failing. If the download provider reports a failed download, the request now ends promptly with DOWNLOAD_FAILED in overallStatus instead of staying in DOWNLOADING. Request and response formats are unchanged, and no change is needed if you use an official SDK.
  • YouTube Shorts links work again. YouTube Shorts links submitted to POST /api/files/social are downloaded and analyzed.
Changes

Breaking changes

None.

Changes to be aware of

  • Scans still analyzing after 2 hours now end as UNABLE_TO_EVALUATE. If a scan stays in ANALYZING for 2 hours (previously 12 hours), overallStatus on GET /api/media/users/{requestId} is set to UNABLE_TO_EVALUATE. Resubmit the file to try again. The official SDKs already treat UNABLE_TO_EVALUATE as a final status, so no code change is needed.
Fixed

Breaking changes

None.

Fixed

  • YouTube links no longer fail. YouTube links submitted to POST /api/files/social are downloaded and analyzed again. Since the beginning of October, these requests could end with DOWNLOAD_FAILED in overallStatus. If you have YouTube requests that failed this way since then, resubmit the links. YouTube Shorts links aren’t supported yet. Request and response formats are unchanged, and no change is needed if you use an official SDK.
Model updatesImproved

Breaking changes

None.

Model updates

  • Image model retrained on newer image-to-video generators. The image model is retrained with frames from videos made by newer image-to-video generators, including Google’s Veo 3.1, to improve detection of this content. The response format is unchanged. Scores for the same image can differ from earlier results.

Improved

  • More stable video processing. Video scans are more stable and less likely to slow down during heavy video traffic.
Model updates

Breaking changes

None.

Model updates

  • Image model retrained for OpenAI GPT Image content. The image model is retrained to catch more images created with OpenAI’s GPT Image models. The response format is unchanged. Scores for the same image can differ from earlier results.
Improved

Breaking changes

None.

Improved

  • Social media link imports now accepts TikTok tiktok.com/t/<code> short links and TikTok mobile links m.tiktok.com.
Model updatesImprovedFixed

Breaking changes

None.

Model updates

  • Faster video results. The video model was optimized for speed, so video results return faster.

Improved

  • More social media link formats. Social media link imports now accept mobile and shortened links: mobile.twitter.com, m.facebook.com, and web.facebook.com links; fb.me, instagram.com/share/ short links; Threads share links; and vm.tiktok.com and vt.tiktok.com short links for TikTok videos.

Fixed

  • Facebook share links for images. Images shared with facebook.com/share/p/ links were sometimes treated as videos and returned DOWNLOAD_FAILED. They now import and are analyzed as images.
  • More reliable presigned uploads. A presigned URL could stop working minutes after it was requested, before the lifetime it was given, because it was signed with credentials that were about to expire. These URLs now stay valid for the lifetime they were issued with.
ImprovedFixed

Breaking changes

None.

Improved

  • Image support for social media links. Social media link imports now download and analyze images, not just videos. Image scans are supported only for public posts on X, Facebook, Instagram, and Threads. For posts with multiple images, only the first image is analyzed. Mobile and shortened links aren’t supported yet.

Fixed

  • Social media imports no longer stall. Imports that could stay in a downloading state indefinitely now time out. A timed-out import returns DOWNLOAD_FAILED in overallStatus, and socialLinkDownloadFailed is true.
BreakingImproved

Breaking changes

  • Presigned uploads use a new domain. The signedUrl returned by POST /api/files/aws-presigned now points to api.prd.realitydefender.xyz instead of AWS S3. If your network or client restricts which hosts it can upload to, allow api.prd.realitydefender.xyz. The official SDKs already use this flow, so no code change is needed if you use one. If you upload with your own HTTP client, send a PUT request to the signedUrl exactly as returned, with the file as the raw request body and a Content-Type: application/octet-stream header.

Improved

  • 30-minute limit on video. Video files longer than 30 minutes are no longer analyzed. Split longer recordings into segments of 30 minutes or less before uploading. Media Detail returns a not-applicable result instead of a processing error. overallStatus and resultsSummary.status are both NOT_APPLICABLE. resultsSummary.metadata.reasons contains:
  • Context-aware explainability fields withheld. Media Detail still returns the image or video context model. Its data keeps score and decision. The sources field remains only when reverse-image results are already included. These undocumented fields are omitted: explanation, image_description, manipulation_reasons, potential_source, image_type, and recognizable_identities. aggregation.json for scans run after this change omits contextResult. Files written earlier still include it. If your integration reads those fields, use the documented response fields only.
  • Faster video results. Video analysis now completes about twice as fast.
  • Shorter queue times during busy periods. Video processing capacity now scales with demand.
Changes

Breaking changes

None.

Changes to be aware of

  • API keys and webhooks have their own pages. In the web app, API keys and webhooks are now separate pages in the main navigation, each with a tab per product. To manage RealScan keys, go to API keys > RealScan. To manage RealScan webhooks, go to Webhooks > RealScan.
Improved

Breaking changes

None.

Improved

  • Faster results for social media links that can’t be downloaded. When a submitted social media URL can never be downloaded, such as a post with no media attached or content that requires app-only access, the request now fails after the first confirmed error instead of retrying up to five times. You receive the failure result sooner. The set of links that fail is unchanged.