No original-video upload
The complete source video remains in the browser and is not written to ACEAI artifact storage.
Privacy-first AI coaching
Privacy-first AI tennis coaching keeps the full video local.
ACEAI analyzes selected key frames instead of uploading the original video file. This page explains the practical data flow, the role of retained evidence, and the limits of the privacy model.
For the binding legal terms, always read the current ACEAI Privacy Policy.
Data flow overviewEvidence review
One clip · two clearly separated layers
01
BrowserOriginal video, playback, motion scan, and frame selection.
02
AnalysisSelected JPEG frames, pose evidence, and request context.
03
Account memoryDerived results retained for progress and reuse.
Explicit retained evidence
Selected frames, pose evidence, prompts, model responses, and derived reports may be stored for account-scoped reuse.
Private infrastructure
Production analysis artifacts use private object storage and are addressed by internal user identifiers rather than public email addresses.
A transparent workflow
From court footage to a focused next step
ACEAI separates observable evidence from coaching interpretation so you can understand where each suggestion comes from.
- 1
Open locally
The browser reads the video you selected and previews it without first sending the complete file to the server.
- 2
Reduce locally
Motion, sharpness, and pose signals help choose a small chronological set of evidence frames.
- 3
Analyze selected data
The chosen JPEG frames and compact context are sent to the configured vision analysis service.
- 4
Retain derived evidence
Account-scoped artifacts and database records support cached results, coaching memory, and later comparisons.
What remains on the device
The source video is used as a local working file
When you choose a video in the ACEAI web app, the browser loads it for playback, seeking, frame capture, and local pose measurements. The app does not send the complete original video file to its analysis API or save that source file in its production artifact store.
Keeping the source local reduces the amount of data transferred and avoids retaining a full practice recording when a small number of still frames can support the requested review. It does not mean the entire analysis runs offline: selected evidence leaves the device.
What leaves the device
Selected evidence frames carry the visual information needed by the model
The browser sends up to eight JPEG frames selected around useful motion windows. It also sends frame timestamps, compact pose evidence, the declared analysis focus, sampling details, and racket-hand evidence. These items let the service reason about sequence and visible body structure without receiving the whole clip.
A third-party model provider processes the selected images and request context to generate a candidate report. Like any external AI request, that processing has provider-specific handling and operational risks. ACEAI’s privacy policy should be used to evaluate the current providers and applicable terms before uploading sensitive material.
Key-frame images
Compressed still images from selected moments, not a continuous copy of the source video.
Pose evidence
Normalized joint coordinates and confidence values measured for the same evidence frames.
Request context
Shot focus, timing, sampling information, and racket-hand evidence required to interpret the sequence.
Model response
Structured coaching suggestions plus the raw response needed for traceability and future parsing.
Why retain analysis evidence
Incremental memory depends on reusable intermediate results
ACEAI is designed to avoid running the full analysis again whenever a coaching memory changes. Selected frames, model output, normalized observations, and processing metadata form an immutable analysis artifact. The memory system can then update a skill’s confidence, trend, status, and practice plan from those intermediate results.
This separation also allows a new memory algorithm or display label to work from preserved evidence rather than asking the model to reinterpret the original video every time. The tradeoff is that key frames and derived records remain associated with the account according to the service’s retention and deletion practices.
Cache identical analysis
A known video hash and analysis version can avoid unnecessary duplicate model work.
Track changing skills
New observations can strengthen, weaken, resolve, or reactivate an existing coaching memory.
Preserve provenance
A stored artifact records which evidence, prompt, model, and schema produced a report.
Support future formats
Raw model material helps recover information when the normalized schema evolves.
Account and operational security
Private storage is a control, not a promise of zero risk
ACEAI uses Google identity for account access and a time-limited application session. Production database access is kept inside the deployment network, and analysis objects are stored in a private bucket rather than exposed through anonymous public links.
No internet service can honestly claim that storage, software, identity providers, or model processors carry no risk. Use a clip that contains only what is needed for technique review, avoid filming bystanders when possible, and do not upload confidential material. Consult the Privacy Policy for current retention, deletion, contact, and cross-border processing details.
Questions, answered
Frequently asked questions
Is the entire tennis video uploaded to ACEAI?
No. The web workflow keeps the original video in the browser. It sends a small set of selected JPEG frames plus timestamps, pose evidence, and analysis context. Those selected frames and derived artifacts may be retained with the account.
Does local frame selection mean the analysis is fully offline?
No. Motion scanning, frame selection, and pose extraction happen in the browser, but selected frames are sent to the server and configured vision model. The resulting evidence and report support cloud-based account memory.
Why does ACEAI retain selected key frames?
They provide provenance for the report and allow memory, caching, and future schema updates to reuse intermediate evidence without obtaining or reprocessing the original video.
Are stored analysis artifacts publicly accessible?
The intended production configuration uses private object storage and does not expose anonymous read access. Application authorization, deployment security, and provider infrastructure still matter, so private storage should not be interpreted as zero risk.
Is this page the legal Privacy Policy?
No. This is an educational explanation of the product architecture. The Privacy Policy governs the service and should contain the current legal details, contact route, retention terms, and user rights.
Train with evidence
A smaller input can still support a focused coaching conversation.
Choose a short tennis clip, keep the original on your device, and review what the selected evidence can reasonably show.