A few things worth knowing.
Clear answers about how frame capture works and where its limits are.
Does my video get uploaded?
No. The selected video and extracted frames are processed locally by this tool. Your browser still requests website pages and assets from the host. Contact inquiries send only the fields you choose to submit.
Which video formats work?
It depends on the codec your browser can decode, not just the filename. MP4 with H.264 is a practical first choice. MOV and WebM compatibility varies. There is no universal fallback decoder.
Are the exported images lossless?
PNG avoids an additional lossy image encode. JPG and WebP use lossy quality settings. Images keep the decoded frame’s dimensions, but source blur, video compression, and browser color handling still apply.
Can I download a real ZIP?
Yes. The ZIP contains the captured images with numbered filenames and requested timestamps. The gallery allows up to 120 frames and 96 MB of encoded images. Download and clear it before collecting more.
What does the sharpness score mean?
It is a heuristic estimate from Laplacian variance on a downscaled whole-frame sample, not an AI model or a calibrated quality test. Noise and busy backgrounds can increase it. Inspect your subject before choosing a frame.
Does the tool export every source frame?
No. Batch extraction samples requested times at the chosen interval. Seeking is handled by the browser and is not guaranteed to match an exact source frame index.
Is there a file size limit?
Frame extraction is limited to 2 GiB and 8,294,400 decoded pixels per frame; other tools have separate limits. Performance also depends on codec, browser, and device memory. Use a shorter or smaller compatible clip if playback fails.
Can I use images commercially?
Extraction does not grant rights in the video. Use footage you own or are permitted to reuse and follow the applicable license and privacy requirements.
Why does a batch omit the end time?
The start is included and end excluded. A 0–10 second range sampled every second requests ten pictures at 0 through 9 seconds. This prevents overlapping batches from requesting the same boundary twice. Use a valid timestamp below the duration for a separate end capture.
Can I extract a first and last reference together?
Use the endpoint extractor. The last capture seeks just before the reported duration; it is not an exact last-source-frame enumeration. A fade or black ending may require an earlier moment.
What is inside frames.json?
When you enable ZIP metadata, it records source filename and exported picture names, dimensions, formats, player times and capture roles. It does not contain YOLO labels, COCO boxes or certified source-frame identifiers.
Why do different tools have different limits?
Capturing a still, re-encoding a video and running a model have different memory costs. The current limits table separates each workflow. A file admitted by extraction can be too large for compression.
Does image or video upscaling always use AI?
No. Fast resize is the default and enlarges existing pixels. Optional AI modes load an actual model and can infer incorrect detail. The frame sharpness estimate is a heuristic and does not use AI.