How fast are Creaptor’s converters?

What we verified

Conversions completed

180/180 Across these five scenarios and six profiles.

Output files checked

216 Every result fully decoded independently.

Repeat engine downloads

0 B Maximum across the measured repeat runs.

• AAC extraction copies the existing audio. Compatible AAC → M4A needs no re-encoding and no extra encoder download.
• FLAC and batch conversion completed in every tested profile. WebKit used a validated FFmpeg fallback where its browser decoder could not process FLAC.
• Repeat runs reused the engine code. No engine was downloaded before Run, and no measured repeat downloaded engine code again, including WebKit’s FFmpeg path.

How long did conversion take?

First is a fresh browser profile. Repeat runs the same file and operation again in the same tab. Times include code loading, processing and result checks.

Desktop builds on an Apple M4 Pro with 48 GiB RAM. These measurements do not promise every conversion will be faster: extra validation and storage can add time. All six profiles appear below.

How much code was downloaded?

The processing engine loads after Run. This table includes the engine, codecs and any extra interface helpers downloaded before the result was ready.

FFmpeg is loaded only when the lighter path cannot handle the file. Its bytes were reused on repeat runs here; browser storage can still be cleared or evicted.

KB/MB are decimal compressed response-body bytes. Measurements used local HTTP with gzip, without optional telemetry. WASM means WebAssembly; dedicated codecs can include it inside a JavaScript chunk.

Which browsers were tested?

Mobile profiles emulate layout, touch and user agent on this Mac. They do not measure phone hardware. WebKit is Playwright WebKit, not Safari.app. Real iOS Safari, Android devices and embedded Telegram, Instagram or TikTok browsers were not tested.

Full follow-up results

Open a profile for all 5 scenarios. Values show medians, min–max ranges and the number of captured timings (n). On smaller screens, scroll inside the table.

Method & downloadable data

1. Same inputs, three first/repeat pairs. Audio fixtures were 30-second stereo files; AAC extraction used a 12-second 720p MP4; AVI used a roughly four-second clip. Each pair started a new browser process and disk-backed profile. OS caches were not cleared.
2. Time until ready. From Run to the enabled download action, including built-in result validation. Selection, saving and independent checks were outside the timer. Jobs ran sequentially, without network or CPU throttling.
3. Observed code and downloads. Worker, codec and WASM probes recorded the processing paths; a local server recorded gzip response bytes under production cache/CSP rules. Observed codecs include output validation. Probe overhead was not calibrated for this corpus.
4. Every output checked. All 216 files passed full independent FFmpeg decoding, required-track, duration, codec, audio and video checks. AAC extraction also passed packet-identity checks. AVI output retained the documented source timeline normalization.

This follow-up covers five selected scenarios, rather than a fresh run of the earlier 38-scenario suite. It does not establish a full-site success rate, fallback share, large-file safety, hardware acceleration, peak memory or perceptual quality.

Data is a dated snapshot of build 679531e. Evidence downloads only when you choose a link.

• Every test run (CSV) · 88.7 KB
• Every downloaded file (CSV) · 203.0 KB
• Environment and input files (JSON) · 15.1 KB
• Complete evidence and reproduction scripts (ZIP) · 2.88 MB

Earlier 38-scenario report and raw data (ZIP) · 1,368 attempts on build 35d47a0, before the reliability updates. The historical measurements are preserved in that archive.