A video looks right in a browser preview. You click Export, download a file and move on. That proves a file was produced. It leaves several separate questions unanswered: did the look reach the file, did the timing hold, which format did you get, and is the expected audio there?
Here is a small, reproducible check using a real browser export, with the source and result frames below.
Disclosure: published by Valmera, a commercial AI video editor. An AI agent working for Valmera wrote this article and performed the checks on November 7, 2026. This test uses the free local color tool, not the Studio AI editing engine. It is one in-app browser run, not a compatibility or professional color-accuracy benchmark.
The test and what it produced
The input was Valmera's public 25-second motion-graphic source. Download the original source on that page, not its shorter edited example. It contains text, moving graphics and a visible source timecode; it has no audio stream.
In the free local color preview, the test selected Warm, set Saturation to −1%, and left Exposure, Contrast and Temperature at zero. The export ran in the browser and produced a Download WebM link.
| Observation | Source | Downloaded result |
|---|---|---|
| Picture dimensions | 1280 × 720 | 1280 × 720 |
| Video codec | H.264 | VP9 |
| Container | MP4 | WebM |
| Audio stream | None | None |
| Timing evidence | 25.000 seconds reported by the file probe | No container duration value reported; video packet timestamps ran from 0.000 to 25.079 seconds |
| Result size | — | 2,078,604 bytes |
These values came from inspecting the actual source and downloaded file with FFprobe. The packet timestamp span is not a substitute for checking playback, the final frame or the intended runtime.
1. Compare the same source moment
A screenshot of the editor is evidence about the preview. To check the delivery, extract or inspect a frame from the downloaded file and match it to the original moment.
This source has a timecode drawn into the picture. Both frames below show source 00:05:00. The first was extracted at source file second 5.0; the second at downloaded file second 5.1. At output second 5.0, the source timecode was still 00:04:27. Matching the visible source moment avoided comparing different positions of the moving playhead.
Original source frame:
Frame from the downloaded Warm export:
The block shifts toward orange and the background looks darker. That establishes a visible change in this sampled output frame. It does not establish accurate skin tones, a correct log conversion, or consistency across an entire film. This synthetic source contains no skin or camera footage to test those claims.
For a real grading review, include the subjects that matter: a face, a neutral object, important product colors, highlights, shadows and a shot boundary. A preset name alone cannot specify those requirements.
2. Check timing separately from appearance
The matched frames demonstrate why “compare second five in both files” can be insufficient. A real-time capture can have a different relationship between its output timestamps and the source moment shown in the picture.
The downloaded file also did not report a container-level duration to FFprobe. That observation alone does not prove the file is unplayable. It means this run did not establish that every player will display its duration or seek through it correctly.
For delivery, open the downloaded file in the intended player, seek near the end, inspect the final frame and check motion through the important cuts. If exact timing matters, compare visible events or timecodes at the beginning, middle and end. One matched still is not proof of frame-accurate playback throughout.
MDN describes canvas captureStream as a real-time capture of canvas contents. Treat that capture path as something to validate in its own right.
3. Inspect the format you actually received
An MP4 input did not imply an MP4 output in this run: the result was VP9 inside WebM. Renaming the extension would not convert its container or codec.
MediaRecorder.mimeType identifies the recording format selected for the recorder. isTypeSupported is a capability check; MDN notes that recording can still fail when resources are insufficient. Test the resulting file with the destination that will consume it.
Similarly, a working CSS-filter preview does not establish filtered-canvas export support. CanvasRenderingContext2D.filter has its own browser-compatibility considerations.
4. Use a source with audio to test audio preservation
The UI reported an export without audio. The file probe confirmed no audio stream in the result. But the original was silent too: this run neither proves audio preservation nor demonstrates that source audio was lost.
For that check, use a separate source with known speech or sound. Inspect the downloaded audio stream, listen to it, and compare a visible synchronization cue where available. Record that result separately from a silent-file test.
A useful export record keeps the source filename, selected settings, browser/version, output filename, dimensions, container, codec, timing observations and audio result together. Mark unperformed checks as untested instead of turning “download available” into a blanket pass.
Try the check with your footage
The Valmera color tool offers a free local preview and supported exports for clips up to two minutes, with a maximum long edge of 1920 pixels. Format and audio support depend on the browser. No account is needed for that local workflow.
When you want to direct the edit through conversation, its Continue with these settings option prepares a brief for Valmera Studio. You still upload the original there; the local file is not transferred by that button. Account creation and uploads are free; AI editing requires a subscription. Review the Studio result separately from this browser-made convenience export.
