Add Subtitles to Video

From a caption track to a finished clip

The Add Subtitles tool renders supplied timed text into video frames in your browser. A fictional six-second quiet-workshop example shows the decisions involved, from matching the track to checking the downloaded result. It is not an automatically transcribed recording.

Begin with the recording and the words that belong to it

A useful subtitle export starts with two matching sources: the finished video edit and a timed text file prepared for that edit. Choose the actual recording you intend to share, not a similar take with a different opening. Then identify the SRT or WebVTT track that belongs to it. A removed pause can put every later sentence in the wrong place even when both files open successfully. Keep the original video and track in an unchanged folder. The downloaded MP4 will be a new delivery copy, so those originals remain your route back to an uncaptioned picture or revised wording.

This tool does not listen to the recording or invent the missing text. If you have video but no subtitles, prepare and review a timed track elsewhere before starting. If you have only a paragraph of dialogue, it still needs start and end times for each cue. Translation is also a separate editorial job. Check names, quantities, speaker references, and meaningful sounds against the source rather than assuming a valid file contains accurate language. A perfectly synchronized mistake becomes harder to fix after it is printed into every affected frame. Resolve uncertain wording before choosing its visual treatment here.

Processing uses local files in the current browser session, without a private-media upload or account connection. That makes the utility convenient for a short prepared recording, but it does not turn the tab into storage. There is no autosave, remote project history, or recovery promise after the session closes. Work from copies you can find again and download the useful outputs before leaving. Normal site analytics can still operate on the page; local media processing is not a claim that the entire website makes no network requests. Handle confidential material according to your own review and sharing requirements throughout the workflow.

Check the track before giving it a visual style

Use a basic UTF-8 SRT or WebVTT track no larger than 64 KiB, with at most 200 cues and 500 characters in each cue. These are working boundaries for this utility, not universal caption-format limits. Review the first and last lines after loading so an older revision does not slip through because its filename looks familiar. Accents and punctuation should already read correctly. If they appear damaged, repair the encoding at the source instead of accepting the garbled words. The renderer can display supplied text, but it cannot infer which original characters an incorrectly decoded file was meant to contain.

Cue order matters. Each interval needs a start before its end, and the sequence must be chronological without overlaps. The tool rejects conflicting timing instead of silently sorting the file or throwing away simultaneous words. A discussion with overlapping speakers may need an editorially reviewed sequential track for this particular rendering workflow. Unsupported structured WebVTT blocks, including STYLE, REGION, and NOTE, are outside its basic input scope. Obtain a suitable plain working copy rather than expecting the tool to interpret a complex caption document as if every feature were supported. Keep that richer source when its additional information remains important.

Some accepted text can still carry formatting or layout information that will not survive plain conversion. Read the reported losses and acknowledge the simplification only when you understand its effect. A speaker name written as ordinary text can remain useful even when a colored speaker style disappears. Conversely, stripping a positional convention may remove an important distinction that needs another treatment. Use the Subtitle Editor for focused wording and timing repairs before returning to the burn-in step. Editing the source track first is especially helpful when several cues need restructuring rather than one shared timing adjustment.

Walk through a six-second quiet-workshop example

Imagine a fictional six-second workshop recording with no spoken dialogue. A maker places a paper pattern on a table, lines up a ruler, and lifts the finished shape. The supplied track contains three short cues: 0.500 to 2.000 seconds, “Place the pattern flat.”; 2.250 to 4.000 seconds, “Align the ruler.”; and 4.250 to 5.750 seconds, “Check the finished edge.” These are deliberately authored demonstration instructions, not recognized speech. Load a matching local video and that existing track, select the full interval from 0.000 to 6.000, and begin with an offset of zero so the source relationship remains unchanged.

Preview the first instruction where the pattern reaches the table. At 0.500 seconds it should appear; at 2.000 seconds its interval has ended. The next cue begins at 2.250, leaving a quarter-second gap. Those small pauses make the three steps distinct without leaving an instruction on screen during the next action. Play the entire six seconds as well as inspecting individual frames. A still frame can prove that a sentence fits, but only playback shows whether the viewer has time to read it while following the hands. Adjust the source wording or timing if that combined task feels rushed.

Keep the example modest. It demonstrates three supplied cues over six seconds, not a promise that every workshop process should use the same pacing. A real recording may need a sound description, a speaker label, or more time on a difficult instruction. If the ruler covers the bottom area, test a different subtitle position before reducing the lettering. The source is quiet, so preserving its actual audio and choosing explicit mute are different delivery decisions even if both seem similar during a casual listen. After export, verify all three cues in the downloaded MP4 and confirm that the opening and ending gaps remain intentional.

Make room for the text without hiding the demonstration

Subtitle size is a percentage of the shorter output edge, not a fixed measurement of the preview panel. The available range is three to eight percent in half-percent steps. At five percent, a 720-pixel short edge gives a nominal 36-pixel font size, while a 1080-pixel short edge gives 54 pixels. Those calculations explain the control; they do not establish a universal reading standard. Judge the actual words at a realistic viewing size. Long technical terms, a moving background, and a viewer trying to watch the demonstration can all make the same nominal size feel less comfortable.

Choose text color, bold or normal weight, and background opacity together. White lettering may read clearly on a dark bench but disappear against a white paper pattern. A rectangle behind the words can maintain separation while the footage changes, although a heavy rectangle also hides more of the scene. Compare the lightest and busiest moments rather than approving only the opening frame. The system sans-serif treatment is intentionally straightforward. This is not a font-upload or animated-title workflow, and imported caption styling does not become a collection of editable graphic designs. Keep the treatment consistent enough that the instructional content remains the focus.

Top, center, and bottom placement let you move the text away from useful action. Margin is adjustable from zero to twenty percent of output height, while the subtitle area stays within a maximum of eighty-four percent of frame width. Text wraps to at most three lines. Overflow is rejected, not shortened with hidden words or an ellipsis. If a cue fails, split or revise its reviewed source text, choose another supported size, or reconsider the composition. Do not simply remove a necessary warning to make the box fit. Placement here is also not a simulation of every destination app’s buttons, overlays, or cropping behavior.

Choose the excerpt, then inspect its first and last cues

The subtitle track uses the original source timeline. Selecting a later excerpt does not mean you should first rewrite every timestamp to start at zero. Choose the source start and end, then let the tool map the visible portion into the exported clip. For example, an instruction running from source second 18.000 to 21.000, inside a selected interval from 16.000 to 24.000, appears from output second 2.000 to 5.000 when the offset is zero. This distinction prevents a common mistake: subtracting the clip start manually and then asking the tool to subtract it again during export.

A global offset shifts every cue by the same amount before the selected interval is applied. Positive values move subtitles later; negative values move them earlier. The permitted range is minus 300 to plus 300 seconds. Use a shared shift when several checked events show the same timing error. It does not stretch the track, recognize words, or repair an internal edit. If the first sentence matches but later sentences increasingly miss, return to the timing source and investigate. Repeatedly adjusting the offset until one isolated frame looks right can conceal a different problem elsewhere in the recording.

Cues crossing the excerpt boundaries are clipped to the selected interval, and cues wholly outside it are reported rather than shown. At least one cue must remain visible. Review a sentence that is shortened by either edge: the retained words may refer to a step the excerpt no longer contains, or remain visible too briefly to read. Consider moving the trim boundary to include the complete instruction instead of accepting an abrupt fragment. The normalized used-track SRT follows these decisions, with clip-relative timestamps and the offset already applied. It is a companion to this excerpt, not a replacement for the original whole-source track.

Set the MP4 output within the local processing limits

The source can be up to 150 MiB and five minutes long. The selected interval must last between 0.25 and 120 seconds. MP4, MOV, and WebM are possible source containers, but their internal codecs still need browser support. A familiar extension alone does not prove a recording can be decoded. The tool relies on native browser media capabilities, including the required WebCodecs support, so read an unsupported-format message as a real compatibility boundary. Try an appropriate supported preparation workflow instead of renaming the extension and expecting that to change the encoded video inside. The original file should remain untouched during any separate conversion.

Export produces H.264 MP4 at up to thirty frames per second. Long-edge choices are 640, 1280, and 1920 pixels, subject to a maximum of 2,073,600 output pixels. Smaller sources are not enlarged, and the tool does not crop or change the composition to fit a new shape. A vertical recording therefore stays vertical rather than becoming a square placement automatically. Choose a size that suits the receiving workflow, then inspect the resulting subtitle scale. The utility’s 64 MiB output ceiling is a processing constraint, not a statement about any social platform’s upload allowance or a guarantee that a destination will accept the file.

Keeping audio means the requested source audio must survive as AAC at 128 kbps and 48 kHz, or the operation must fail instead of silently delivering a muted result. Choose mute explicitly only when a silent copy is actually wanted. Neither successful video encoding nor visible subtitles prove that the sound is correct. Re-encoding is a new media generation and can change the picture or produce a file larger than expected. Compare the actual result rather than assuming that a lower dimension always means a particular file size. Browser memory and native codec availability also matter independently of the stated input and output byte limits.

Review moving pictures and the actual downloaded file

The frame preview and export use the same subtitle renderer, which makes the preview useful for inspecting wrapping, position, background, and size. Scrub through every cue, including the longest one and any text over a changing background. Check the start and end of each interval as well as its midpoint. A short label can fit beautifully while a later instruction overflows, and a correct sentence can cover the exact detail it explains. Shared rendering reduces one source of disagreement, but it does not replace reviewing the encoded video or guarantee identical playback on every physical device.

Next play the selection continuously with the audio decision in mind. Listen for a missing introduction, an abruptly cut word, or sound that no longer matches the instructional framing. Watch whether an instruction arrives before the relevant action and whether that lead is helpful or misleading. Then watch without relying on sound to identify any important information absent from the supplied track. The tool does not assess caption completeness for you. Speaker changes, relevant sound descriptions, and ambiguous references need editorial attention even when every imported cue passes syntax checks and every line fits the selected subtitle area.

Download the MP4 and open that file for a final pass. Verify the beginning, middle, ending, visible text, duration, and retained audio rather than treating a completed progress display as approval. Review at the intended destination when possible, because its player controls and presentation can differ from the local preview. There is no physical-device compatibility guarantee. If changes are needed, return to the saved source video and track, revise the appropriate input or setting, and create another delivery copy. Burning more text onto an already captioned MP4 would stack the old and new wording instead of replacing the original subtitle pixels.

Deliver the right files and preserve a route to revision

The finished MP4 contains permanent visible text. Viewers cannot switch those rendered words off, select a different caption language, or search them as an embedded closed-caption track. Keep the clean video and a separate reviewed caption file when those capabilities matter. The used-track SRT download contains only cues used in the selected clip, normalized to its new zero point. It can support a later handoff, but the receiving player still needs its own suitable caption workflow. Do not describe that sidecar as automatically embedded in the MP4 or assume that every social composer provides an SRT attachment field.

Save the settings JSON when you need to preserve the text, style, offset, and time choices. It deliberately includes subtitle text, so treat it as a content-bearing document rather than an empty styling preset. It does not contain the video or its source filename, and reopening it is not a substitute for retaining the matching media yourself. Review text, CSV, and PDF help record the preparation decisions. The PDF contains settings, review information, and notes, not the production video or a certified searchable, tagged caption deliverable. Keep file roles clear so an approver does not mistake a review document for media ready to share.

Finally, separate preparation from publication. A platform label gives the project a sharing context; it is not an uploader, crop tool, safe-zone guarantee, or acceptance validator. This utility has no native publishing or direct SMTasker import. Name the approved clip, its language, and its revision in your own delivery folder, then hand it to the person responsible for the destination account. Include the clean source and editable track where the brief requires them. A small, well-identified package is easier to correct later than one unexplained MP4 whose words, timing, and source recording can no longer be traced to their approved versions.

Questions before you add subtitles