SRT to MP4: Match Subtitle Timing to Your Clip

Use Add Subtitles to Video when you need those existing words permanently visible in an MP4. This practical guide works through a map-folding recording with a twelve-second excerpt, including the boundary cues that a simple timestamp subtraction can overlook.

Find out what the SRT was timed against

Before converting anything, establish the relationship between the two inputs. An SRT may describe the full camera recording, a shortened edit, or a previously exported clip whose clock begins at zero. Those are different references even if the visible sentences are identical. Ask which exact video version accompanied the track and check several recognizable events. A file named final-subtitles does not settle the question. If a five-second title was removed after the captions were prepared, their timestamps still refer to the earlier edit until somebody deliberately corrects them. Keep the supplied file unchanged while investigating so the original evidence stays available.

An SRT cannot become meaningful moving footage on its own. It supplies cue intervals and text; the local video supplies the pictures and any audio you request to preserve. The output MP4 receives visible subtitle pixels rather than a newly selectable caption channel. If the delivery only needs an editable caption attachment, burning the words may be the wrong operation. Confirm that requirement before spending time on style. It is possible to deliver both a captioned viewing copy and a clean video with a sidecar track, but those files should have separate names and a clearly stated purpose for the recipient.

For this worked example, imagine a sixty-second recording demonstrating how to fold a printed map. The existing SRT was prepared against that full recording. The useful excerpt runs from source second 42.000 through 54.000, lasting twelve seconds. Independent checks show that every subtitle needs to move 1.250 seconds later. These assumptions are deliberate teaching facts, not values inferred by the tool. Write them in the delivery notes: full-source track, selected source interval, and verified global correction. With those three items fixed, the arithmetic can be checked without guessing whether a timestamp already includes a trim or a previous timing adjustment.

Read timestamps as intervals, not as frame numbers

An SRT timestamp uses hours, minutes, seconds, and milliseconds. The notation 00:00:43,250 means 43.250 seconds from the track’s reference zero. For arithmetic, convert a boundary to integer milliseconds: hours times 3,600,000, plus minutes times 60,000, plus seconds times 1,000, plus the final millisecond value. That representation avoids confusing the comma with a decimal separator in a spreadsheet or treating the last three digits as frames. A WebVTT timestamp uses a period for the fractional part, but the elapsed-time meaning remains the same. Neither notation tells you which video revision supplied the reference clock in the first place.

Each cue has a start and an end, and its duration is end minus start. A cue from 43.000 to 46.500 lasts 3.500 seconds. In this tool, text is active at its starting boundary and inactive at its ending boundary. Two cues can therefore meet at the same timestamp without being displayed simultaneously. A true overlap, where the next cue begins before the previous one ends, is rejected for this workflow. Do not solve that by blindly sorting or deleting rows. Determine whether the source contains simultaneous speakers, an accidental duplicate, or an actual timing mistake before preparing a sequential working track.

Import a basic UTF-8 track within this tool’s 64 KiB, 200-cue, and 500-character-per-cue boundaries. Those limits do not describe all possible SRT files or every destination’s caption rules. Unsupported structured WebVTT blocks such as STYLE, REGION, and NOTE need a different source preparation. Formatting or positioning losses in otherwise accepted input require a reviewed plain-conversion acknowledgement. Check the visible words before doing the calculations: a damaged character or missing speaker label is not a timing issue. It is better to discover that the wrong language file was loaded before producing a perfectly calculated subtitle schedule for the wrong text.

Apply a verified offset once, in the correct direction

The offset relationship is shifted boundary = original boundary + offset. In the map example, the offset is +1.250 seconds, so an original cue from 43.000 to 46.500 becomes 44.250 to 47.750 on the source timeline. Both boundaries move by the same amount. The cue still lasts 3.500 seconds, and every gap between cues is unchanged. A positive correction makes text appear later, which is appropriate when the supplied subtitles lead the event. A negative correction moves text earlier. State the direction in ordinary words before entering the number so a familiar-looking minus sign does not reverse the intended repair.

Choose the correction from matching events, not from the first cue’s timestamp alone. Listen or watch at several points and compare the same identifiable word or action with its supplied cue. In a spoken recording, a pause can make a vague sentence beginning difficult to locate; choose a clearer reference instead of claiming more precision than the evidence supports. For the fictional map recording, assume three checked references all show the same 1.250-second lead. That supports one global shift. The utility applies your chosen offset but does not recognize the audio, detect those reference points, or decide that they match.

The supported offset range is minus 300 to plus 300 seconds. An earlier shift can move a cue partly before video zero. For example, 0.400 to 1.400 with an offset of -0.750 becomes -0.350 to 0.650. If the selected interval begins at zero, only 0.000 to 0.650 can remain visible. A cue moved wholly before the interval contributes no visible text. Review the resulting notes and meaning rather than expecting negative time to become hidden pre-roll. Also check whether the file was already corrected elsewhere: applying +1.250 here after saving that same correction into the SRT would shift the captions twice.

Map the twelve-second excerpt with all its boundary cases

Let A be the selected source start, 42.000, and B the selected end, 54.000. After applying the offset, a cue with boundaries S and E has a visible start of max(S, A) and a visible end of min(E, B). Keep it only when the visible end is greater than the visible start. Then subtract A from both visible boundaries to obtain clip-relative times. This order matters: shift first, intersect with the selected interval second, and establish the new zero third. It handles partial cues without pretending their complete original intervals fit inside the excerpt. The output duration is B minus A, or 12.000 seconds.

The first relevant original cue, “Open the map fully.”, runs from 39.500 to 42.000. Adding 1.250 produces 40.750 to 43.250. Its intersection with the selected interval is 42.000 to 43.250, so the used SRT places it at 0.000 to 1.250. The next cue, “Match the two marked corners.”, originally runs from 43.000 to 46.500. Its shifted interval is 44.250 to 47.750, wholly inside the selection, so its output times are 2.250 to 5.750. That middle cue still lasts 3.500 seconds. Notice the one-second gap after the opening cue: trimming changes the reference zero, not the spacing between surviving events.

The last relevant instruction, “Fold along the center line.”, originally runs from 50.500 to 54.000. After shifting, it occupies 51.750 to 55.250, and the selected end clips it to 51.750 to 54.000. Its output interval is therefore 9.750 to 12.000. An earlier cue from 37.000 to 39.000 ends at 40.250 after shifting and is omitted; a later cue from 55.000 to 56.000 starts at 56.250 and is also omitted. These five nonoverlapping source cues yield three visible output cues. The arithmetic is valid, but the shortened first and last instructions still need a human reading and context check.

Avoid subtracting the trim or adding the offset twice

The used-track SRT is already a transformed delivery artifact. In the example, its first visible cue begins at 0.000, not at the original 39.500 or shifted 40.750. Pair that SRT with the newly exported twelve-second video, begin the selection at that video’s zero, and use zero additional offset unless a new verified reason requires otherwise. Do not load the used track against the sixty-second source and keep the original 42.000-to-54.000 selection unchanged. Its zero-based cues would refer to the wrong part of that source. The word used identifies a relationship to one rendered excerpt, not a universally reusable timing reference.

Adding the selected start back to a used cue recovers its clipped, shifted source interval, not necessarily its original interval. For the first instruction, 0.000 to 1.250 plus 42.000 gives 42.000 to 43.250. Subtracting the 1.250 offset then gives 40.750 to 42.000, which still does not recover the original start of 39.500. That information was removed by clipping at the excerpt boundary. This is why the normalized sidecar cannot replace the original track in an archive. Its timestamps faithfully describe what was used, but they do not preserve every source cue or every discarded portion of a retained cue.

Keep a simple naming distinction outside the tool: map-full-original.srt, map-excerpt-used.srt, and map-excerpt-captioned.mp4, with a revision suffix when needed. The names are your own archive convention, not a filename feature that the tool guarantees. Record whether an offset has already been baked into each track. If a colleague sends an SRT starting at zero, ask whether it belongs to a trimmed video rather than assuming the full recording starts with those words. A quick comparison of two recognizable events often resolves the ambiguity faster than trying several large offsets until one sentence happens to line up with something on screen.

Separate constant delay from drift and internal edits

A constant offset cannot correct a changing timing relationship. Suppose a caption reference at track second 10.000 matches video second 10.500, while a later reference at track second 110.000 matches video second 112.500. The first discrepancy is 0.500 seconds and the second is 2.500. No single shift satisfies both. A possible linear relationship has scale (112.500 – 10.500) / (110.000 – 10.000) = 1.020 and offset 10.500 – 1.020 times 10.000 = 0.300. That is a diagnostic calculation for a separate retiming workflow, not an extra transformation this burn-in tool silently applies when you enter its global offset.

Under that example relationship, track second 60.000 would map to video second 61.500. Check a real middle reference before accepting the model. Two matching endpoints do not establish that all intervening edits share a constant rate. A deleted section, speed ramp, or rearranged sequence can create local changes that a straight-line mapping cannot explain. Scaling also changes cue durations, unlike a simple shift. Use the Subtitle Editor and sync fixer when an existing track needs deliberate timing repairs before rendering. Preserve the repaired track separately and begin its burn-in review with a clear record of any correction already incorporated into its timestamps.

A sudden mismatch after one cut calls for another investigation. If captions align until a three-second pause is removed, later cues may need a local correction while earlier cues stay unchanged. Moving the entire file would damage the portion that already matches. Likewise, a track prepared for a different take can contain correct sentences in a different order, which no offset resolves. Compare the video edit itself, not only the numerical schedule. After repairing the appropriate source sections, replay the transition through each edit. A syntactically tidy sequence can still pair an instruction with the wrong demonstrated action or leave a relevant sound description behind.

Check what the mathematics does to reading time

Clipping reduces visible duration without automatically rewriting the sentence. The first map instruction originally lasted 2.500 seconds but remains on screen for only 1.250 seconds in the excerpt. The final instruction retains 2.250 seconds of its original 3.500-second duration. Ask whether the viewer can read the complete phrase while understanding the action now visible. A mathematically valid boundary is not an editorial endorsement. You might move the excerpt start earlier, extend its end, or revise the approved cue structure. Do not hide a missing opening step behind a subtitle that assumes the viewer already saw it just before the selected interval.

Readability also depends on the frame geometry. The tool allows a font size of three to eight percent of the shorter output edge, with top, center, or bottom placement and an adjustable margin. Cue text wraps into no more than three lines within the allowed subtitle area. Overflow stops the operation rather than dropping the last words. Test the longest retained cue at the selected output dimensions, not only at an enlarged browser window. Splitting a long cue needs a meaningful text division and appropriate timing in the existing track; it is not solved by squeezing necessary instructions into tiny lettering.

Preview uses the same text renderer as export, which helps check wrapping and placement, but exported video still consists of discrete frames. A timestamp expressed to the millisecond does not promise that a viewer will observe a transition on a frame at that exact millisecond. Export is limited to up to thirty frames per second, and this workflow does not promise frame-exact trimming. Scrub around the boundaries, then play through them at normal speed. Look for a flash of text at the opening, a final instruction cut off mid-thought, or an awkward empty interval that made sense only in the full recording.

Verify the MP4 and keep the timing evidence together

Use a supported local MP4, MOV, or WebM source up to 150 MiB and five minutes long. The selected clip must last 0.25 to 120 seconds, so the twelve-second example fits the duration boundary without saying anything about a destination’s upload rules. Native browser codec support remains necessary. The output is H.264 MP4, no larger than 64 MiB, with a maximum 1920-pixel edge and 2,073,600 pixels. There is no crop or upscale. If preserving sound, requested audio must be retained as AAC at 128 kbps and 48 kHz or the conversion fails; explicit mute is a different choice.

Open the actual downloaded MP4 and compare it with the used SRT. In the worked example, verify the opening text, the 2.250-to-5.750 middle instruction, and the 9.750-to-12.000 final instruction. Confirm that the omitted source cues do not appear. Check the selected content and audio as well as the caption timing. The sidecar does not make the burned text toggleable, and this tool does not embed a selectable closed-caption channel. A separate player may use the sidecar with a clean matching clip when its own caption workflow supports that arrangement. Review that destination separately instead of treating a successful local export as acceptance everywhere.

Archive the original video, original SRT, corrected working track if one exists, final MP4, and normalized used SRT with a short note explaining their relationship. Settings JSON preserves transcript text and chosen settings, but not the video or source filename; review PDF and text or CSV records are supporting evidence, not replacement media. Treat those text-bearing files according to their contents when sharing. The browser session has no autosave and no automatic publication step. A clear handoff should let the next person identify the timeline zero, offset history, selected source interval, and approved revision without recreating your calculations from an unexplained exported file.

Questions about SRT to MP4 Timing

Continue with another subtitle preparation guide

Prepare your own clip with the Add Subtitles to Video, or continue with a focused guide. Find more utilities in all free tools.