Video Frame Extractor & Contact Sheet

Use your original local MP4, MOV, or WebM file. Video processing happens in your browser without a social login or private video upload. This tool does not download videos from links, recognize scenes, predict winning images, or post to a platform. Find related preparation utilities in the free tools directory.

Choose the output before collecting the frames

Start by deciding what someone needs to do with the images. A repair note may need one clear view of a part before removal and another after replacement. A workshop record may need several stages together for comparison. An editor may want a few source stills to consider as illustration candidates. These jobs need different selections, even when they use the same recording. Write the intended use first so you do not mistake a large capture set for a useful finished result.

Think of capture and selection as separate decisions. Capture asks the browser to decode images near the requested source times. Selection decides which of those candidates belong in the current sheet and exports. You can collect a wider set for review and retain a smaller set for a handoff. Deselecting a candidate does not alter the recording, erase a source frame, or improve the quality of the candidates that remain. It only changes the set you intend to use.

Choose clean individual PNGs when each image needs room or will be placed elsewhere. Choose a contact sheet when the relationship between selected views matters more than full-size inspection of one view. Choose a review package when another person needs pictures together with source-time records and settings. A contact sheet is a single comparison image, not a video edit, a slideshow with playback timing, or a simulation of how multiple attachments will appear on a social platform.

Keep the promise of the selection proportional to what stills can show. They can document a visible state, help someone locate a recorded moment, and make alternatives easier to compare. They omit movement and audio between those states. They do not establish an unseen cause, verify a spoken claim, or measure audience response. If the important explanation depends on a continuous movement, retain the original recording as the reference rather than trying to make isolated endpoints carry the whole account.

Prepare a compatible source without losing its identity

Choose an original local MP4, MOV, or WebM recording that you are ready to inspect. The file extension describes a container, not a guarantee that the current browser can decode its video. Supported AVC, VP8, or VP9 footage still depends on available decoding support. Use SDR material within the tool’s checks. If a file is rejected, read the specific reason and prepare a suitable local copy in another editor; renaming an extension does not convert the encoded video.

The input limits are 100 MiB, a duration from 0.5 seconds to 10 minutes, display and coded edges no larger than 3840 pixels, no more than 8.3 megapixels, and an aspect ratio no wider or taller than 4:1. These bounds describe this utility, not a platform upload specification. A source outside them may need trimming, resizing, or conversion elsewhere. Keep the original separately and identify which prepared version you actually used for the frame selection.

Source identity matters because a different trim changes the timeline. A request at 8.00 seconds in a shortened excerpt can refer to a different moment than 8.00 seconds in the full recording. Keep a small project note describing the take or revision without putting private details into a public sheet title. When reproducing a selection later, reselect the matching source rather than assuming another file with a similar scene or duration is interchangeable with the original.

Review the source for visible private information before generating shareable images. A fleeting notification, a paper label, or a reflected screen can become readable in a still. This tool does not redact, remove backgrounds, replace faces, or clean confidential text. Local processing means the selected video is not uploaded for extraction; it does not mean the page loads no normal site resources, and it does not make the exported image contents harmless. Choose a safely prepared source before continuing.

Use even intervals to survey an unfamiliar clip

Even-interval mode helps survey a range before choosing meaningful moments. Set the start and end, then choose between one and twelve captures. A single capture uses the start only. Begin with enough points to understand the recording, not necessarily the maximum. In a short equipment inspection, a survey might reveal the front, rear, open compartment, and final assembled view. It cannot judge which view matters, but it helps locate regions worth inspecting more closely.

The end value is exclusive. With a start of 0.00 seconds, an end of 3.00 seconds, and four captures, the requested times are 0.00, 1.00, 1.99, and 2.99 seconds. For multiple captures, the first request is at the start; the last is one hundredth of a second before the end, with intermediate points spread evenly. Do not expect a request exactly at 3.00 or interpret the range as four consecutive source frames. These are four time-based requests across that range.

An evenly distributed set can include motion blur, a transition, or a view partly covered by someone’s hand. That is not proof that the source lacks a better image nearby. Use the survey to decide where to look next, then switch to specific times around a settled view. Save any survey export you need before changing the capture parameters. A new range, count, or capture size creates a new result, not an additional permanent layer of images stored behind the current gallery.

Choose the longest output edge according to the detail you need: 640, 960, or 1280 pixels. The tool does not upscale, so a smaller source cannot supply extra detail merely because a larger option is chosen. The complete capture set must remain within 24 million output pixels and 40 MiB of combined PNG data. These processing bounds may require a smaller size or fewer captures. They are limits for managing the work, not guarantees about completion time on every device.

Request specific moments and read the returned frame times

Use specific-times mode when you already know which source moments matter. Enter seconds one per line with up to two decimal places, in chronological order, keeping the set between one and twelve requests. A single custom time is valid. For a maintenance record, you might request 1.25 for the original assembly, 7.80 for the exposed part, and 14.50 for the reassembled result. These numbers illustrate a plan, not values that will produce the same stages in another recording.

Duplicate custom request values are refused. However, two different request values can still return the same decoded source frame, especially when source frames are sparse. Two separate frames can also look identical because nothing moved. Inspect the returned images instead of treating an accepted list of times as a guarantee of unique visual moments. If a set contains redundant candidates, either choose a more useful time and recapture or deselect the extra image from the exports you are preparing.

The requested time is where you asked to look. The actual decoded frame start is when the returned source image begins on the presented source timeline. A request can fall between recorded frame starts, so these values need not match. For illustration, a request at 7.80 seconds might return an image beginning at 7.77 seconds. Read the real record for your file; that example is not a fixed correction to add or subtract from every request.

Keep both values when someone needs to trace a selection. They are relative to the presented source origin, not wall-clock time or a promised exact frame index. The extractor does not interpolate a new image at the requested instant or automatically identify a scene boundary. If the returned image precedes a useful state, request a slightly later point and inspect again. Use actual visible content to approve the image, and preserve the time record to explain how it was obtained.

Review candidates before committing to a selection

Open each candidate and ask whether it performs its intended job. A before image should show the relevant starting state, not merely a similar-looking earlier view. An after image should show the settled result, not a moment while the part is still moving. Compare framing, visibility, and any labels you expect someone to read. A technically successful extraction can still be an unhelpful illustration. The judgment about meaning stays with the person who understands the recording and the intended explanation.

Select the candidates that belong in the current output and deselect those that are redundant, obscured, or irrelevant. You do not need another video decode to revise this selection. Keep a fuller internal review set only if the extra views answer an actual question. A public image sequence may need fewer pictures, while an inspection handoff may need more context. Neither use requires you to pretend that the omitted states never occurred or that the chosen states represent every outcome.

Separate a source problem from an export problem. If the camera never showed a connector clearly, more columns or a larger PNG setting will not reveal its hidden contacts. If the source is clear but the sheet tile is too small, use an individual PNG or fewer columns. If the requested point catches movement, choose a settled moment. This diagnosis avoids repeating the same capture with settings that cannot fix the actual reason the image is difficult to interpret.

Before replacing the capture set, export any version another person is already reviewing. A changed request list should not silently inherit an earlier approval. Keep a brief note about the purpose of the selection and any important omissions, such as a covered label or missing intermediate stage. The extractor does not score faces, rank scenes, or predict performance. Your selection should be defensible because its images communicate the intended information, not because a count, timestamp, or file format appears to validate them.

Build a contact sheet without changing the source selection

A contact sheet arranges selected captures into one PNG for comparison. Request two, three, or four columns, with three as the default, and add an optional title up to 80 characters. The sheet uses no more columns than selected images, so a single image gets one effective column. Use a title identifying the review subject, not an unearned approval. Titles are measured rather than silently shortened; revise one if needed. Keep confidential project names out of sheets intended for wider sharing.

Sheet layout and capture are independent. Changing columns, title, or which existing candidates are selected does not require decoding the video again. It also does not create new source images or alter their recorded times. If a view is missing, return to capture rather than expecting a layout change to reveal it. This distinction is useful during review: you can try a more readable arrangement without changing the underlying evidence that a colleague is comparing or discussing.

Time labels belong on the contact sheet. Individual PNG frame downloads are clean, without those labels or a compulsory advertising watermark. Use the sheet when a reviewer needs a compact overview, and the clean images when detail or separate placement matters. The PNG sheet is bounded at 12 megapixels and 6000 pixels in height. If a large arrangement is refused or becomes hard to read, reduce the selected set or change the layout rather than assuming every candidate must fit in one montage.

Inspect the completed sheet at a realistic viewing size. Small text in the source may be readable in an individual capture but not in a multi-column overview. Give important views their own files when needed and keep explanatory text alongside them. The sheet is not a native social preview and does not guarantee an attachment order, crop, or video-cover assignment on any destination. Prepare a separate crop or padded asset with the image resizer only after preserving the clean extracted source.

Match each export to the person receiving it

Download individual PNGs for image use and a PNG contact sheet for visual comparison. Choose ZIP when the recipient needs the selected PNGs with the manifest, settings, and notes together. The ZIP is limited to 64 MiB, so a large package may need a smaller selection or output size. Do not describe a bundle as a copy of the original recording: it is a package of chosen stills and supporting records, not a video backup or an edited clip with sound.

CSV and TXT provide companion records that can be read and handled separately from the pictures. Use them to preserve the distinction between requested seconds and actual frame starts and to carry written review context. JSON settings preserve capture parameters and chosen times, not video, image pixels, or the private source filename. Reopening settings requires reselecting the matching local source. This lets you resume the setup, but it does not make a settings file into a self-contained archive of captured images.

The PDF is a review document with selected still-image content, explanatory material, and the complete settings and manifest attachment. Its pages are rasterized. Do not assume its visible text can be selected, searched, or read as tagged document structure. Supply TXT or CSV when accessible companion text or searchable records are important. A complete attachment preserves supporting data, but it does not transform the raster pages into an accessible text document or replace a recipient’s need for the actual images.

Review the files that contain pictures before sharing them. PNGs, the contact sheet, ZIP, and PDF can reveal details from the source even though the settings exclude video and pixels. Open the exported result and inspect labels, screens, reflections, and identifying details at full size. Keep distribution limited to the intended audience for that content. A local workflow, a neutral title, or a JSON file without a private filename does not remove information already visible in an exported still image.

Finish with a reproducible handoff and a separate next step

Before leaving the page, save the outputs you need and keep the matching source in your own project storage. There is no autosave to rely on as a backup of the session. Give a collaborator the selected images, the relevant time record, and a sentence explaining the decision they are being asked to make. For example, ask whether the maintenance note clearly distinguishes the original and reassembled states. That is more actionable than sending a contact sheet labeled final with no review instruction.

If processing is cancelled, times out, or rejects a combination of settings, treat the message as a boundary of the current operation, not a verdict on the recording’s usefulness. Reduce the capture count or size where appropriate, or prepare a compatible source elsewhere when the file itself is unsupported. Check the completed gallery and actual exported result after another attempt. Do not assume a partial or earlier preview represents the new set, and do not promise a particular processing speed to someone using another device.

For social use, attach and review the final assets in the destination yourself. Platform contexts can help you plan a useful explanation, but native placement, cropping, and cover controls vary. The extractor neither posts nor guarantees performance. Use the media specs checker for a separate technical preparation check when relevant. Keep a caption or image description that explains the selected content; a timestamp alone rarely tells a new reader why a particular source moment matters.

SMTasker is an optional next step for supported account routines, not part of the extraction engine or an importer for this settings JSON. It uses a Windows or macOS client with a connected Android device. Its built-in Skye manager requires your own OpenAI or Anthropic key, with provider usage billed separately, and external MCP access is another option. Review those operational capabilities independently after the image handoff is complete. A five-day trial is for assessing the workflow, not a promise of account growth or automatic publishing everywhere.

Questions about extracting video frames