Video Frame Extractor: MP4, TikTok, Instagram Reels to Images avatar

Video Frame Extractor: MP4, TikTok, Instagram Reels to Images

Pricing

from $1.40 / 1,000 frames

Go to Apify Store
Video Frame Extractor: MP4, TikTok, Instagram Reels to Images

Video Frame Extractor: MP4, TikTok, Instagram Reels to Images

Extract still frames from MP4, WebM and MOV links, TikTok videos and Instagram reels: every N seconds, N evenly spaced, at the times you list or at each scene change. One row per frame with its image link, timestamp, width and height. Max resolution, blank frame skip. No login.

Pricing

from $1.40 / 1,000 frames

Rating

0.0

(0)

Developer

The Mine Works

The Mine Works

Maintained by Community

Actor stats

0

Bookmarked

1

Total users

0

Monthly active users

8 days ago

Last modified

Features

Limited permissions

Share

11 frames from a video link, one row per frame with its image link, in 4 seconds

From The Mine Works, makers of Instagram Reel Scraper and B2B Leads Finder, with over 290,000 runs across 310+ public actors.

Updated 30 Oct 2026: first release. Direct MP4, WebM, MOV and MKV links, TikTok videos and Instagram reels; a frame every N seconds, N frames spread evenly, the exact times you list, or one per scene change; max resolution, JPG, PNG or WebP, blank frames stepped past. One row per frame with its image link. Bookmark this Actor on its Store page so it is one click away next time.

Why choose this actor?

  • TikTok and Instagram links work as they are. Paste a TikTok video link or an Instagram reel link next to your MP4 links. The actor finds the video file behind the post and downloads the smallest copy that reaches the size you asked for, so a 720 px frame from a reel does not pay for a 1080p download. In our test, 40 of 40 TikTok and Instagram links from four public profiles returned frames.
  • Long files cost only the parts you use. For interval, count and timestamps modes the actor reads only the stretches of the file around each frame. 10 frames of a 296 MB, 15 minute film read 24 MB in our test, not the whole file.
  • One row per frame, ready for a spreadsheet or an AI model. Every row has the image link, timestamp_sec and timestamp (00:01:23.500), frame_index, width and height, plus the video's length, size and codec. The links open without an Apify login, so you can pass them straight to GPT, Claude or Gemini.
  • No black frames by default, and you do not pay for them. A frame that is one flat colour (a fade from black, a cut to black) is swapped for the first frame with content up to 2 seconds later; if there is none, it is dropped and not charged. Links that cannot be read, the summary row and duplicates are never charged.

Run it on Apify

Part of The Mine Works Social media and video family: Threads Scraper, Reddit Scraper, Threads Search Scraper, Instagram Profile Scraper, Instagram Followers & Following, Reddit Search Scraper.

Try it in one minute

Paste this into the input (JSON view) and press Start. It returns one frame every 5 seconds from the 52 second Sintel trailer, 11 frames in all. It took 4 seconds in our test run, plus the few seconds Apify needs to start the run. The first frame of this trailer is black, so it comes from 2 seconds in and says so in stepped_from_sec:

{
"videoUrls": ["https://media.w3.org/2010/05/sintel/trailer.mp4"],
"mode": "interval",
"intervalSeconds": 5
}

Add more links on separate lines, mix MP4 files, TikTok videos and Instagram reels, and pick another mode below. Every frame is a row; open image_url to see it.

Apify's free plan includes $5 of credit every month, which covers about 2,500 frames at this actor's price.

Copy to your AI assistant

themineworks/video-frame-extractor on Apify. Turns video links into still images: direct video files (MP4, WebM, MOV, MKV, M4V), TikTok video links and Instagram reel or video post links, with no login. Returns one dataset row per frame with a public image link. Call ApifyClient("TOKEN").actor("themineworks/video-frame-extractor").call(run_input={"videoUrls": ["https://media.w3.org/2010/05/sintel/trailer.mp4"], "mode": "interval", "intervalSeconds": 5}), then client.dataset(run["defaultDatasetId"]).list_items().items. Give videoUrls (list of links, up to 200 per run; YouTube is not supported). Optional: mode (interval, count, timestamps or scenes; default interval), intervalSeconds (default 5), frameCount (count mode, default 10), timestamps (list such as "90", "1:30", "00:02:15.5", "1m30s"), sceneThreshold (scenes mode, default 0.3, lower finds more cuts), startTime and endTime (same time formats), maxFramesPerVideo (default 50, up to 1000), maxResolution (longest side in px, default 1280, 0 keeps the original), imageFormat (jpg, png or webp; default jpg), skipBlankFrames (default true; steps past one colour frames up to 2 s, not in timestamps mode), maxDownloadMbPerVideo (default 500). Frame rows have no _type field and carry video_url, page_url, platform, video_id, frame_index, timestamp_sec, timestamp, mode, scene_score, stepped_from_sec, image_url, image_key, image_format, width, height, image_bytes, video_duration_sec, video_width, video_height, video_codec, video_fps, video_title, author_username, scraped_at. Each run also writes one row with _type "summary", one with _type "info" when frames were delivered, and a row with _type "error" for each link that could not be read; none of these is charged. The images are records of the run's default key-value store (image_key). Full spec: GET https://api.apify.com/v2/acts/themineworks~video-frame-extractor/builds/default (Bearer TOKEN), which returns inputSchema and readme. Token: https://console.apify.com/account/integrations?fpr=ymnoit&utm_source=apify-readme&utm_medium=referral

Key features

  • Four ways to pick frames. interval: one frame every N seconds (0.5 gives two a second). count: N frames, each from the middle of an equal slice of the video. timestamps: the exact times you list, in seconds, clock time or 1m30s. scenes: the opening shot plus one frame at each cut, with a sensitivity you can set; over the cap, the strongest cuts are kept.
  • Start and end time. Limit any mode except timestamps to a part of the video, for example the first minute of a webinar or 5:00 to 10:00 of a match.
  • Max resolution on the longest side. 1280 gives 1280x720 from a landscape video and 720x1280 from a vertical TikTok. Frames are never enlarged; 0 keeps the original size. Phone videos are turned the right way up.
  • The right copy of a TikTok or reel. TikTok lists each video in several sizes and codecs, Instagram in several quality tiers up to 1080p. The actor takes the smallest one that reaches your max resolution, and the full 1080p copy when you ask for the original.
  • Accurate times. Each frame is cut at the time asked, not at the nearest keyframe. In scenes mode scene_score says how strong each cut was, from 0 to 1.
  • Safe on big and odd files. A per video download limit, hosts that do not allow partial reads (read once, front to back), links that open a web page instead of a file, and dead links each get a plain reason in an error row, and are not charged. A dead link costs one request; two videos in a row that a site refuses end the run, so a blocked run takes seconds.

How to use it

Thumbnails or a storyboard for each video

{
"videoUrls": [
"https://media.w3.org/2010/05/sintel/trailer.mp4",
"https://media.w3.org/2010/05/bunny/trailer.mp4"
],
"mode": "count",
"frameCount": 8,
"maxResolution": 640
}

Eight evenly spaced frames per video at 640 px. Good for a preview strip, a course catalogue or a quick look at many files.

TikTok and Instagram reels for an AI model

{
"videoUrls": [
"https://www.tiktok.com/@nasa/video/7700397098994273549",
"https://www.instagram.com/reel/Dd9Etk7R91I/"
],
"mode": "interval",
"intervalSeconds": 2,
"maxResolution": 768
}

A frame every 2 seconds is enough for a vision model to describe, tag or moderate a short video. Pass the image_url links to the model in the order of frame_index; video_title and author_username give it the post's caption and creator.

One frame per scene of an ad or a trailer

{
"videoUrls": ["https://download.blender.org/durian/trailer/sintel_trailer-1080p.mp4"],
"mode": "scenes",
"sceneThreshold": 0.3
}

In our test this returned 7 frames for the 52 second Sintel trailer: the opening shot and 6 cuts, each with its scene_score. Lower the threshold to 0.2 to catch soft cuts and dissolves.

Exact moments you already know

{
"videoUrls": ["https://archive.org/download/Sintel/sintel-2048-surround.mp4"],
"mode": "timestamps",
"timestamps": ["0:45", "3:12.5", "7m30s", "12:00"]
}

Times past the end of a video are skipped and not charged. In timestamps mode the frames are always at the exact times you gave, blank or not.

A long recording, one frame a minute

{
"videoUrls": ["https://archive.org/download/Sintel/sintel-2048-surround.mp4"],
"mode": "interval",
"intervalSeconds": 60,
"maxResolution": 1280
}

The actor reads only the parts of the file around each frame, so a long file is fast and light. Use startTime and endTime to cover part of it.

Input parameters

ParameterTypeDefaultWhat it does
videoUrlsarray of stringsrequiredDirect video file links, TikTok video links, Instagram reel or video post links. Up to 200 per run.
modestringintervalinterval, count, timestamps or scenes.
intervalSecondsnumber5Interval mode: seconds between frames, 0.1 to 3,600.
frameCountinteger10Count mode: frames per video, spread evenly.
timestampsarray of stringsemptyTimestamps mode: 90, 90.5, 1:30, 00:01:30.250 or 1m30s.
sceneThresholdnumber0.3Scenes mode: 0.05 to 0.95, lower finds more cuts.
startTime, endTimestringemptyPart of the video to use, same time formats. Not used in timestamps mode.
maxFramesPerVideointeger50Most frames per video, 1 to 1,000.
maxResolutioninteger1280Longest side of each image in pixels; 0 keeps the original size.
imageFormatstringjpgjpg, png or webp.
skipBlankFramesbooleantrueStep past one colour frames up to 2 s; drop and do not charge them if nothing shows.
maxDownloadMbPerVideointeger500Most video data read for one video, in MB.

What data do you get?

One row per frame. Frame rows have no _type field; the summary, info and error rows do, so a filter on _type separates them.

FieldWhat it is
video_urlThe link as you gave it
page_urlThe TikTok or Instagram page of the video (platform links)
platformdirect, tiktok or instagram
video_idTikTok video id or Instagram shortcode
frame_indexPosition of the frame in its video, from 1
timestamp_sec, timestampWhere the frame is, in seconds and as HH:MM:SS.mmm
modeHow the frame was chosen
scene_scoreScenes mode: strength of the cut, 0 to 1 (empty for the opening shot)
stepped_from_secSet when a blank frame was swapped for a later one: the time first planned
image_urlLink to the image
image_keyThe image's key in the run's key-value store
image_format, width, height, image_bytesThe image file
video_duration_sec, video_width, video_height, video_codec, video_fpsThe video file that was read
video_titleTikTok description, Instagram caption or the file name
author_usernameTikTok or Instagram username
scraped_atWhen the frame was cut

Summary row (_type: "summary", never charged): videos_requested, videos_read, videos_failed, frames_planned, frames_delivered, charged_for, frames_past_end, frames_over_cap, blank_frames_stepped, blank_frames_skipped, scene_cuts_found, downloaded_megabytes, ended_on_deadline, stopped_by_max_cost, shortfall_reason when fewer frames came back than planned, and per_video, one entry per video with its status, frames, size, MB downloaded and the reason when it failed. The same summary is saved as the OUTPUT record of the run.

Error rows (_type: "error", never charged): one per link that could not be read, with error_type (invalid_url, youtube_not_supported, not_a_video_file, not_a_video, not_found, refused, rate_limited, too_large, unreadable, no_frames, blocked, fetch_error), identifier, reason and http_status when the host answered.

Stable fields for automations

These were present in every frame row of our test runs, and their names will not change:

FieldWhat it is
video_urlThe link you gave
platformWhere the video came from
frame_indexOrder of the frame
timestamp_secTime in seconds
timestampTime as text
image_urlThe image
image_keyThe image record
width, heightImage size
image_bytesImage file size
video_duration_secVideo length
video_codecCodec of the file read

Optional: page_url, video_id and author_username (TikTok and Instagram only), scene_score (scenes mode), stepped_from_sec (only on frames that were moved past a blank frame), video_fps (when the file states it).

Output examples

A frame from a TikTok video, a frame every 3 seconds at 1280 px (test run, 30 Oct 2026; on Apify image_url carries the real store id and a signature):

{
"video_url": "https://www.tiktok.com/@nasa/video/7700397098994273549",
"page_url": "https://www.tiktok.com/@nasa/video/7700397098994273549",
"platform": "tiktok",
"video_id": "7700397098994273549",
"frame_index": 2,
"timestamp_sec": 3,
"timestamp": "00:00:03.000",
"mode": "interval",
"image_url": "https://api.apify.com/v2/key-value-stores/STORE_ID/records/7700397098994273549-408bde-0002-00003000ms.jpg",
"image_key": "7700397098994273549-408bde-0002-00003000ms.jpg",
"image_format": "jpg",
"width": 720,
"height": 1280,
"image_bytes": 150829,
"video_duration_sec": 16.903,
"video_width": 720,
"video_height": 1280,
"video_codec": "hevc",
"video_fps": 30,
"video_title": "NASA 🤝 college football. We heard y’all like a good air show. Roll Tide!",
"author_username": "nasa"
}

A frame from an Instagram reel in the same run (caption trimmed):

{
"video_url": "https://www.instagram.com/reel/DfHwGxbBD_K/",
"platform": "instagram",
"video_id": "DfHwGxbBD_K",
"frame_index": 2,
"timestamp_sec": 3,
"timestamp": "00:00:03.000",
"image_key": "DfHwGxbBD_K-4db3ae-0002-00003000ms.jpg",
"width": 720,
"height": 1280,
"image_bytes": 53842,
"video_duration_sec": 26.86,
"video_codec": "vp9",
"video_title": "To capture Jim Morrison's incredible accomplishment, the team was willing to go beyond ...",
"author_username": "natgeo"
}

A scene frame from the Sintel 1080p trailer, moved half a second past a cut to black (same day):

{
"video_url": "https://download.blender.org/durian/trailer/sintel_trailer-1080p.mp4",
"platform": "direct",
"frame_index": 7,
"timestamp_sec": 28.5,
"timestamp": "00:00:28.500",
"mode": "scenes",
"scene_score": 0.6823,
"stepped_from_sec": 28,
"image_key": "sintel_trailer-1080p.mp4-fd7bd5-0007-00028500ms.jpg",
"width": 640,
"height": 360,
"video_duration_sec": 52.208,
"video_width": 1920,
"video_height": 1080,
"video_codec": "h264",
"video_fps": 24
}

An error row for a YouTube link (nothing charged):

{ "_type": "error", "error_type": "youtube_not_supported", "identifier": "https://www.youtube.com/watch?v=aqz-KE-bpKQ", "reason": "YouTube links are not supported yet. Use a direct link to the video file, a TikTok video or an Instagram reel. Not charged." }

Pricing

Pay per event: you pay for the frames you receive, plus a flat $0.003 per run, whatever memory you choose. Scene detection, max resolution, blank frame checks and the platform lookups cost nothing extra.

EventWhen it is chargedFree planBronze (Starter)Silver (Scale)Gold (Business)
frame-extractedOnce per frame row stored in your dataset$1.80 per 1,000$1.70 per 1,000$1.55 per 1,000$1.40 per 1,000
run-startOnce per run, just before the first frames are delivered$0.003$0.003$0.003$0.003

The price of one frame is the price per 1,000 divided by 1,000.

Never charged:

  • links that cannot be read: YouTube, web pages instead of files, dead or private links, photo posts (each gets an error row);
  • frames that fail, blank frames that were dropped, and timestamps past the end of a video;
  • frames past your maxFramesPerVideo;
  • a video that a run already read (the same link twice);
  • the summary, info and error rows;
  • a run in which no frame is delivered: the start fee is charged only right before the first frames are delivered;
  • anything past your "Max cost per run". The actor checks the budget before it cuts the frames of each video and cuts only what the budget can pay for, so the run ends SUCCEEDED with a partial result and a status message saying so.

Run it on a schedule

  1. Open the actor in Apify Console, set your input and click Save as a new task.
  2. In the task, open Schedules, click Add schedule and pick how often.
  3. For a stream of new videos, start the actor from your own pipeline instead (Make, Zapier, n8n or the API) with the new links each time; a run with one video is fine.
  4. Optional: add an integration (Google Sheets, Slack, a webhook) so each run's frames land where your team works.

FAQ

Direct links to video files: MP4, M4V, MOV, WebM, MKV, MPEG TS and anything else ffmpeg reads, from your own storage (S3, R2, Google Cloud Storage), a CDN or a website. TikTok video links (tiktok.com/@user/video/..., vm.tiktok.com and vt.tiktok.com share links) and Instagram reel or video post links (instagram.com/reel/..., instagram.com/p/...; each video of a carousel is read). Google Drive and Dropbox share links open a web page: use their download link (Dropbox: add ?dl=1).

Does it work with YouTube?

Not in this version. A YouTube link gets an error row and costs nothing.

Do you download the whole video?

Not for long files in interval, count and timestamps modes: the actor reads only the parts around each frame. Files up to 24 MB, hosts that do not allow partial reads, and scene mode read the whole file once. downloaded_megabytes in the summary shows what each video took.

Why are some frames moved by a second or two?

With skipBlankFrames on (the default), a frame that is one flat colour, such as the black at a fade in or a cut, is replaced by the first frame with content up to 2 seconds later, never past the next frame. Those rows carry stepped_from_sec. Turn the option off to get every frame at its planned time. Timestamps mode never moves a frame.

How many frames can I get?

Up to 1,000 per video and 200 videos per run. count never returns more frames than you ask for; interval and timestamps keep the first maxFramesPerVideo; scenes keeps the strongest cuts.

How long does a run take?

In our tests on a laptop: 4 seconds for the "Try it" input, 3 seconds for scene detection on the 52 second 1080p Sintel trailer, 39 seconds for 10 frames from a 296 MB film on a slow office line, and 325 seconds for 613 frames from 40 TikTok and Instagram videos on the same line. On Apify the default 1 GB of memory comes with a quarter of a CPU, so cutting frames takes longer than on a laptop; give the run more memory for a faster run. Compute is billed on memory times time, so a faster run costs about the same. The default time limit is 3,600 seconds; the actor stops before it and keeps every frame.

The images are records of the run's default key-value store and are kept as long as Apify keeps that run's storage (see your plan's data retention). Download them, or copy them to your own storage, if you need them for longer.

Do I need an account, cookies or a proxy?

No. TikTok and Instagram links are read over Apify Proxy, which is included. Direct links are read straight from their host.

Which formats can I download?

Rows as JSON, CSV, Excel, XML, RSS or HTML from the run's Storage tab or the API. The images are JPG, PNG or WebP files in the run's key-value store; the dataset's Frames view shows each one in the table.

Can I use it from Claude, ChatGPT or Cursor (MCP)?

Yes, through Apify's MCP server. Connector URL: https://mcp.apify.com/?tools=themineworks/video-frame-extractor

  • Claude: Settings > Connectors > Add custom connector, paste the URL, sign in with Apify.
  • ChatGPT: turn on developer mode, add an MCP connector with the URL, sign in with Apify.
  • Cursor or VS Code: add it as an HTTP MCP server with that URL.
  • Claude Code: claude mcp add -t http video-frame-extractor "https://mcp.apify.com/?tools=themineworks/video-frame-extractor"

The actor reads only public videos, as anyone can see them without logging in. You are responsible for having the right to process each video you submit, including the terms of TikTok and Instagram, copyright and data protection law such as GDPR and CCPA. This is general information, not legal advice.

Integrations

  • Google Sheets: send each run's rows to a sheet; image_url shows as a link per frame.
  • Make, Zapier and n8n: start a run with new video links and read its dataset, for example "frames of each new upload to our moderation queue".
  • Webhooks: have Apify call your URL when a run succeeds, then fetch the rows and images.
  • API and SDKs: run it from Python or JavaScript with the Apify client, or with a plain HTTP call (see "Copy to your AI assistant" above).
  • MCP clients: Claude, ChatGPT, Cursor and VS Code can run it as a tool through the connector URL in the FAQ.

More from The Mine Works

Social media and video

Leads and business directories

Marketing, SEO and reviews

LinkedIn

Real estate

Science, health and government data

Jobs and hiring

Company and business data

E-commerce and marketplaces

Food and local services

Developer and AI tools

More tools

Support

Found a link that should work and does not, or a frame at the wrong time? Open an issue in the Issues tab with the input and the run ID, and we will look at it. For a mode or a source you need, email dmineworks@gmail.com.

Paste MP4 links, TikTok videos or Instagram reels and get still frames back as JPG, PNG or WebP, every N seconds, at set times or at each scene change, one row per frame with its image link, no login.