Video Frame Extractor: Thumbnails, Scene Keyframes, Contact Sheet & GIF

Video Frame Extractor turns video files into still images: a frame every N seconds, N evenly spaced thumbnails, or one keyframe per scene change. Add a contact sheet (one grid image of all frames), the video metadata (duration, resolution, FPS, codecs), the audio track as M4A, MP3 or Opus, and a short animated GIF clip, all in one run. Paste direct video links or point it at your own uploads. Everything runs with FFmpeg inside the actor: no API key, no external service.

• ✅ Interval frames: one every N seconds, or exactly N frames spread over the whole video.
• ✅ Scene-change keyframes: one frame at the start of each shot, with adjustable sensitivity and minimum scene length.
• ✅ JPEG, PNG or WebP, with quality and max width; frames are saved upright (phone rotation applied).
• ✅ Contact sheet, GIF clip and audio track (M4A, MP3, Opus; copied without re-encoding when it is already in that codec) as optional extras.
• ✅ Video metadata for every file: duration, width, height, FPS, video and audio codecs, bit rate, rotation, audio channels.
• ✅ Any common video: MP4, MOV, MKV, WEBM, AVI, MPEG-TS, FLV, 3GP; H.264, HEVC, VP8/VP9, AV1, ProRes and more.
• ✅ Batch-safe: a 404, a web page, a corrupt or too-long file never fails the run. You get an item with the reason, for free.
• ✅ Pay per result: $0.005 per video + $0.003 per frame. No start fee, compute included.

Start in one minute

Click Start with the prefilled input:

{
  "urls": [
    "https://images-assets.nasa.gov/video/KSC_69-71212-sRGB/KSC_69-71212-sRGB~small.mp4"
  ],
  "mode": "interval",
  "frameCount": 12
}

You get: 12 evenly spaced JPEG frames from an 84-second 1969 NASA launch film, with frameUrls, durationSeconds, width, height, fps and costUsd.

Cost: $0.041: 1 video × $0.005 + 12 frames × $0.003. No start fee.

Use with an AI assistant

Connect the Apify MCP server to Claude, ChatGPT, Cursor or any MCP client (https://mcp.apify.com?tools=cprussin/video-frame-extractor loads only this actor), then paste:

Use the Apify actor cprussin/video-frame-extractor to extract one frame per scene from https://images-assets.nasa.gov/video/KSC_69-71212-sRGB/KSC_69-71212-sRGB~small.mp4, and describe what happens in each scene.

Recipes

Thumbnails and a contact sheet

12 WebP thumbnails plus one grid image of them all.

{
  "urls": [
    "https://images-assets.nasa.gov/video/KSC_69-71212-sRGB/KSC_69-71212-sRGB~small.mp4"
  ],
  "frameCount": 12,
  "frameFormat": "webp",
  "maxWidth": 960,
  "contactSheet": true
}

One keyframe per scene

A frame at each scene change, up to 30.

{
  "urls": [
    "https://images-assets.nasa.gov/video/KSC_69-71212-sRGB/KSC_69-71212-sRGB~small.mp4"
  ],
  "mode": "scene",
  "sceneThreshold": 35,
  "maxFrames": 30
}

A GIF preview

A 5-second animated GIF from 0:10, no frames.

{
  "urls": [
    "https://images-assets.nasa.gov/video/KSC_69-71212-sRGB/KSC_69-71212-sRGB~small.mp4"
  ],
  "mode": "none",
  "gifClip": true,
  "gifStartSeconds": 10,
  "gifDurationSeconds": 5
}

The audio track as MP3

Video metadata and the audio track, no frames.

{
  "urls": [
    "https://images-assets.nasa.gov/video/KSC_69-71212-sRGB/KSC_69-71212-sRGB~small.mp4"
  ],
  "mode": "none",
  "audioFormat": "mp3"
}

What can I use a video frame extractor for?

• Thumbnails and previews: pick cover images for a video library, CMS or e-commerce product videos.
• AI and computer vision: feed frames to a vision model (GPT-4o, Claude, Gemini) for captioning, tagging, moderation or search.
• Video summaries: one keyframe per scene gives a storyboard of a talk, ad or tutorial; the contact sheet shows it at a glance.
• Datasets: sample frames from your own footage for training or labeling.
• Social content: grab stills and a GIF from your own clips, and the audio track for a podcast or transcript.
• Automation and AI agents: add a "get frames from this video" step to Make, Zapier, n8n or an MCP-enabled agent through the Apify API.

How does it work?

1. Each video is downloaded (with size and time limits), or read from your key-value store. Web pages, streaming playlists and social links are refused for free.
2. FFprobe reads the metadata. Videos over Max video length are skipped (free).
3. Interval mode seeks straight to each timestamp, so only a few frames are decoded per image. Scene mode decodes the whole video at low resolution and measures how much each frame differs from the previous one (FFmpeg's scene score); a jump over the threshold starts a new scene.
4. Frames are scaled to Max width and saved as JPEG, PNG or WebP in the run's key-value store. The contact sheet, GIF and audio track are made from the same video.
5. One dataset item per video lists every frame (time, timecode, link, size), the extras and the metadata.

How do I extract frames from a video?

Field	Description	Default
urls	Direct links to video files, one per line	a NASA video
startUrls	Same, in Apify request-list format (combined with urls)	none
keyValueStoreRecords	Uploaded videos, as storeId/recordKey or username~store-name/recordKey	none
mode	interval, scene (scene changes) or none (metadata, GIF and audio only)	interval
intervalSeconds	Interval mode: one frame every N seconds from 0:00	10
frameCount	Interval mode: N evenly spaced frames instead (0: use intervalSeconds)	0
sceneThreshold	Scene mode: sensitivity, 1-99 (lower finds more cuts)	30
minSceneSeconds	Scene mode: ignore cuts closer than this to the previous one	1
maxFrames	Max frames per video (at most 1000)	50
frameFormat	jpg, png or webp	jpg
frameQuality	JPEG/WebP quality, 1-100	85
maxWidth	Scale wider frames down to this width (0: original size)	1280
contactSheet, contactSheetColumns	Also save a grid image of the frames, with this many columns	off, 4
gifClip, gifStartSeconds, gifDurationSeconds, gifWidth, gifFps	Also save an animated GIF clip (at most 15 s)	off, 0, 5, 480, 10
audioFormat	Also save the audio track: none, m4a, mp3 or opus	none
maxDurationMinutes	Longer videos are skipped with a free error (at most 180)	60
maxFileSizeMb	Larger downloads are skipped with a free error (at most 4000)	1000
timeoutSecs	Max download time per video	600

Processing your own files: upload them to a key-value store in your Apify account (Storage → Key-value stores → your store → upload a record) and list them in keyValueStoreRecords, e.g. my-user~videos/demo.mp4. Private record URLs (https://api.apify.com/v2/key-value-stores/.../records/...) also work in urls; the actor reads them with your run's token.

YouTube, TikTok, Instagram, Facebook, X, Vimeo, Twitch and other social or video-platform links are refused with a free error item, as are web pages and streaming playlists (.m3u8, .mpd). Use a direct link to a video file you have the rights to.

What data do you get?

One item per video (shortened):

{
  "url": "https://images-assets.nasa.gov/video/KSC_69-71212-sRGB/KSC_69-71212-sRGB~small.mp4",
  "sourceType": "url",
  "fileName": "KSC_69-71212-sRGB~small.mp4",
  "bytes": 15526635,
  "mode": "interval",
  "durationSeconds": 84.2,
  "width": 480,
  "height": 480,
  "fps": 30,
  "videoCodec": "h264",
  "audioCodec": null,
  "container": "mov,mp4,m4a,3gp,3g2,mj2",
  "bitRate": 1475214,
  "rotation": 0,
  "hasAudio": false,
  "frameFormat": "jpg",
  "frameCount": 12,
  "frames": [
    {
      "index": 1,
      "timeSeconds": 3.508,
      "timecode": "00:00:03.508",
      "sceneScore": null,
      "key": "video-0001-frame-0001.jpg",
      "url": "https://api.apify.com/v2/key-value-stores/.../records/video-0001-frame-0001.jpg",
      "width": 480,
      "height": 480,
      "bytes": 18446
    }
  ],
  "frameUrls": [
    "https://api.apify.com/v2/key-value-stores/.../records/video-0001-frame-0001.jpg"
  ],
  "contactSheetUrl": "https://api.apify.com/v2/key-value-stores/.../records/video-0001-contact-sheet.jpg",
  "gifUrl": "https://api.apify.com/v2/key-value-stores/.../records/video-0001-clip.gif",
  "audioUrl": null,
  "charges": { "video-processed": 1, "frame": 13, "audio-or-gif": 1 },
  "costUsd": 0.049,
  "warning": null,
  "error": null
}

• Files are in the run's key-value store as video-0001-frame-0001.jpg, video-0001-contact-sheet.jpg, video-0001-clip.gif and video-0001-audio.m4a (numbered by input order), linked from every item.
• In scene mode, sceneCount is the number of scenes found and each frame has its sceneScore (0-1; 1 for the first frame).
• Failed videos are returned with an error and no files, such as Download failed: HTTP 404., This link is a web page, not a video file., The video is 75.3 min long, over the 60 min limit (maxDurationMinutes). or The file has no video track (it is audio only).
• warning notes things like frames capped by maxFrames, a frame that could not be decoded or a video without audio.

How much does video frame extraction cost?

Pay per event, compute included, no start fee:

Event	Price	When
video-processed	$0.005	Each video delivered (metadata included). Failed videos are free.
frame	$0.003	Each frame image. A contact sheet counts as one frame.
scene-minute	$0.01	Scene mode only: each started minute of video analyzed.
audio-or-gif	$0.005	The audio track copied as is, or each GIF unit (see below).
audio-minute	$0.002	Each started minute of video when the audio track is re-encoded.

• Larger videos take longer to decode: above 1080p, scene minutes and GIF events count once per 1080p-sized area (1440p 2x, 4K 4x).
• GIF clip: one event per 5 s of a 480x480 px, 10 fps clip, so the default clip (5 s, 480 px wide, 10 fps) of a landscape video is 1 event; a 15 s, 1080 px, 25 fps clip is 22.
• Audio track: copied for a flat $0.005 when it is already in the chosen codec (AAC for M4A, MP3 for MP3, Opus for Opus); otherwise re-encoded for $0.002 per started minute.

Examples:

Run	Price
1 video, 12 thumbnails	$0.041
1 video, 12 thumbnails + contact sheet + GIF	$0.049
10 min video, scene mode, 25 keyframes	$0.18
Same 10 min video in 4K	$0.48
1 video, metadata + audio track copied (M4A)	$0.01
60 min video, audio re-encoded to MP3	$0.125

If you set a maximum cost per run, the actor never starts work it can't pay for: a video whose frames, scene minutes or extras don't fit the remaining budget is skipped with a free error item, and scene mode keeps only the strongest cuts the budget pays for.

Tips

• Speed: on the default 4 GB memory (1 CPU core), 24 frames from a 2-minute 720p video take about 12 s; scene detection takes about 5 s per minute of 720p video and 9 s per minute of 1080p, plus the download. The audio track is copied in seconds when it is already in the chosen codec (e.g. AAC in an MP4, for M4A); re-encoding takes about 1 s per minute.
• Choose a mode: interval mode is fastest and cheapest for thumbnails. Scene mode suits edited videos (ads, talks, trailers); for one-shot footage (a webcam, a launch) it finds few scenes.
• Sensitivity: if scene mode finds too many frames (flicker, fast motion), raise sceneThreshold or minSceneSeconds; if it misses soft cuts or fades, lower the threshold to 15-20.
• File size: JPEG at quality 85 is a good default; WebP is about 30% smaller; PNG is lossless and several times larger.
• Long videos: the default limit is 60 minutes (180 at most). Interval mode spreads maxFrames over the whole video when more frames would be needed.

FAQ

Can I extract frames from YouTube or TikTok videos? No. Social and video-platform links are refused: downloading from them breaks their terms of service. Use direct links to video files you own or have the rights to, or upload them to a key-value store.

Which codecs are supported? Everything FFmpeg 7.1 decodes natively: H.264, HEVC/H.265, VP8, VP9, AV1, MPEG-2, MPEG-4, ProRes, DNxHD, MJPEG and many more, in MP4, MOV, MKV, WEBM, AVI, TS and other containers.

Are my videos sent anywhere else? No. Everything runs inside the actor's container on Apify.

Can AI agents use it? Yes, through the Apify API, the Apify MCP server or any Apify integration (Make, Zapier, n8n, LangChain).

Disclaimer: You are responsible for having the right to process the videos you submit.

Open-source software

This actor runs FFmpeg 7.1 in an LGPL build compiled from the official source, with dav1d (BSD), libwebp (BSD), libopus (BSD) and LAME (LGPL). See THIRD_PARTY_NOTICES.md in the source for all notices. It is not affiliated with or endorsed by the FFmpeg project.

Related actors

• Whisper Audio & Video Transcriber: Audio, video and podcast transcription with Whisper, with SRT and VTT subtitles.
• AI Image Upscaler: Upscale and enhance images 2x, 3x or 4x with Real-ESRGAN.
• YouTube Transcript Scraper: YouTube transcripts as text, segments, SRT or VTT, for AI and RAG.