Cinematic Shot Breakdown: Cuts, Camera & AI Video Prompts
Pricing
$10.00 / 1,000 10 seconds of video analyseds
Cinematic Shot Breakdown: Cuts, Camera & AI Video Prompts
Cinematic shot breakdown for a clip up to 120 s: every cut, shot timing, estimated camera movement, lighting, palette and rhythm, plus Sora, Veo, Kling and Runway prompts per shot. Optional BYOK vision. $0.01 per 10 seconds of video.
Pricing
$10.00 / 1,000 10 seconds of video analyseds
Rating
0.0
(0)
Developer
Ferrolark
Maintained by CommunityActor stats
0
Bookmarked
2
Total users
1
Monthly active users
4 days ago
Last modified
Categories
Features
Share
Cinematic Shot Breakdown: Cut Detection, Camera Moves & AI Video Prompts
Upload a clip of up to 2 minutes and get a shot-by-shot breakdown: every cut with timecodes, shot length and rhythm, estimated camera movement, lighting, colour palette and shot size, plus ready-to-edit prompts for Sora, Veo, Kling and Runway for each shot.
- Cuts: PySceneDetect (adaptive + content detectors combined) on a letterbox-cropped proxy, with flash-frame merging.
- Edit rhythm: average and median shot length (ASL), cuts per minute, pace label, and "held shot / flash cut" per shot.
- Camera movement: optical flow estimates pan, tilt, push in / pull out, handheld shake and speed, with a confidence value.
- Look: palette with colour names and shares, key (high/low/mid), contrast, colour temperature, saturation, vignetting.
- Shot size: close-up / medium / wide from detected faces (heuristic).
- Prompts: four model-specific prompts per shot. Without a vision model the subject and setting stay as
[subject]/[setting]placeholders for you to fill. - Optional BYOK vision: add your own OpenAI, Gemini, OpenRouter or Anthropic key and a vision model describes subject, setting, angle, lens and mood, which fill the prompts.
Typical uses: studying the edit of an ad or trailer you are allowed to analyse, briefing a re-shoot, building a shot list for AI video generation, and comparing pacing between your own cuts.
Real example
The Sintel trailer (Blender Foundation, CC-BY 3.0), 52 s at 854×480:
{"videoUrl": "https://example.com/your-own-clip.mp4","rightsConfirmed": true}
- Run time: 14 s on Apify at 2,048 MB. Platform usage: $0.0022.
- Your cost: 6 started 10-second blocks × $0.01 = $0.06.
- Result: 18 shots, ASL 2.9 s, median 1.27 s, 19.5 cuts per minute ("brisk"), palette dominated by black, dark brown and tan.
| Shot | Start | Seconds | Camera | Lighting |
|---|---|---|---|---|
| 1 | 00:00.000 | 13.5 | slow push in / zoom in | mid-key, medium contrast, desaturated, vignetted |
| 2 | 00:13.500 | 0.4 | static / locked off | flat, low contrast, deep shadows (flash cut) |
Shot 1's Veo prompt: "A cinematic medium close-up of [subject] [action] in [setting]. The camera uses a slow push in / zoom in. Balanced / mid-key, medium contrast, neutral colour temperature, desaturated, vignetted, light pooled on the centre. Colour grade built on black, charcoal, dark brown, grey. Duration about 14 seconds. No on-screen text."
Price and limits
- $0.01 per started 10 seconds of analysed video ($0.06 per minute). A 30 s ad costs $0.03, a 2-minute trailer $0.12.
- Charged only after a successful analysis. Clips that cannot be downloaded or decoded, are too long, or come from a refused source are not charged.
- Limits: 120 s per clip (or turn on
trimToLimitto analyse the first 120 s), 250 MB per file. - Vision-model costs (BYOK) go to your own provider account. With small models (for example gpt-5-nano or gemini-2.5-flash-lite) and 2 frames per shot this is typically well under a cent per clip.
Your rights and what we refuse
You must own the clip or be allowed to analyse it (rightsConfirmed: true). The Actor does not download from platforms whose terms forbid it: links to YouTube, TikTok, Instagram, Facebook, X, Vimeo, Twitch, Netflix, Prime Video and similar sites are refused. Use a direct file link (your storage, S3, Drive direct link) or base64.
Nothing from the clip is redistributed: keyframes are analysed in memory; saveKeyframes keeps them only in your private run storage.
Input
| Field | Default | What it does |
|---|---|---|
videoUrl / videoBase64 | none | A direct link to an MP4/MOV/WebM/MKV file, or base64 for n8n and Make binaries. |
rightsConfirmed | false | Required for your own clips (not for the public Sintel sample). |
maxSeconds / trimToLimit | 120 / false | Longer clips fail (not charged) unless trimming is on. |
cutSensitivity | 3.0 | Lower finds more cuts, higher fewer. |
minShotSeconds | 0.4 | Merges cuts closer than this. |
visionProvider | none | openai, gemini, openrouter or anthropic (BYOK). |
visionApiKey / visionModel | none | Your key (secret, never logged) and an optional model id. |
maxShotsForVision | 60 | Caps your provider cost on very fast edits. |
saveKeyframes | false | Keep analysed frames in your private run storage. |
Call it from code or an AI agent
curl -X POST "https://api.apify.com/v2/acts/ferrolark~cinematic-shot-breakdown/run-sync-get-dataset-items?maxTotalChargeUsd=0.2" \-H "Authorization: Bearer $APIFY_TOKEN" -H "Content-Type: application/json" \-d '{"videoUrl": "https://example.com/your-own-clip.mp4", "rightsConfirmed": true}'
The run summary (ASL, cuts per minute, palette, shot types) is in the OUTPUT record of the run's key-value store.
AI agents (MCP): add the Apify MCP server (https://mcp.apify.com) and allow ferrolark/cinematic-shot-breakdown.
FAQ
How accurate is the camera movement?
It is an optical-flow estimate on a small proxy. Pans, tilts and push-ins are usually right; complex moves (orbit, crane plus pan) come out as the dominant component. Each shot has a confidence value.
Why are subject and setting placeholders?
Without a vision model the Actor only measures what it can compute. Add your own vision key to fill them in, or edit the prompts yourself.
Can I analyse a YouTube video?
No. Download platforms' terms forbid it, so those links are refused. Use a file you own.
More Ferrolark tools
- Brand Name Availability Checker: domains, GitHub, npm, PyPI and more in one report.
- Domain Intelligence Report: WHOIS, DNS, email security, SSL and hosting for any domain.
- Launch Asset Pack: favicons, PWA icons, OG image and head tags from one logo.
- Live Deploy QA: meta tags, canonical, robots, OG image and favicon of your deployed routes, with diffs between deploys.
- Sitemap Delta: new, removed and broken URLs in any sitemap since the last run.
- RSS Delta: only new RSS, Atom and JSON Feed items, de-duplicated.
- PDF Merge API: merge PDFs from URLs or base64 with page ranges and a SHA-256 manifest.
- PDF Split API: split a PDF by page ranges or every N pages, or extract pages, with a ZIP and manifest.
- Image Convert: PNG, JPG, WebP, AVIF and HEIC conversion (SVG input too) from URLs or base64.
- Image Compress & Resize: resize images and compress them to a target size in KB, WebP or AVIF.
- PDF to Image: PDF pages to PNG, JPG or WebP at the DPI you choose.
- Image to PDF: JPG, PNG, HEIC and WebP images to one PDF, JPEGs embedded without recompression.
- PDF Compress: smaller PDFs, lossless or with image downsampling; charged only when smaller.
- HTML to PDF & URL to PDF: web pages or raw HTML to PDF with headless Chrome.
- Markdown to PDF: Markdown with tables, code and footnotes to a clean PDF.
- DOCX to PDF: Word, Excel, PowerPoint and ODT to PDF with LibreOffice.
- Data Convert: CSV, JSON, NDJSON and Excel XLSX in any direction.
- Criteria Checker: pass/review/fail QA gate for AI output, emails and posts (BYOK TypeSafe).
- Support Ticket Triage: route tickets to teams with P1-P4 priority and escalation flags (BYOK TypeSafe).
- Amazon Listing Checker: title rules, restricted claims and search-term audit before upload.
Built by Ferrolark.