Video 2 Social Post

Video to Social Post is a video to social media post ai workflow that downloads one accessible video, transcribes speech locally, and creates reviewable drafts for selected platform, tone, language, and model combinations. It returns one structured Dataset row per completed generation task and can attach shared source screenshots or styled image variants.

• Six explicit inputs. Choose one video, image mode, language, one or more platform profiles, tones, and model routes.
• Nine-field task rows. Every saved draft includes content, targeting labels, images, provider-reported tokens, and provenance stamps.
• Bounded media retrieval. The Actor tries direct extraction first, then no more than three public-proxy fallbacks with an eight-second socket timeout.
• Self-contained transcription. FFmpeg and the CPU-int8 Whisper small model are bundled in the 2 GiB runtime image.

A smallest paid-plan request contains one platform, one tone, and one model, so it requests one post call. If that call reports 128 text tokens, the FREE event schedule adds $0.004992 for text plus $0.02 for Actor Start at 2 GiB, before Actor Usage or optional image events.

Why Choose Video to Social Post

The Actor joins media retrieval, local speech recognition, prompt routing, and structured delivery in one run. That reduces the integration work required to turn an authorized source video into drafts that editors can compare, review, and revise.

The media stage uses yt-dlp and accepts an absolute HTTP(S) URL rather than a narrow platform allowlist. Compatibility therefore follows the installed extractors and the current source site. Direct retrieval is always attempted first. Only after it fails can the Actor try up to three shuffled public proxies; it does not spend paid residential-proxy traffic.

FFmpeg extracts 16 kHz mono audio when possible. The bundled faster-whisper small model transcribes locally on CPU with int8 compute, so speech does not need to be sent to the post-generation model as audio. The resulting timestamped segments become prompt context; the transcript itself is not part of the nine-field public Dataset.

Paid-plan task count is the product of selected platforms, tones, and models. A request with two platforms, three tones, and two models asks for 12 post calls. Duplicate selections intentionally create duplicate tasks. FREE-plan runs are narrower: DeepSeek, the first platform, the first tone, and no images.

The Actor creates drafts only. A platform label is an internal writing profile, not a publishing connection, current policy guarantee, or promise of engagement.

Quick Start Guide

Step 1: Configure

Open the Actor input page. Begin with the prefilled TikTok video, None image style, and one value in each array.

{
  "video_url": "https://www.tiktok.com/@openai/video/7521583654028512525",
  "image_style": "None",
  "language": "English",
  "platforms": ["X-Free"],
  "tones": ["Questioning"],
  "models": ["DeepSeek"]
}

Step 2: Run

Click Start. The Actor creates a unique workspace, retrieves the video, extracts audio, transcribes speech, and runs the selected generation tasks. Image selection follows successful post generation when a paid-plan image mode is active.

Step 3: Review

Open Dataset and inspect content, labels, tokens, and run status together. For image modes, also inspect the default Key-Value Store. Treat every draft as review material: compare names, numbers, quotations, language, tone, claims, and platform suitability with the source before publishing elsewhere.

Input Parameters

The six required inputs match runtime order and validation exactly. Console prefills do not make a property optional.

Parameter	Type	Choices	Description
video_url	string	absolute HTTP(S) URL	One source handled by the current extractor; playlists are disabled.
image_style	string	None, Original, or 8 styles	None skips images; Original selects screenshots; other values request styled variants.
language	string	12 languages	Requested output language; correctness still requires review.
platforms	array	12 profiles	X-Free, X-Premium, Instagram, LinkedIn, Facebook, Pinterest, Reddit, TikTok, YouTube, Rednote, Threads, or Mastodon.
tones	array	12 tones	Questioning, Emotional, Contrasting, Controversial, Storytelling, Urgent, Humorous, Motivational, Educational, Personal, Professional, or Casual.
models	array	6 routes	DeepSeek, Claude, Gemini, Grok, Qwen, or GPT.

Unknown fields, empty values, non-string array members, invalid enums, and non-HTTP(S) video URLs are rejected before retrieval. Arrays have a minimum of one item and no schema maximum. Deduplicate them in the caller unless repeated generations are deliberate, and use a run charge cap sized for the requested product.

Model labels map to current OpenRouter routes at build time. The Dataset returns the stable uppercase label, such as DEEPSEEK, rather than the provider route ID. Provider availability, context limits, token accounting, and model output can change independently.

Output Data Schema

Each Dataset item is one completed social-post task with nine ordered fields. Empty provider content is not written as a row.

Field	Type	Meaning
content	string	Reviewable generated draft for the task.
platform	string	Requested internal platform profile.
model	string	Uppercase selected model label.
tone	string	Requested tone label.
language	string	Requested output language label.
images	array	Shared KVS URLs; empty when images are off or unavailable.
tokens	integer	Provider-reported total tokens for this post call only.
processor	URL	Actor Store URL that created the row.
processed_at	datetime	ISO 8601 UTC row timestamp.

{
  "content": "What skill matters when tools change faster than job titles? Adaptability turns every new tool into leverage.",
  "platform": "X-Free",
  "model": "DEEPSEEK",
  "tone": "Questioning",
  "language": "English",
  "images": [],
  "tokens": 128,
  "processor": "https://apify.com/agentx/video-to-social-post",
  "processed_at": "2026-08-13T17:13:13+00:00"
}

The input URL and local transcript are intentionally absent from task rows. Preserve the run ID and INPUT record when downstream lineage needs them. tokens excludes local transcription and separate image-discovery or selection calls.

Integration Examples

Use the REST API, Apify clients, Schedules, webhooks, or MCP to place the Actor inside a review workflow. Keep the Apify token outside code and set a deliberate charge cap.

{
  "method": "POST",
  "url": "https://api.apify.com/v2/acts/agentx~video-to-social-post/runs?waitForFinish=300&maxTotalChargeUsd=1",
  "headers": {
    "Authorization": "Bearer YOUR_APIFY_TOKEN",
    "Content-Type": "application/json"
  },
  "body": {
    "video_url": "https://www.tiktok.com/@openai/video/7521583654028512525",
    "image_style": "None",
    "language": "English",
    "platforms": ["LinkedIn"],
    "tones": ["Professional"],
    "models": ["DeepSeek"]
  }
}

After completion, retrieve defaultDatasetId, compare item count with the requested task product, and route drafts to human approval. A charge-limit stop can end the product early. Webhooks should branch on terminal run status as well as Dataset count.

Pricing

Charges follow Actor memory and provider-reported token events, not the number of saved rows alone. At 2 GiB, Actor Start emits two $0.01 events. Actor Usage is $0.0001 per event.

Text Token pricing per reported token is FREE $0.000039, BRONZE $0.000036, SILVER $0.000033, and GOLD/PLATINUM/DIAMOND $0.000030. Image Token pricing is FREE $0.000074, BRONZE $0.000072, SILVER $0.000070, and GOLD/PLATINUM/DIAMOND $0.000069.

Text events cover post generation and the text calls used by paid image discovery and selection. Image events apply to styled-image model usage; Original screenshots do not use an image-token event. Events are charged when provider usage is reported, before the final Dataset batch, so charged tokens and saved-row count can differ after a later interruption.

Use Cases

Editorial teams can repurpose authorized interviews, product explainers, webinars, and short videos into a controlled draft queue. A reviewer can compare model or tone variants while retaining a single structured contract for downstream tools.

Localization teams can request one of 12 languages as a starting point, then verify meaning, idioms, names, numbers, cultural fit, and required disclosures. Research or support teams can extract candidate takeaways from a spoken update without treating generated text as a transcript or source of truth.

Original-image mode can attach selected source screenshots when reuse rights exist. Styled modes can create visual variants from selected frames, but neither mode guarantees six images or a particular aesthetic result.

Alternatives

Choose a transcript Actor when you need source text, a downloader when you need media, or this Actor when the deliverable is platform-oriented drafts. Video Transcript exposes transcript-oriented output. Video Captions Downloader retrieves source captions when available. All Video Scraper targets broader video observations.

A custom pipeline offers complete control over extraction, ASR, prompts, models, storage, and moderation, but you must operate every component and billing boundary. Manual editing gives the strongest contextual judgment for a few videos. This Actor is useful when repeated, structured draft generation outweighs that integration work.

Limits and Troubleshooting

Most failures belong to one of four stages: retrieval, transcription, generation, or image storage. Start with the terminal status and the last completed stage.

• Retrieval failures can reflect removed media, authentication, geo restrictions, extractor changes, or lack of a usable format. Public proxy fallback does not add source cookies.
• Transcription failures can reflect silent media, unsupported audio, corrupt downloads, or speech the local model cannot recognize reliably.
• Generation failures can reflect missing provider availability, route changes, safety controls, timeouts, or an empty provider response. Zero successful posts ends the run as failed.
• Image failures can reflect unusable timestamps, KVS storage, provider output, or charge limits. Post rows can still succeed with an empty images array.

Long media increases download, FFmpeg, and CPU transcription time before the first post appears. Large selection products increase provider calls sequentially. Use the smallest representative request first, then expand while observing run duration, events, and output quality.

Trust and Reliability

The public contract is intentionally explicit about what is local, what is external, and what is best-effort. Media files are isolated by a unique run workspace. Candidate MP4 files must pass ffprobe. Audio extraction checks FFmpeg completion. The bundled model avoids a cold transcription-model download.

Empty provider content is excluded, and a zero-row generation outcome is a failed run rather than a silent success. Token charges skip zero usage. Dataset timestamps and processor URLs make rows attributable to a run surface, but they do not certify factual accuracy.

No video site, social network, OpenRouter provider, transcription project, or regulator endorses the Actor. Review the source and destination rules independently.

Legal and Compliance

Use only media you are authorized to download, transcribe, transform, store, and reuse. A public URL does not automatically grant copyright, privacy, publicity, model-processing, or commercial rights.

Review generated claims, quotations, endorsements, trademarks, likenesses, safety, accessibility, advertising disclosures, and synthetic-media labels before publication. Define retention for INPUT, Dataset, and KVS records. Do not use drafts to impersonate speakers, fabricate endorsements, harass people, or evade destination-platform rules. This is operational guidance, not legal advice.

Frequently Asked Questions

How does video to social media post ai generation work?

The Actor downloads the video, transcribes speech locally, creates tasks from the selected arrays, sends text prompts through configured model routes, and saves non-empty responses as nine-field rows.

Which video sites are supported?

Compatibility follows the installed yt-dlp extractors and current site behavior. The supported-sites list is useful guidance, not a guarantee that every listed URL works today.

How many post rows will a run create?

Paid intent is platform count × tone count × model count, including duplicates, until a provider error or charge limit interrupts. FREE-plan runs use only the first platform, first tone, and DeepSeek.

Can the Actor publish directly to social networks?

No. Platform names are writing profiles. Review and publish approved drafts through your own authorized workflow.

Why can token charges and Dataset count differ?

Provider tokens are charged as calls complete, while rows are pushed after generation and optional image work. A later failure or charge-limit stop can leave fewer rows than charged calls.

Are images or styled variants guaranteed?

No. None returns no images, Original selects up to six source screenshots, and styled modes attempt variants from selected frames. Any image stage can return fewer or none.

Is the local transcript included in each row?

No. It is internal prompt context. Use a transcript-specific Actor when transcript delivery is the product requirement.

How should generated drafts be reviewed?

Compare them with the source for facts and context, verify the requested language and tone, check current destination limits, and require human approval before publication.

Related AgentX Actors

AgentX publishes a broad catalog of Actors organized by product category. Closest to this Actor:

• Video Transcript — return transcript-oriented output from accessible video.
• Video Captions Downloader — retrieve source-provided subtitle files when available.
• All Video Scraper — collect broader video metadata and media-oriented output.

Business and Market Intelligence

• All Web Search Scraper — multi-engine web search across ten independent indexes
• Google Keyword Trends — Google Trends keyword interest, compared five at a time
• Google Trends Scraper — Google Trends trending searches
• Google Maps Store Scraper — Google Maps business listings
• LinkedIn Company Lookup — LinkedIn company records
• LinkedIn Profile Lookup — LinkedIn profile records
• Weather Forecast API — forecast and historical weather data

Jobs and Hiring

• All Jobs Scraper — multi-platform job listings
• Bayt Jobs Scraper — Bayt job listings
• Glassdoor Jobs Scraper — Glassdoor job listings
• Indeed Jobs Scraper — Indeed job listings
• JobStreet Company Profile — JobStreet employer records
• Jooble Jobs Scraper — Jooble job listings
• LinkedIn Jobs Scraper — LinkedIn job listings
• Naukri Jobs Scraper — Naukri job listings
• Talent Jobs Scraper — Talent.com job listings
• ZipRecruiter Jobs Scraper — ZipRecruiter job listings

Social Media

• All Forums Thread Scraper — multi-platform forum and Q&A threads
• Instagram Creator Scraper — Instagram creator profiles
• Instagram Reels Scraper — Instagram Reels metadata
• Instagram Trending Scraper — Instagram trending content
• Medium Scraper — Medium articles
• Medium Profile Scraper — Medium author profiles
• Reddit Profile Scraper — Reddit user profiles
• Reddit Viral Scraper — Reddit viral posts
• SubReddit Info Scraper — subreddit metadata
• Subreddit Members Scraper — subreddit member lists
• Subreddit Posts Scraper — subreddit post feeds
• Telegram Chat Scraper — Telegram chat messages
• Telegram Info Scraper — Telegram channel metadata
• Telegram Member Adder — Telegram group membership management
• Telegram Member Scraper — Telegram group members
• Telegram Private Group Scraper — private Telegram group data
• TikTok Creator API — TikTok creator profiles
• TikTok User Lookup — TikTok account records
• X Twitter Community API — X community data
• X Twitter Profile Lookup — X profile records
• YouTube Creator Email Scraper — YouTube creator contact data

Video, Transcripts and Downloads

• All Video Scraper — multi-platform video metadata
• All Short Videos Scraper — short-video keyword search across two independent indexes
• All Image Scraper — image search and download across five engines
• All Short Drama Downloader — short-drama episodes from fifteen official apps
• Video Transcript — multi-platform transcripts
• Video Captions Downloader — caption files
• YouTube Transcript — YouTube transcripts
• TikTok Transcript — TikTok transcripts
• X Twitter Transcript — X video transcripts
• Facebook Transcript — Facebook video transcripts
• Bilibili Transcript — Bilibili transcripts
• Dailymotion Transcript — Dailymotion transcripts
• Rutube Transcript — Rutube transcripts
• Loom Transcript — Loom transcripts
• Wistia Transcript — Wistia transcripts
• Instagram Reels Downloader — Reels downloads
• Kick Clip Downloader — Kick clip downloads
• Linkedin Video Downloader — LinkedIn video downloads
• Pinterest Video Downloader — Pinterest video downloads
• Reddit Video Downloader — Reddit video downloads
• Snapchat Video Downloader — Snapchat video downloads
• TED Talk Downloader — TED talk downloads
• TikTok Live Downloader — TikTok live downloads
• Twitch VOD Downloader — Twitch VOD downloads
• Zoom Recording Downloader — Zoom recording downloads

E-Commerce and Retail

• All Shopping Scraper — multi-platform product data
• AliExpress Product Scraper — AliExpress products
• Amazon Storefront Scraper — Amazon Brand Store pages
• Bol Product Scraper — Bol.com products
• eBay Seller Leads API — eBay seller records
• Hepsiburada Product Scraper — Hepsiburada products
• Kakaku Product Scraper — Kakaku.com products
• Rakuten Product Scraper — Rakuten products

Classifieds and Automotive

• All Vehicle Scraper — multi-platform vehicle listings
• AutoTrader Vehicle Search — AutoTrader US listings
• AutoTrader UK Vehicle Search — AutoTrader UK listings

Real Estate

• All Property Scraper — multi-platform property listings
• Homes Property Scraper — Homes.com listings
• MagicBricks Property Scraper — MagicBricks listings
• NoBroker Property Scraper — NoBroker listings
• Property24 Property Scraper — Property24 listings
• Realtor Property Scraper — Realtor.com listings
• Zillow Property Scraper — Zillow listings

Support and Community

Use the Store page to run the Actor, the API page for integration details, and the Issues page for reproducible failures. Include the run ID, UTC time, build, redacted source URL, input selections, expected task count, terminal status, event counts, Dataset count, and failure stage. Never send API keys or private source credentials. Product questions belong in the AgentX community.

AgentX is an Arcyton brand — arcyton.com.

Last Updated: August 13, 2026