TrendLoom — Source Research, Content Ideas & Script Packages

Turn a research question into primary-paper references, three original content ideas, or a six-scene Short script draft. TrendLoom searches real primary-paper abstracts through the official Europe PMC API, accepts only explicit commercially reusable CC BY or CC0 license metadata, and preserves their PubMed citation links and attribution. API access and model processing are included: you do not need a YouTube or OpenRouter key.

This release does not use the YouTube API, fetch channel pages, collect or analyze YouTube metrics, or access private channel analytics. For channel-aware ideas, you may paste a short manual summary and a few recent titles you own or are authorized to share in styleNotes. TrendLoom will use them only as context for distinct ideas: no automatic channel lookup, performance estimates, title copying or metric output. Do not paste full transcripts or material you lack permission to use.

Choose one result

• research — $0.02: up to three relevant, explicitly licensed primary-paper abstracts, titles, attribution and citation links. No AI generation.
• ideas — $0.05: source research plus three original topics, titles, hooks, angles, source references and supporting quotations.
• package — $0.10: source research, three ideas and one six-scene production draft: narration, three title variants, description, tags, thumbnail brief, image/camera prompts and source references. ideaIndex chooses one of the ideas generated in this run.

One selected custom result event is charged per successfully delivered result, not per source or per scene. A package costs $0.10 in TrendLoom result charges, not $0.02 + $0.05 + $0.10. The Actor refuses a non-zero apify-default-dataset-item event because it would add a second per-item charge. Apify may separately charge its platform-managed apify-actor-start event (default $0.00005/run when enabled) and compute usage; those platform fees are not part of the TrendLoom result price. Check the active Apify pricing details and your run spending limit before starting. Upstream model processing is included: no OpenRouter key is required.

Quick start

{
  "action": "research",
  "query": "sleep memory consolidation",
  "language": "en"
}

Use formal research keywords rather than a headline. The default query performs a real source lookup, not a generated demo.

Ideas tailored to your channel — manual input

Choose Three content ideas and fill in:

• query: the scientific topic to research (still required).
• channelDescription: your own short description of the channel topic, audience and editorial style (optional, up to 1000 characters).
• exampleTitles: up to five previous titles you own or are authorized to share (optional, up to 180 characters each).
• styleNotes: additional creative/visual preferences (optional).

The service uses this context to suggest distinct new angles, not to analyze channel performance. No channel/video link, ID or handle is fetched; no views, subscribers, scores or growth predictions are produced. The context is not factual evidence: claims still need the licensed primary abstracts. Exact and conservatively detected near-copy titles trigger the existing bounded repair or a clear failure, not silent replacement. This heuristic is not a semantic originality or rights guarantee.

Use your own description or authorized brief notes; do not paste full transcripts, statistics, private data or material you lack permission to process. Existing inputs using only styleNotes remain supported. The same optional channel fields can guide the ideas stage of a script package.

Example:

{
  "action": "ideas",
  "query": "sleep memory consolidation",
  "language": "de",
  "channelDescription": "Wissenschaft für neugierige Einsteiger: alltagsnahe Erklärungen, verständlich und ohne Fachjargon.",
  "exampleTitles": ["Warum uns vertraute Gerüche nicht mehr auffallen", "Wie Erinnerungen im Schlaf verändert werden"],
  "styleNotes": "Einfache visuelle Metaphern; ruhige Papieranimation."
}

Supported video languages: English, German, Turkish. Source abstracts retain their original language. Visual prompts may be English. Supported target durations: 45, 60 or 90 seconds. Durations are planning estimates, including four seconds reserved for an outro—not measured voiceover.

Output and exports

Each successful run writes one complete result to its default dataset and the OUTPUT key-value record. Dataset overview columns include action, query, title, source count and idea count. Nested fields retain the sources, ideas, script and warnings. Use JSON export to preserve nested structures.

For script packages, additional files are available in the run's key-value store:

• PACKAGE.md: metadata and production package.
• SCENES.json: the six-scene draft.
• PRODUCTION-PROMPT.txt: a standalone production brief.

Actual citations come from source lookups; they are not invented by the model. Literal-quotation and identifier checks help catch malformed evidence. These checks do not establish that every generated claim is factually supported. Read the actual paper, preserve population/method/limitations, and review all narration and visuals before use. Every package is an unapproved draft, even if an automated review passes.

What is not included

• No rendered video, image generation, audio/TTS, animation or publication.
• No guaranteed factual accuracy, virality, reach, sales or originality clearance.
• No complete systematic literature review or full-paper access. An abstract may omit important qualifications; clinical/medical content is not medical advice.
• No YouTube views, rankings, channel scraping, private analytics or video downloads.
• Not all news, sport or general-interest questions have suitable primary research. If no relevant abstract with an accepted reuse license is found, the run fails clearly rather than inventing sources.

Limits and reliability

Inputs are bounded: a 3–180-character query, at most 1000 characters each of style notes and channel description, up to five example titles of at most 180 characters each, exactly three generated ideas and exactly six script scenes. Work runs through a budget-guarded TrendLoom service with two concurrent operations and an initial fair-use cap of ten operations per Apify customer per UTC day. Shared upstream/cost limits can temporarily reject new work; such a rejection is not a successful result.

No automatic replay is made after an uncertain service timeout. The service binds each request to its Apify run identity and retains idempotency state to prevent repeated paid work for that run. Start a new run only after inspecting the previous run's status. Upstream source and model availability varies.

API and integrations

Run this Actor from Apify Console, schedules, Make, n8n, Zapier, the Apify API or an AI agent. Authenticate to Apify with your Apify API token, never with a TrendLoom service secret. After publication, use the API tab for exact code examples and URLs.

curl --request POST \
  --header "Authorization: Bearer $APIFY_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{"action":"research","query":"sleep memory consolidation","language":"en"}' \
  'https://api.apify.com/v2/acts/genkenobi~trendloom/run-sync-get-dataset-items?timeout=420&memory=256&maxTotalChargeUsd=0.11'

The result dataset belongs to your Apify run. Keep your token private. Longer scripts can exceed a synchronous client's deadline; use the asynchronous run endpoint and retrieve the dataset after completion if needed.

Privacy and use terms

By running the Actor, you request processing of the supplied query, manual channel description, title examples, creative notes and selected settings. Do not include secrets, sensitive personal information, confidential manuscripts or material you do not have permission to process.

The Actor uses Apify infrastructure and a developer-operated TrendLoom service. For run authentication, it forwards only the current run's ephemeral limited-permissions Apify token to that fixed HTTPS service; the service uses it to verify this run's Actor ID and user identity, and does not store or log the token. It does not receive your full-account API token. Source keywords are sent to the relevant public literature APIs. For ideas and scripts, your query, channel description, example titles, creative notes, selected abstracts and draft content are sent to OpenRouter and its selected model provider. These third parties have their own privacy policies: Apify, OpenRouter, Europe PMC.

TrendLoom's separate service keeps completed result payloads for up to seven days for idempotency/recovery, then deletes those payloads. Minimal request identity, authorization and cost-accounting records remain for operation and abuse prevention. It does not read the owner's private Studio storage. Your Apify inputs, logs, datasets and exports are retained according to your Apify account's retention settings, independently of the TrendLoom service. You can delete run storage in Apify; use this Actor's Issues tab for support or a service-side deletion request and include only the run ID, not API tokens.

Source results include the index-reported CC BY or CC0 license and available author attribution. Preserve that attribution and the applicable license; check the original paper for the authoritative terms and any third-party exclusions. Missing, non-commercial, no-derivatives or unknown licenses are not accepted. A citation or abstract license is not permission to reproduce unrelated full-paper images or datasets. You are responsible for editorial review, permissions, final production and any publication. TrendLoom is not affiliated with PubMed, Europe PMC, Crossref, YouTube or the cited authors.