Replicate AI Model Explore Scraper

Scrape Replicate's public model catalog (replicate.com/explore) — the directory of AI models runnable via API for image generation, LLMs, video, speech, and more. Browse trending & featured models, filter by one of 39 task collections (text-to-image, LLMs, video, audio, and more), browse a creator's full model list, search by keyword, or look up specific models by owner/model-name. Extracts exact run counts, descriptions, cover images, published per-unit pricing, and latest version IDs straight from each model's own page. HTTP-only, no login, no API key, no proxy required.

What this actor does

• Five modes: trending, byCollection, byOwner, byModelIds, search
• Live per-model detail lookup: every record is built from that model's own Replicate page — exact run count, official-model flag, hardware, pricing, and latest version are always current
• Finite collection enum: 39 official Replicate task collections (text-to-image, text-to-video, LLMs, speech-to-text, upscaling, background removal, and more)
• Filters: minimum run count, official-models-only
• Published pricing: per-unit price (per image, per second of video, per input/output token, etc.), including resolution/mode-gated tiers when a model has more than one
• Empty fields are omitted from every record

Output per model

• modelId — owner/model-name (e.g. black-forest-labs/flux-dev)
• owner, name — split components of modelId
• displayName — model's human-readable title (e.g. FLUX.1 [dev])
• description — model tagline/summary
• runCount — exact total API run count
• isOfficial — true if Replicate marks this as an official, publisher-maintained model
• userType — organization or user
• hardware — GPU/CPU class the model runs on (e.g. H100, CPU)
• visibility — public or private
• latestVersion — latest model version ID (hash)
• latestVersionCreatedAt — ISO timestamp of the latest version
• pricing[] — Replicate's published per-unit pricing ({title, price, metric, note, variant}); one entry per billing tier x metric, e.g. per output image, per second of video, per input/output token. variant distinguishes resolution/mode-gated tiers when a model has more than one (e.g. 480p vs 1080p video)
• avatarUrl — model owner's avatar image
• coverImageUrl — model's cover/example image
• modelUrl — the Replicate page for this model
• sourceUrl — same as modelUrl (canonical page scraped)
• recordType: "model", scrapedAt

Input

Field	Type	Default	Description
mode	string	trending	trending / byCollection / byOwner / byModelIds / search
collection	string	text-to-image	(mode=byCollection) One of 39 task collections
ownerName	string	–	(mode=byOwner) Replicate username/org, e.g. black-forest-labs
modelIds	array	–	(mode=byModelIds) owner/model-name strings to look up directly
searchQuery	string	–	(mode=search) Keyword matched against owner, name, and blurb in the trending catalog
minRuns	int	0	Only include models with at least this many total runs
officialOnly	boolean	false	Only include Replicate-official models
maxItems	int	10	Maximum models to return (1–200)

trending / byCollection / byOwner / search each list from a single Replicate listing page, so the number of models actually available is capped by how many that page shows (typically 80–140 depending on the page) — maxItems above that ceiling returns everything the page has rather than an error. Use byModelIds to fetch any additional specific models by name.

Examples

Example: trending models

{
  "mode": "trending",
  "maxItems": 20
}

Example: browse the text-to-video collection

{
  "mode": "byCollection",
  "collection": "text-to-video",
  "maxItems": 10
}

Example: all models by a creator

{
  "mode": "byOwner",
  "ownerName": "black-forest-labs",
  "maxItems": 30
}

Example: look up specific models

{
  "mode": "byModelIds",
  "modelIds": ["black-forest-labs/flux-dev", "anthropic/claude-4.5-sonnet"]
}

Example: search with a popularity floor

{
  "mode": "search",
  "searchQuery": "upscale",
  "minRuns": 100000,
  "maxItems": 10
}

Use cases

• AI tooling research — track which image/video/LLM models are trending on Replicate
• Vendor comparison — compare run counts and hardware requirements across models in a task category
• Model discovery — find every model published by a specific lab (e.g. black-forest-labs, bytedance)
• Catalog mirroring — sync a curated model directory for an internal AI tools portal
• Market signal tracking — monitor a collection (e.g. language-models, text-to-video) for new releases and their adoption via runCount

FAQ

Is this affiliated with Replicate? No — this is an independent, third-party actor that reads Replicate's public model pages. It is not affiliated with or endorsed by Replicate.

Does this actor run models or call the Replicate API? No. It only reads publicly published model listing pages — it does not run predictions, and no Replicate API token is used or required.

What does runCount measure? The total number of times a model's API has been called, as published on the model's own Replicate page.

Why does search only match the trending catalog? Replicate's full-text search is a client-side JavaScript widget with no public unauthenticated API. search mode filters the same trending/featured catalog used by trending mode by keyword. For an exhaustive search of one creator's or one task's models, use byOwner or byCollection instead.

Why do some models lack a hardware or latestVersion field? Those fields are omitted (not returned as empty) when a model's page doesn't publish them — for example, some API-only models list CPU while GPU-hosted models list the exact GPU class.

How fresh is the data? Every run fetches live pages directly from replicate.com — there is no caching layer.