PicDefense
Identity & SecurityConnect AI agents to PicDefense, an image copyright and theft risk scanner. Agents check an image for copyright exposure before a flow publishes it, and read the analysis behind the score.
What This Integration Enables
PicDefense answers a question marketing teams usually answer with a shrug: where did this image come from, and is anyone else claiming it. It runs reverse image search, reads embedded EXIF copyright and artist tags, detects visible stock agency watermarks, and picks out faces, landmarks, logos, and content labels, then synthesizes all of it into a single Forensic Risk Score returned as picrisk with a value of high, medium, or low. Every action takes a publicly accessible image URL in jpg, png, or webp and PicDefense fetches and processes it server side, so the flow passes a link rather than a file.
Read the vendor's own framing of that score and then build to it, because it is unusually honest: the Forensic Risk Score is a triage signal, not a legal verdict. This connector sorts images into piles. It does not clear them. That is exactly the right shape for an agent, and it is why the integration belongs in a publishing flow rather than in a legal review. The agent handles what is mechanical, which is fetching the analysis for every asset in a queue and assembling the evidence into one readable summary. It hands back what is not, which is whether the license the company actually holds covers this particular use.
The cost model shapes flow design more than most connectors do. Check Image Risk consumes 0.80 account credits per successful call, Get Backlinks consumes 0.30, and every other analysis action consumes 0.03. Get Credits is free. So a nightly sweep across a large asset library is a spend decision, not a background task, and the cheap way to run one is a cascade: call the 0.03 actions first, let Detect Watermark and Get EXIF Data eliminate the obviously clean and the obviously licensed, and reserve the full Check Image Risk for the images that survive the first pass. There are no triggers on this connector. A flow calls it inline before it publishes, or on a schedule against a library it walks itself. Nothing here fires on its own.
Without FlowRunner
With FlowRunner
Use Case Scenarios
A publish step that will not run on an unchecked image
A content flow is ready to push a blog post with a hero image to Webflow. Before the publish action runs, the agent calls Get EXIF Data on the image URL, which returns any embedded copyright and artist tags plus a derived copyright flag. Then it calls Detect Watermark, which identifies a visible stock agency mark and returns an OCR snippet of the watermark text and a confidence value. An image carrying a stock agency watermark never reaches Check Image Risk, because the answer is already known and the 0.80 credits would tell nobody anything new. It goes straight to the brand owner with the agency name attached. Images with no watermark and no embedded copyright tag get the full Check Image Risk call, and only a picrisk of low proceeds to publish unattended.
A library sweep that respects its own budget
A team inherits several thousand images from an agency handover and nobody knows the provenance of any of them. The agent calls Get Credits first, which costs nothing, and works out how deep a sweep the balance supports. It then runs the cheap cascade across the whole library: Get EXIF Data and Detect Watermark at 0.03 each, plus Detect Logo to flag trademark considerations and Detect Face to flag anything with a person in it. Every result is written to Google Sheets as a row with the image URL, the flags, and the credit cost. Only images that come back with no watermark, no embedded copyright tag, and a detected face or logo get escalated to Get Backlinks and Check Image Risk, because those are the ones where the cheap signals disagree with each other. The team gets a triaged library instead of a folder, and they get it without exhausting the account on images that were obviously fine.
Finding who is reusing your own photography
The direction runs both ways. A brand with original product photography wants to know where its images have travelled. A monthly flow calls Get Backlinks on each hero image. That returns the pages where the image or a modified copy of it appears across the web, each with a similarity score from 0 to 100 and the direct image URL on that page, plus the list of domains it was found at. The agent filters out the company's own domains and its authorized retailers, then logs the remainder to a sheet and posts a summary to the brand channel in Slack. High similarity matches on unfamiliar commerce domains get flagged first. Nobody sends anything anywhere automatically, because a takedown notice is a claim a company makes about its own rights, and that is not a thing a similarity score decides.
Human-in-Loop Highlight
The gate worth building here is not the high band. A picrisk of high is easy, because it stops the flow and everyone agrees it should. The gate is the image that comes back low with a face in it. Detect Face reports whether an image contains a human face, and the vendor documents exactly why that matters: faces carry higher privacy release and model release requirements. A model release is a signed document sitting in a filing cabinet or a contracts folder. It is not in the pixels, it is not in the EXIF, and no reverse image search will find it. So a picrisk of low on an image containing a face means precisely one thing: PicDefense found no evidence of a copyright claim. It does not mean the person in the photograph agreed to appear in a paid ad. Treating those two statements as the same is how a company ends up with a problem that no amount of image analysis was ever going to catch. The agent therefore refuses to publish on a low band alone whenever Detect Face returns true. It posts the whole picture to the brand owner: "Image cleared at picrisk low, no watermark, no embedded copyright tag, 2 backlink matches both on our own domain. Detect Face returned true. Do we hold a model release for this shoot?" A person answers from the contracts folder, and the flow proceeds or stops on that answer. This is human-in-the-loop as a design principle rather than an approval step: the agent is not asking permission, it is escalating the specific thing its instruments cannot measure.
Agent Capabilities
11 actionsRisk Triage
2- Check Image Risk Runs the full multi-engine forensic analysis on a publicly accessible image URL and returns a synthesized copyright Forensic Risk Score as picrisk with a value of high, medium, or low. Combines reverse image backlinks, where-found domains, stock library detection, EXIF copyright tags, visible watermark detection, face, landmark, and logo detection, and content labels into a single triage result. The most complete operation and the most expensive, at 0.80 account credits per successful call.
- Get Credits Returns the number of analysis credits remaining on the authenticated account, and consumes none itself. Call it at the top of any sweep so the flow knows what it can afford before it starts spending.
Provenance
4- Get Backlinks Runs a reverse image search and returns the pages where the image or a modified copy of it appears across the web, each with a similarity score from 0 to 100 and the direct image URL on that page, plus the list of domains it was found at. The action for reconstructing provenance and for spotting unlicensed reuse of your own work. Consumes 0.30 credits per successful call.
- Get EXIF Data Extracts embedded EXIF metadata including camera make and model, original dimensions, capture and modify dates, editing software, and any embedded copyright and artist tags, returning a derived copyright flag and copyright holder when a copyright tag is present. Cheap at 0.03 credits and often decisive, which makes it a natural first call.
- Detect Watermark Detects a visible stock agency or photographer watermark and, when detected, identifies its likely source along with a short OCR snippet of the watermark text and a confidence value between 0 and 1. A detected agency watermark usually ends the investigation on its own. Consumes 0.03 credits.
- Extract Text Runs OCR and returns any text in the image as full text and as an array of individual words, with a flag for whether the extraction was truncated. Useful for reading copyright notices and watermark text that has been baked into an image rather than embedded in its metadata. Consumes 0.03 credits.
Content Detection
5- Detect Face Reports whether the image contains a human face. Faces carry higher privacy release and model release requirements, which makes this a compliance pre-check rather than a curiosity, and the trigger for the review path this page argues for. Consumes 0.03 credits.
- Detect Landmark Reports whether the image contains a recognizable landmark and returns its name when found. Consumes 0.03 credits.
- Detect Logo Reports whether the image contains one or more brand logos and returns any detected logo names. Detected logos indicate potential trademark considerations, which is a different review path from copyright and belongs with different people. Consumes 0.03 credits.
- Detect Labels Returns descriptive content labels detected in the image, both as a plain list and as label:confidence pairs. Useful for tagging an asset library and for routing images to reviewers by subject matter. Consumes 0.03 credits.
- Check Safe Search Returns a content safety assessment across the adult, spoof, medical, violence, and racy categories, each rated on a likelihood scale of VERY_UNLIKELY, UNLIKELY, POSSIBLE, LIKELY, or VERY_LIKELY. Worth running on anything user submitted before it reaches a public surface. Consumes 0.03 credits.
Frequently Asked Questions
What can FlowRunner do with PicDefense?
FlowRunner agents can run Check Image Risk, Get Backlinks, and Detect Watermark in PicDefense, plus 8 more actions.
Does connecting PicDefense to FlowRunner require OAuth?
No. PicDefense connects to FlowRunner with an API key, no OAuth flow required.
Can PicDefense trigger a FlowRunner workflow automatically?
PicDefense doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with PicDefense
$100 in credits. No card required. Connect in minutes.