Early concept

Content removal
is not enough.

When harmful content is taken down, it doesn't disappear — it comes back, often in modified forms. Rechase tracks and prevents what happens after removal.

[ See how it works ↓ ]

All processing runs in your browser — no data leaves your device

The problem

Content removal doesn't solve the problem — it only delays it

—Content comes back
—Often in modified forms
—Spreads across platforms
—Enforcement is reactive
Result:ongoing harm
The insight

Takedown is a temporary fix — not a resolution

Current systems stop at removal.
They don't track what happens after.

—No persistence after removal
—No cross-platform visibility
—No behavioral tracking over time
The solution

Rechase focuses on the post-takedown lifecycle

Instead of just detecting content —
it tracks what happens after removal.

+Tracks removed content
+Detects variations
+Identifies repeated behavior
+Flags reappearance early
How it works

The post-takedown lifecycle

01

Content removed

A piece of harmful content is reported and taken down from the platform.

02

Signal captured

Rechase captures a unique signal of the removed content.

03

Reappearance detected

The same content comes back — modified, re-uploaded, or regenerated. Rechase recognizes it.

04

Flagged early

Rechase identifies it before it spreads.

Validation

What professionals say

“Takedown is often a first step rather than a final solution.”

— Legal & Privacy

“It becomes an ongoing issue rather than a one-time fix.”

— Trust & Safety
The Problem

Why exact matching fails

Your platform probably uses file hashing to detect duplicate content. Change a single pixel — the hash changes completely. Removed content comes back under a different hash.

Drop any image

PNG, JPG, WebP — stays on your device

∅

Drop an image to see its hash.

Detection

We catch every modification

Drop an image, apply any transformation, and watch pHash and CLIP match it while SHA-256 fails every time.

Drop image to fingerprint

This becomes the 'removed content' in the database

←

Drop an image to start

or use the sample image

Try It

Bring your own modification

Modify an image any way you want — Photoshop, phone, AI. Drop both files here.

We don't care how it was modified. We care what it looks like now.

Original

Original image

The 'removed' version

Modified anywhere you want

Modified image

Your version — cropped, filtered, AI-upscaled

LEVEL UP

Try to fool it
with AI

Take a portrait, regenerate it with AI — different angle, lighting, background. Drop both here. See if face recognition still finds the same person.

1

Upload original photo

Drop a portrait photo

Any photo with a visible face

2

Generate AI variant

Take the photo to ChatGPT, Midjourney, or any AI tool. Use this prompt:

Take the photo shown on the left and recreate it using AI. Keep the same person's identity and appearance, but change: the head angle or pose, the lighting direction, and the background. The result should look like a different photo of the same person taken in a different situation.

ChatGPT (GPT-4o) · Midjourney · Firefly · Grok ↗

AI Variants

Even AI-generated variants can't hide

When pHash fails on heavy AI transformations, CLIP embeddings catch what the human visual system sees as the same content — regardless of how the pixels changed.

AI ModificationSHA-256pHashCLIP
AI Upscale4× resolution boost via Stable Diffusion upscalerMISSMATCH (7)MATCH (0.94)
Style TransferMonet watercolor style applied via LoRAMISSMISS (28)MATCH (0.87)
InpaintingWatermark removed with SDXL inpaintingMISSMATCH (4)MATCH (0.96)
Background SwapSubject extracted, new background generatedMISSMISS (35)MATCH (0.82)
Face SwapIdentity swapped via deepfake pipelineMISSMISS (42)MATCH (0.83)
Color GradeProfessional LUT applied, saturation boostedMISSMATCH (6)MATCH (0.93)

CLIP catches 6/6 cases — including deepfakes and style transfers that break every pixel-based approach.

Dashboard

What happens after detection

Every re-upload attempt is logged with the detection method, confidence score, and timing — a complete record of what was caught and how.

dashboard.rechase.io

Re-uploads blocked

1,847

today

Detection rate

99.3%

across all methods

Avg latency

38ms

end-to-end

CLIP catches

31%

of daily blocks

Live block feed

LIVE
EventConfidenceAction
EVT-48210.3s ago94%✗ Blocked
EVT-482012s ago98%✗ Blocked
EVT-48191m ago87%✗ Blocked
EVT-48184m ago100%✗ Blocked
EVT-48179m ago96%✗ Blocked
Early access

Rechase is currently in validation

Built for platforms, Trust & Safety teams, and legal professionals.

[ Request early access ]