Model reference — Pictures
This service finds images that are the same or nearly the same — the identical photograph uploaded twice, a resized or cropped copy, a re-photographed version, or simply a picture of the same object from a similar angle.
It works by turning each image into a numeric fingerprint and looking for fingerprints that sit close together. The threshold decides what "similar" means, and that is a business question: a claims team hunting for the same damage photograph submitted twice needs a strict setting, while a catalogue team looking for pictures of the same product needs a loose one. Fingerprints are computed once per image, so the ongoing cost is small even across millions of pictures.
Models in this group take a single image as input: a photograph, a scan or a screenshot. Each specification table gives three hardware tiers — Minimum, the smallest setup on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is the number of sample inputs processed per hour on that hardware. Use these rates for initial sizing. Before production, benchmark your own data to validate accuracy, latency, throughput and cost. The sample input here is one photograph at about 2 megapixels.
Vendor: Meta
What it does: produces fingerprints that group pictures of the same object even when the angle, lighting and background change. The strongest general choice for visual similarity.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 8,000 | ≈ 30,000 | ≈ 85,000 |
Vendor: OpenAI
What it does: fingerprints images in a way that also allows searching them with a written description, so one index serves both duplicate detection and text search.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 6 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 15,000 | ≈ 60,000 | ≈ 170,000 |
Vendor: Google
What it does: a stronger model of the same kind, more reliable at separating images that are genuinely similar from images that merely share a subject.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 12,000 | ≈ 48,000 | ≈ 140,000 |
Vendor: Meta
What it does: fingerprints tuned to fine visual detail, which is what distinguishes two near-identical products or two prints of the same document.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 12,000 | ≈ 48,000 | ≈ 140,000 |
Vendor: Google
What it does: general image fingerprints usable for similarity, classification and search from the same single pass over the collection.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 10,000 | ≈ 40,000 | ≈ 110,000 |
Vendor: Microsoft
What it does: a classical method needing no GPU that finds exact and lightly edited copies extremely cheaply. It misses anything reshot or re-cropped, so it is used as the free first pass before a model looks at the rest.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 4 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 8 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 200,000 | ≈ 800,000 | ≈ 2,400,000 |
The right model and threshold depend on what you consider a duplicate and how the images have been altered. Our consultants test the candidates against known duplicates in your own collection, then recommend a model and a threshold with its measured miss and false-match rates.
At the start of a project we may run a short proof of concept on a sample of your own data, measuring the accuracy and the throughput the model actually achieves on your material. That replaces the estimates on this page with real figures, so the cost and the schedule for the full engagement are known before it is committed.
Share a representative sample, expected volume, latency target and deployment location for Visual similarity / duplicate image detection. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster.