Model reference — Pictures
This service hides a small piece of binary data — an identifier, a signature, a key, a serial number — inside a photograph so that the picture appears unchanged and the data can be read back by software.
Binary payloads differ from hidden text in that every bit must be recovered exactly: a single flipped bit invalidates a signature, where a single wrong character in hidden text is often still readable. That makes error-correcting codes essential, and they consume part of the capacity. Practical capacities are therefore small — tens to a few hundred bits for methods that survive recompression and reprinting, kilobytes for methods that require the file to arrive untouched. Steganography hides data; it does not protect it, so a confidential payload must be encrypted before embedding.
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 2-megapixel photograph with a 100-bit payload embedded and read back.
Vendor: Adobe Research
What it does: embeds a short payload that survives recompression, resizing and cropping and reads back reliably. The most robust general choice for identifiers.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 3,000 | ≈ 12,000 | ≈ 24,000 |
Vendor: University of California, Berkeley
What it does: embeds data that survives being printed and photographed again, which is what allows a physical print to carry a machine-readable identifier.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 2,000 | ≈ 8,000 | ≈ 18,000 |
Vendor: Stanford University
What it does: embeds and recovers a fixed-length payload with strong resistance to common transformations, in a very small model suited to high volume.
| 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) | ≈ 6,000 | ≈ 20,000 | ≈ 50,000 |
Vendor: Data to AI Lab, Massachusetts Institute of Technology
What it does: embeds a payload robust to heavy recompression and social-media processing, at a small cost in image quality.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 2,400 | ≈ 9,000 | ≈ 20,000 |
Vendor: community-maintained open-source project
What it does: a classical method needing no GPU, surviving ordinary recompression but not printing. By far the cheapest per image.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | No GPU required | No GPU required | No GPU required |
| VRAM | — | — | — |
| vCPUs | 2 | 8 | 32 |
| RAM | 4 GB | 16 GB | 64 GB |
| Server | CPU instance, 2 vCPU | CPU instance, 8 vCPU | CPU instance, 32 vCPU |
| Rate (images/hour) | ≈ 20,000 | ≈ 80,000 | ≈ 300,000 |
Vendor: community-maintained open-source projects
What it does: least-significant-bit tools that carry kilobytes rather than bits, but lose everything if the file is recompressed or resized. Suited to images transferred untouched.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | No GPU required | No GPU required | No GPU required |
| VRAM | — | — | — |
| vCPUs | 1 | 4 | 16 |
| RAM | 2 GB | 8 GB | 32 GB |
| Server | CPU instance, 1 vCPU | CPU instance, 4 vCPU | CPU instance, 16 vCPU |
| Rate (images/hour) | ≈ 40,000 | ≈ 160,000 | ≈ 600,000 |
The method follows from your payload size and the transformations it must survive, and those two pull against each other. Our consultants review your images, payload and distribution path, then recommend a method and report the measured bit-recovery rate on your own material.
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 Image steganography — hide small binary data in photo. 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.