AI image generators can reproduce copyrighted images from their training data.
Researchers have extracted memorised training images from diffusion models and shown near-copies appearing in output, especially for images repeated in the training data. The US Copyright Office has reviewed the same evidence. The claim is that reproduction can happen, not that most outputs are copies or that any particular output infringes copyright.
Current
Cases on this claim
Supporting
1
Disputing
0
1 case from 1 contributor. These numbers count cases people have written. They are not votes, and Koinovera does not decide whether the claim is true.
Original case
The case Psami put this claim forward with on 6 Oct 2026. It is the publisher's case, open to the same scrutiny as every other.
+Supports
Analyses replication of training images and identifies training-data duplication as a contributing factor. Supports the technical possibility of near-copy outputs.[1]
Summary of the cases and their sourcesAI analysisGenerated 6 Oct 2026
AI summary
Written by AI from the published cases and the sources they cite. It describes what each side argues and what the readable sources show, and is not a ruling from Koinovera. The generation date does not guarantee that the claim wording or sources are unchanged. It only changes when someone asks for an update.
AI generatedCurrent evidence: Mostly supportive
An AI reading of the cases so far and the sources they cite. The label above rests on the sources alone, not on what the cases assert or how many take a side. It is not a verdict from Koinovera on whether the claim is true, and it may change as evidence is added or questioned.
Research indicates that diffusion-based image generation models can replicate content from their training datasets, with some models producing near-copies of training samples. [1][2] Studies have demonstrated that these models can regurgitate training images even when datasets have undergone de-duplication processes. [2]
Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Models
arXiv.org
View PDF
HTML (experimental)
Abstract:Cutting-edge diffusion models produce images with high quality and customizability, enabling them to be used for commercial art and graphic design purposes. But do diffusion models create unique works of art, or are they replicating content directly from their training sets? In this work, we study image retrieval frameworks that enable us to compare generated images with training samples and detect when content has been replicated. Applying our frameworks to diffusion models trained on multiple datasets including Oxford flowers, Celeb-A, ImageNet, and LAION, we discuss how factors such as training set size impact rates of content replication. We also identify cases where diffusion models, including the popular Stable Diffusion model, blatantly copy f
A Reproducible Extraction of Training Images from Diffusion Models
arXiv.org
View PDF
HTML (experimental)
Abstract:Recently, Carlini et al. demonstrated the widely used model Stable Diffusion can regurgitate real training samples, which is troublesome from a copyright perspective. In this work, we provide an efficient extraction attack on par with the recent attack, with several order of magnitudes less network evaluations. In the process, we expose a new phenomena, which we dub template verbatims, wherein a diffusion model will regurgitate a training sample largely in tact. Template verbatims are harder to detect as they require retrieval and masking to correctly label. Furthermore, they are still generated by newer systems, even those which de-duplicate their training set, and we give insight into why they still appear during generation. We extract training ima
Every argument made about this claim, each with its own reasoning and sources.
Strongest orders cases by how readers have rated their reasoning, weighted by the standing of those readers and lowered when a case rests on poorly rated sources. Cases nobody has rated sit in the middle, newest first. It never counts how many cases or sources there are.