PAPER: INTRODUCTIONCurrently, US copyright law offers no protection for artistic styles, likely due to challenges in defining style infringement and a prior lack of necessity. However, with highly capable text-to-image generative models transforming the art landscape, there have been increasingly more calls to reconsider this lack of protection. Namely, many fear that models like Stable Diffusion, Imagen, Mid-Journey, and DeepFloyd (Rombach et al., 2021;Saharia et al., 2022;DeepFloyd, 2023;Podell et al., 2024) may make replication of an artist's unique style as simple as providing an adequate prompt, potentially inundating the market with imitations that devalue the original artist's work and threaten their livelihood. This matter has gained widespread attention, as it is fundamentally interdisciplinary (engaging legal, artistic, and technical communities), and has serious material and human consequences. In fact, multiple legal cases regarding artistic copyright are ongoing (Brittain, 2024;Poritz, 2024), and Adobe (2023) has already called for new provisions in copyright law to protect artistic style. Thus, with generative AI creating a pressing need to answer previously unsettled questions, it is now critical for us to rethink artistic copyright.We begin by surveying existing copyright law around art, finding that it has historically relied heavily on qualitative judgments, such as the 'substantial similarity test' (Goldstein, 2014). The subjective nature of existing law has led to judgments that vary from case-to-case (e.g. on if characters are protected (DCC, 2015;MGM, 1995;Kli, 2014) or not (Sony, 1998)), which will prove problematic when dealing with AI-powered mimicry at unprecedented scale. An automatic quantitative approach could help make judgments quicker and more consistent. However, any technical solution must also be easy to understand and based in legal literature, as the audience whose decisions we intend to assist1 is largely non-technical (e.g. judges, juries, and artists). Thus, we set out to develop an intuitive, automatic, and legally-grounded manner to quantitatively argue artistic style infringement.Figure 1: Our primary contribution is an interpretable and quantitative framework, rooted in legal scholarship, for arguing style infringement from the perspective of classification. Given artworks by Canaletto, our tool, ArtSavant, automatically identifies a unique style and recognizes said style in generated art, summarizing its findings in an easy to understand yet quantitative report.Our work is structured around key questions we identify that, when answered, can form the basis of an argument for artistic style infringement. Namely, given an artist, i. do they have a unique style? If so, ii. to what degree does it appear in generated art? And lastly, iii. in what precise way (with respect to specific artistic elements present in both the original and generated art) is the style copied? Inspired by legal arguments presented by Sobel (2024) and O'Connor (2022), we first frame artistic style as characterized by a set of elements that co-occur frequently across an artist's body of work, where each element may not be protectable on its own, but together, they could represent an unique stylistic signature.To prove the uniqueness of an artistic style, we propose a simple logical argument, which serves as our core ideological contribution. Namely, if an artist's works are consistently recognized as their own, that artist must have some unique style, with which their work can consistently be classified back to them (instead of hundreds of other potential creators). Therefore, the task of showing the existence and uniqueness of artistic styles can be studied from the lens of classification -something deep networks are particularly adept at doing. To perform this classification, we curate a reference dataset of artworks from 372 artists, and implement two classification methods, taking holistic and analytic approaches.Current copyright law dictates that similarity between artworks must be evaluated in analytic and holistic terms (Tuf, 2003;Goldstein, 2014) (see § 2). For example, we could analyze Vincent Van Gogh's style as comprised of expressive wavy lines, bright unblended coloring, post-impressionism, choppy textured brushwork, etc, or we could make an intuitive, holistic judgement: e.g. in Figure 3, we can tell the look and feel of the generated images capture Van Gogh's style, even without articulating the shared stylistic elements. In order to make these notions more concrete and quantitative, we develop two complementary methods corresponding to holistic and analytic ways of evaluating style similarity.The first method -DeepMatch -is simply a neural network classifier, with which we show that for 89.2% of artists, their held-out works can consistently (i.e. over half the time) be mapped back to them (over 371 other artists). This represent a significant empirical finding, as it quantitatively shows that unique artistic styles exist for an overwhelming majority of artists we study -a necessary precondition for adding legal protections for artistic style. DeepMatch can be thought of as capturing neural signatures for each artist. Namely, the classification head vector consists of a combination of neural features that encode that artist's style. As these are not interpretable, we complement the holistic DeepMatch with TagMatch, a novel, inherently interpretable, attributable, analytic approach.Figure 2: We define artistic style as a set of elements (or signature) that appear frequently over a body of work, and reduce the problem of style copy detection to classification of sets of images to artists. (left) We offer proof-of-concept via two ways to recognize artistic styles over image set, including a novel inherently interpretable and attributable tag-based method. (right) In an empirical study of 372 prolific artists, we find generative models potentially copy artistic styles for 20.2% of these artists.TagMatch maps a collection of works to an artist by first assigning diverse stylistic tags in a zero-shot manner to each work, and then efficiently searching over the space of tag compositions to surface unique tag signatures. We show empirically that searching over tag compositions is critical, as no tag on its own is unique to an artist -recall this idea was described by legal scholars Sobel (2024) and O'Connor (2022); in our work, we make this quantitative. Further, importantly, TagMatch allows for articulating the specific combination of elements that uniquely appear in artist's works and in (infringing) generated work. Such interpretable output can be particularly useful for the legal settings where arguments around artistic copying ultimately are to be made.We package these methods in ArtSavant, a tool that for any artist can produce a report like Figure 1 in minutes. ArtSavant shows our logical framework to argue style infringement is practically feasible, automatically generating a quantitative assessment of style copying in a way that can be understood by the broad set of relevant stakeholders. Finally, we use ArtSavant to provide a quantitative snapshot of the prevalence of style copying with today's generative models under simple prompting. After generating images in the style of artists from our dataset with 3 popular text-to-image models and using ArtSavant to assess copying, we find 20% of the artists we study to be at risk, suggesting style copying may indeed by possible with today's generative AI.In summary, we make the following contributions:• We develop an automatic, interpretable, and legally-grounded manner to quantitatively argue artistic style infringement, leveraging a simple classification argument.• We introduce ArtSavant, a practical tool to quantify and communicate style copying for any artist, consisting of a reference dataset of artworks from 372 prolific artists, and two complementary methods, including a novel, highly interpretable and attributable one. Notably, our two methods directly relate to the principles of "analytic" and "holistic" style similarity used in the substantial similarity test in existing copyright law.• With ArtSavant, we perform a large-scale empirical study to measure style copying across 3 popular text-to-image generative models, finding that generated images (using simple prompting) from 20% of the artists examined appear to be at risk of style copying.
================================================================================

REVIEW
--------------------------------------------------------------------------------
# Summary Of The Paper

The paper presents **ArtSavant**, a novel tool aimed at addressing the growing legal and societal concern of artistic style infringement caused by generative AI models. The central thesis is that artistic style can be quantified and analyzed using a combination of **holistic** (neural-based) and **analytic** (tag-based) methods. The authors propose two complementary approaches: **DeepMatch**, a neural classifier that identifies an artist's unique style through image embeddings, and **TagMatch**, an interpretable method that detects style copying by analyzing combinations of stylistic tags. Together, these methods are integrated into **ArtSavant**, a practical tool that generates quantitative reports on the potential infringement of an artist's style by generative models. The paper conducts a large-scale empirical study on 372 artists from the **WikiArt** dataset, revealing that 20.2% of these artists may be at risk of style copying when generative models are prompted with basic instructions.

---

# Strengths

1. **Clear Motivation**: The paper is grounded in a compelling and timely legal and societal issue — the rise of generative AI and its potential to infringe on artistic styles. This provides strong relevance to both legal and technical communities.

2. **Dual Methodology**: The introduction of **DeepMatch** and **TagMatch** represents a comprehensive approach to detecting style copying. DeepMatch provides a holistic, high-accuracy method, while TagMatch introduces an interpretable and analytically grounded alternative — a balance often lacking in AI research.

3. **Legal Alignment**: The paper explicitly aligns its framework with the **substantial similarity test** used in U.S. copyright law, which enhances its applicability in legal contexts. It also acknowledges the importance of interpretability and clarity for non-expert users such as judges and lawyers.

4. **Empirical Study**: The authors conducted a large-scale study on 372 artists using three prominent text-to-image models. Their conclusion that 20.2% of artists are at risk of style copying under simple prompting adds empirical grounding to the discussion.

5. **Practical Tool Development**: The integration of both methods into **ArtSavant** is a tangible contribution, offering a user-friendly interface that can be deployed by artists to assess potential style copying.

---

# Weaknesses

1. **Lack of Statistical Rigor**: Many results are presented without accompanying **confidence intervals**, **p-values**, or **error margins**, which limits the reliability of the conclusions. For example, the reported 20.2% risk rate is stated without uncertainty bounds.

2. **Ambiguous Definition of "Unique Style"**: The paper defines an artist's style as a set of co-occurring elements, but this definition is not theoretically justified or validated. It relies almost entirely on **empirical results from DeepMatch**, without formal proofs or alternative measures.

3. **Overgeneralization Based on a Narrow Dataset**: The **WikiArt** dataset is heavily skewed toward Western classical art, with a focus on historical artists. This raises concerns about the **generalizability** of the findings to contemporary, non-Western, or digital artists.

4. **Limited Validation of TagMatch Outputs**: Despite the emphasis on **interpretability**, the paper provides insufficient validation of the **semantic meaning** and **correctness** of the atomic tags. The human study validates only **precision**, not **recall** or **completeness** of the tag assignments.

5. **Insufficient Error Analysis**: Both **DeepMatch** and **TagMatch** suffer from a lack of **confusion matrices**, **false positives/negatives breakdown**, and **case-by-case analysis**. This hinders a deeper understanding of the models’ performance and biases.

6. **No Discussion of False Positives/Negatives in Legal Context**: The paper assumes that the **tool can reliably detect infringement**, but it fails to address how **false positives** (incorrectly accusing an artist) or **false negatives** (failing to detect actual copying) might affect legal outcomes.

---

# Questions For The Authors

1. What is the **formal justification** for the claim that “if an artist's works are consistently recognized as their own, that artist must have some unique style”? Is this claim purely empirical, or is there a theoretical basis?

2. How is the **z-score threshold of 1.75** for assigning atomic tags determined? Is this threshold optimized via cross-validation, or is it arbitrary? Are there alternative thresholds that yield better performance or interpretability?

3. The paper states that **89.3% of artists are recognized by DeepMatch**, but only **20.2% are flagged as being at risk of style copying**. What accounts for this large discrepancy? Is it due to **model sensitivity** or **prompting strategy**?

4. How is the **“percentile of recognizability”** calculated in the ArtSavant report? Is this percentile derived from a **control group** or from the **distribution of all artists** in the dataset?

5. The **human validation study** indicates that only 5% of Safe artists’ generated images are judged to be similar to their style. How were the **annotators instructed** to interpret the task? Were they given **explicit criteria** for judging style similarity?

6. The paper cites **Sobel (2024)** and **O’Connor (2022)** in Section 4.1 but omits them in the **Related Work** section. Why is this omission made, and how does the paper situate itself in the broader scholarly conversation?

7. How does the paper **address intent and market harm**, which are essential components of the **substantial similarity test** in U.S. copyright law? How are these non-quantifiable factors handled in the proposed framework?

8. The **TagMatch method** is claimed to be the first **interpretable method** for detecting artistic style copying. However, **Concept Bottleneck Models (CBMs)** and **Grad-CAM** have been used for similar tasks. How does TagMatch differ in functionality and effectiveness from these existing methods?

9. The paper does not provide a **detailed explanation** of the **iterative tag composition algorithm** (Algorithm 1). What is the **computational complexity** of computing the power set of atomic tags, and how scalable is this approach for larger datasets?

10. How does the **method handle artists with very similar styles** (e.g., family members or contemporaries)? Does the framework account for this ambiguity in legal determinations?

---

# Limitations Not Addressed By The Authors

1. **Ethical Concerns**: The paper does not adequately address the **ethical implications** of labeling artists as “at risk” based on algorithmic assessments. This could lead to **false accusations** or **bias against certain artists**.

2. **Generalization Beyond Western Classical Art**: The **WikiArt** dataset is heavily biased toward Western classical artists. The paper does not address whether the findings are applicable to **digital artists**, **contemporary artists**, or **non-Western artists**.

3. **Interpretability vs. Accuracy Trade-off**: While **TagMatch** is emphasized for its interpretability, the **trade-off between accuracy and interpretability** is not thoroughly discussed. The paper does not compare the **utility** of DeepMatch versus TagMatch in legal or artistic contexts.

4. **Absence of Ground Truth for Style Copying**: The paper assumes that **generated images that match an artist’s style in DeepMatch** correspond to **actual style copying**, but this is not validated against **ground truth** or **expert judgment**.

5. **Lack of Reproducibility Information**: The paper does not provide **code availability**, **data access**, or **reproducibility guidelines**, limiting independent verification of the results.

---

# Soundness: 3 (Good)

While the paper presents a coherent and well-motivated approach to detecting artistic style copying, several **key limitations** undermine the **soundness** of the methodology. The **lack of statistical rigor**, **ambiguous definitions**, and **limited validation** of the tag-based method weaken the credibility of the results. Although the **large-scale empirical study** is commendable, the absence of **confidence intervals**, **cross-validation**, and **error analysis** prevents a thorough evaluation of the framework's reliability.

---

# Contribution: 3 (Good)

The paper contributes a **novel tool (ArtSavant)** that combines **DeepMatch** and **TagMatch** for detecting artistic style copying. Its **alignment with legal standards** and **focus on interpretability** are notable. However, the **contribution is somewhat incremental** given the **existing literature on image similarity, style transfer, and concept bottleneck models**. More work is needed to establish **theoretical foundations** and **broader applicability**.

---

# Confidence: 3 (Moderately Confident)

The paper presents a **clear framework** and **empirical results**, but the **lack of statistical rigor**, **validation of assumptions**, and **replicability information** leaves gaps in confidence. The **authors are moderately confident** in the validity of the framework, but **additional experimental and theoretical work** is required for full trust.

---

# Rating: 7 (Accept)

The paper presents a **valuable and timely contribution** to the intersection of **AI, copyright law, and artistic creativity**. While the **results are promising**, the **limitations in rigor, validation, and generalization** prevent a **stronger rating**. The **framework is acceptable for publication**, pending improvements in **experimental rigor**, **theoretical grounding**, and **ethical considerations**.

---

# Brief Justification For Rating

The paper introduces a **novel and practical framework** for detecting artistic style copying using **both holistic and analytic methods**, with **real-world applications in legal and artistic domains**. However, the **lack of statistical rigor**, **ambiguity in definitions**, and **limited validation** prevent a **higher rating**. The **results are promising**, but **additional experimentation and theoretical refinement** are needed to strengthen the contribution.

================================================================================
