mlsanityData to 7 Oct 2026

Iteratively Prompting Multimodal LLMs to Reproduce Natural and AI-Generated Images

arXiv 2404.13784 ↗First posted 21 Apr 2024cs.CR cs.CL cs.CVcited by 9

On the map: Multimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness

Abstract (arXiv)

With the digital imagery landscape rapidly evolving, image stocks and AI-generated image marketplaces have become central to visual media. Traditional stock images now exist alongside innovative platforms that trade in prompts for AI-generated visuals, driven by sophisticated APIs like DALL-E 3 and Midjourney. This paper studies the possibility of employing multi-modal models with enhanced visual understanding to mimic the outputs of these platforms, introducing an original attack strategy. Our method leverages fine-tuned CLIP models, a multi-label classifier, and the descriptive capabilities of GPT-4V to create prompts that generate images similar to those available in marketplaces and from premium stock image providers, yet at a markedly lower expense. In presenting this strategy, we aim to spotlight a new class of economic and security considerations within the realm of digital imagery. Our findings, supported by both automated metrics and human assessment, reveal that comparable visual content can be produced for a fraction of the prevailing market prices (0.23 - 0.27 per image), emphasizing the need for awareness and strategic discussions about the integrity of digital media in an increasingly AI-integrated landscape. Our work also contributes to the field by assembling a dataset consisting of approximately 19 million prompt-image pairs generated by the popular Midjourney platform, which we plan to release publicly.

Key facts. Cited by 9 papers in the map (data through 7 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 29 of its 32 references that are in the map. First posted on arXiv 21 Apr 2024.

Builds on

29 of 32 references.

References · 29

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  2. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  3. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  4. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,322
  5. 2023BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Modelscited by 6,069
  6. 2022BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generationcited by 4,310
  7. 2021Zero-Shot Text-to-Image Generationcited by 4,276
  8. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,131
  9. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,388
  10. 2022Scaling Autoregressive Models for Content-Rich Text-to-Image Generationcited by 1,287
  11. 2022PaLI: A Jointly-Scaled Multilingual Language-Image Modelcited by 762
  12. 2016Generative Adversarial Text to Image Synthesiscited by 735
17 more
  1. 2021CogView: Mastering Text-to-Image Generation via Transformerscited by 691
  2. 2017AttnGAN: Fine-Grained Text to Image Generation with Attentional Generative Adversarial Networkscited by 664
  3. 2016StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networkscited by 575
  4. 2022GIT: A Generative Image-to-text Transformer for Vision and Languagecited by 536
  5. 2023Tree-Ring Watermarks: Fingerprints for Diffusion Images that are Invisible and Robustcited by 179
  6. 2023StableRep: Synthetic Images from Text-to-Image Models Make Strong Visual Representation Learnerscited by 177
  7. 2023A Recipe for Watermarking Diffusion Modelscited by 136
  8. 2015Generating Images from Captions with Attentioncited by 135
  9. 2023EditGuard: Versatile Image Watermarking for Tamper Localization and Copyright Protectioncited by 109
  10. 2023Effective Prompt Extraction from Language Modelscited by 83
  11. 2020X-LXMERT: Paint, Caption and Answer Questions with Multi-Modal Transformerscited by 58
  12. 2022A Taxonomy of Prompt Modifiers for Text-To-Image Generationcited by 54
  13. 2023Prompt Stealing Attacks Against Text-to-Image Generation Modelscited by 48
  14. 2024Prompt Stealing Attacks Against Large Language Modelscited by 39
  15. 2023Language Models as Black-Box Optimizers for Vision-Language Modelscited by 36
  16. 2022StyleT2I: Toward Compositional and High-Fidelity Text-to-Image Synthesiscited by 33
  17. 2023BeautifulPrompt: Towards Automatic Prompt Engineering for Text-to-Image Synthesiscited by 15

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 3; 2025: 5; 2026: 1.

9 papers.

  1. 2024Backdooring Bias (B2) into Stable Diffusion Modelscited by 12
  2. 2025Vulnerability of Text-to-Image Models to Prompt Template Stealing: A Differential Evolution Approachcited by 5
  3. 2024Data-Centric AI in the Age of Large Language Modelscited by 2
  4. 2025PROMPTMINER: Black-Box Prompt Stealing against Text-to-Image Generative Models via Reinforcement Learning and Fuzz Optimizationcited by 2
  5. 2024Red-Teaming Text-to-Image Models via In-Context Experience Replay and Semantic-Preserving Prompt Rewritingcited by 1
  6. 2025EDITOR: Effective and Interpretable Prompt Inversion for Text-to-Image Diffusion Modelscited by 0
  7. 2025Reinforcement Learning-Based Prompt Template Stealing for Text-to-Image Modelscited by 0
  8. 2025Text-to-Image Models Leave Identifiable Signatures: Implications for Leaderboard Securitycited by 0
  9. 2026Identifying Models Behind Text-to-Image Leaderboardscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Towards Effective Prompt Stealing Attack against Text-to-Image Diffusion Models
  2. 2025LLM-I: LLMs are Naturally Interleaved Multimodal Creators
  3. 2024Evaluating the Efficacy of Prompt-Engineered Large Multimodal Models Versus Fine-Tuned Vision Transformers in Image-Based Security Applications
  4. 2025Using LLMs as prompt modifier to avoid biases in AI image generators
  5. 2026Image-based Prompt Injection: Hijacking Multimodal LLMs through Visually Embedded Adversarial Instructions
  6. 2026Penny Wise, Pixel Foolish: Bypassing Price Constraints in Multimodal Agents via Visual Adversarial Perturbations
  7. 2024Harnessing GPT-4V(ision) for Insurance: A Preliminary Exploration
  8. 2025Safer Prompts: Reducing Risks from Memorization in Visual Generative AI