# Vision-Language Binding in In-Context Image Generation

arXiv: https://arxiv.org/abs/2605.24624 · HTML: https://mlsanity.com/paper/2605.24624/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 20 of its 27 references that are in the map.
- First posted on arXiv 23 May 2026.

## Builds on (most cited first, 20 of 20)

- [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://mlsanity.com/paper/1505.04597/) (2015)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Scalable Diffusion Models with Transformers](https://mlsanity.com/paper/2212.09748/) (2022)
- [Scaling Rectified Flow Transformers for High-Resolution Image Synthesis](https://mlsanity.com/paper/2403.03206/) (2024)
- [InstructPix2Pix: Learning to Follow Image Editing Instructions](https://mlsanity.com/paper/2211.09800/) (2022)
- [Prompt-to-Prompt Image Editing with Cross Attention Control](https://mlsanity.com/paper/2208.01626/) (2022)
- [FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space](https://mlsanity.com/paper/2506.15742/) (2025)
- [Dissecting Recall of Factual Associations in Auto-Regressive Language Models](https://mlsanity.com/paper/2304.14767/) (2023)
- [OmniGen: Unified Image Generation](https://mlsanity.com/paper/2409.11340/) (2024)
- [How to use and interpret activation patching](https://mlsanity.com/paper/2404.15255/) (2024)
- [In-Context Edit: Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformer](https://mlsanity.com/paper/2504.20690/) (2025)
- [DreamBench++: A Human-Aligned Benchmark for Personalized Image Generation](https://mlsanity.com/paper/2406.16855/) (2024)
- [What's in the Image? A Deep-Dive into the Vision of Vision Language Models](https://mlsanity.com/paper/2411.17491/) (2024)
- [Localizing and Editing Knowledge in Text-to-Image Generative Models](https://mlsanity.com/paper/2310.13730/) (2023)
- [ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features](https://mlsanity.com/paper/2502.04320/) (2025)
- [Diffusion Lens: Interpreting Text Encoders in Text-to-Image Pipelines](https://mlsanity.com/paper/2403.05846/) (2024)
- [Padding Tone: A Mechanistic Analysis of Padding Tokens in T2I Models](https://mlsanity.com/paper/2501.06751/) (2025)
- [Group Relative Attention Guidance for Image Editing](https://mlsanity.com/paper/2510.24657/) (2025)
- [Follow the Flow: On Information Flow Across Textual Tokens in Text-to-Image Models](https://mlsanity.com/paper/2504.01137/) (2025)
- [I2AM: Interpreting Image-to-Image Latent Diffusion Models via Bi-Attribution Maps](https://mlsanity.com/paper/2407.12331/) (2024)

## Built on by (most cited first, 1 of 1)

- [Images Talk Back: Guiding MMDiTs by Suppressing Image-to-Text Feedback](https://mlsanity.com/paper/2610.13941/) (2026)

## Abstract (arXiv)

In-context image generation models such as FLUX.2 take a text prompt and an optional reference image as visual conditioning for the output. Internally, all three inputs -- text, reference image, and the noise tokens -- are concatenated and processed through a single attention stream, where all tokens can attend to one another. This leaves open how reference information flows through the model to produce the output image. We show that an implicit cross-modal binding emerges between the text tokens and the reference image: the text tokens absorb visual reference content during the forward pass, and that absorbed content causally influences the generated output. We surface this binding with three causal interventions on FLUX.2: T2I Lens, which decodes intermediate text-token activations through a text-to-image path; Attention Knockout, which severs specific attention edges; and I2I-to-I2I Patching, which copies text token activations between editing runs. Across 2,875 editing tasks on various images, including SUN397 and DreamBench++ datasets and images collected online, we observe a consistent division of labor: properties of the reference image, like color, style, and scene setting, are first written into the text tokens, which carry them to the generated image; pixel-exact properties like a specific face or instance identity bypass the text tokens and flow directly from reference to image through image-to-image attention. We further localize the reference-text binding to the padding tokens of the text sequence. These results show that text tokens in a multimodal DiT are not just prompt holders, but a structured channel for reference image content. More broadly, they suggest that even in unified-attention multimodal generative models, token modality structures how conditioning information is represented and routed across the network.