Research map / Vision-language and multimodal / Human and embodied modeling
Generating Images with Multimodal Language Models
Jing Yu Koh, Daniel Fried, Ruslan Salakhutdinov
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use GILL (LLM encoder + GILLMapper) instead of Stable Diffusion text encoder, with 5 captions | Stable Diffusion backbone · VIST | CLIP Similarity: 0.598 -> 0.612 | Improved |
| provide full multimodal context (5 captions, 4 images) instead of 5 captions | GILL · VIST | CLIP Similarity: 0.612 -> 0.641 | Improved |
| GILL vs Stable Diffusion with a single caption | GILL · VIST 1 caption | CLIP Similarity: 0.592 -> 0.581 | Hurt |
| GILL vs Stable Diffusion with 10 dialogue rounds | GILL · VisDial | CLIP Similarity: 0.622 -> 0.645 | Improved |
| GILL vs Stable Diffusion with 1 dialogue round | GILL · VisDial | CLIP Similarity: 0.552 -> 0.528 | Hurt |
| add retrieval loss to training objective (ablation) | GILL · VIST (5 captions, 4 images) | CLIP Similarity: 0.636 -> 0.641 | No clear effect |
| use GILLMapper transformer instead of linear, MLP or transformer-encoder mapping (ablation) | GILL · CC3M val FID and VIST CLIP Similarity | FID / CLIP Similarity | Improved |
| increase number of [IMG] tokens r (ablation) | GILL · image generation | generation quality | Improved |
| GILL vs Stable Diffusion on MS-COCO captions | GILL · MS-COCO 2014 val | FID: 9.22 -> 12.2 | Hurt |
Closest papers by meaning
- LMFusion: Adapting Pretrained Language Models for Multimodal Generation (NeurIPS 2025 · similarity 0.93)
- TIGeR: Unifying Text-to-Image Generation and Retrieval with Large Multimodal Models (ICLR 2025 · similarity 0.91)
- A Comprehensive Study of Decoder-Only LLMs for Text-to-Image Generation (CVPR 2025 · similarity 0.91)
- The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models (NeurIPS 2025 · similarity 0.90)
- MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer (ICLR 2026 · similarity 0.90)
- Vision as a Dialect: Unifying Visual Understanding and Generation via Text-Aligned Representations (NeurIPS 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
