Research map / Vision-language and multimodal / Multimodal LLMs
X-Fusion: Introducing New Modality to Frozen Large Language Models
Sicheng Mo, Thao Nguyen, Xun Huang, Siddharth Srinivasan Iyer, Yijun Li, Yuchen Liu, Abhishek Tandon, Eli Shechtman, Krishna Kumar Singh, Yong Jae Lee, Bolei Zhou, Yuheng Li
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 Dual Tower (frozen text tower plus trainable vision tower) instead of fine-tuning a single tower (ablation) | X-Fusion · COCO text-to-image | FID: 19.10 -> 14.20 | Improved |
| keep language weights frozen with a separate vision tower (ablation) | X-Fusion · MMLU 5-shot | accuracy: 25.0 -> 32.2 | Improved |
| use Dual Tower instead of Gated Tower (ablation) | X-Fusion · COCO captioning | BLIP-ITM: 14.5 -> 31.3 | Improved |
| use Dual Tower instead of Dual Projection (ablation) | X-Fusion · COCO text-to-image | FID: 20.22 -> 14.20 | Improved |
| use clean images (no noise) in image-to-text samples (ablation) | X-Fusion · COCO generation and understanding | FID and BLIP | Improved |
| increase image understanding (I2T) data share (ablation) | X-Fusion · COCO text-to-image | FID | Improved |
| add text-to-image generation data to help understanding (ablation) | X-Fusion · COCO captioning | BLIP-ITM | No clear effect |
| add CLIP feature alignment loss (ablation) | X-Fusion 1B and 3B · COCO generation and understanding | FID and BLIP | Improved |
| add CLIP feature alignment loss (ablation) | X-Fusion 8B · COCO generation and understanding at 100k iterations | FID and BLIP | Hurt |
| add X-Fuse layer fusing features from both towers (ablation) | X-Fusion · COCO generation and understanding | FID and BLIP | Improved |
| initialize vision tower from pretrained DiT | X-Fusion 8B · image generation | qualitative | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- Janus — Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation (CVPR 2025) · better on 0 of 1 shared result
- MetaMorph — MetaMorph: Multimodal Understanding and Generation via Instruction Tuning (ICCV 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LMFusion: Adapting Pretrained Language Models for Multimodal Generation (NeurIPS 2025 · similarity 0.90)
- SPHINX-X: Scaling Data and Parameters for a Family of Multi-modal Large Language Models (ICML 2024 · similarity 0.89)
- Multi-Layer Visual Feature Fusion in Multimodal LLMs: Methods, Analysis, and Best Practices (CVPR 2025 · similarity 0.89)
- LoMo: Local Modality Substitution for Deeper Vision-Language Fusion (NeurIPS 2026 · similarity 0.89)
- Where Does Vision Meet Language? Understanding and Refining Visual Fusion in MLLMs via Contrastive Attention (CVPR 2026 · similarity 0.89)
- Vision as a Dialect: Unifying Visual Understanding and Generation via Text-Aligned Representations (NeurIPS 2025 · similarity 0.89)
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.
