Latent Reasoning in Continuous Space for Unified Multimodal Models
Latent Reasoning in Continuous Space for Unified Multimodal Models
The LARC framework alleviates the bottleneck of applying linguistic reasoning in image generation for unified multimodal models by reasoning on continuous hidden states, improving generation quality and achieving test-time scaling without discrete token bottlenecks. The framework includes two stages: curriculum supervised fine-tuning (SFT) and information gain reinforcement learning (RL). In text-to-image generation benchmark tests, LARC improved performance by 7% over BAGEL on GenEval and demonstrated clear test-time scaling behavior.
Published on Oct 30
Authors:
,
,
,
Abstract
LARC enables continuous latent reasoning for unified multimodal models, improving image generation quality and enabling test-time scaling without discrete token bottlenecks.
Unified Multimodal Models (UMMs) achieve remarkable success on diverse multimodal understanding and generation tasks, yet they still struggle to effectively leverage language-based chain-of-thought reasoning for image generation. As a result, existing reasoning-augmented UMMs often fail to inherit the test-time scaling and fine-grained controllability observed in large language models. We attribute this failure to a fundamental modality mismatch: reasoning is performed in a discrete token space, whereas image generation operates in a continuous space. To bridge this gap, we introduce a Latent Reasoning framework in Continuous space (LARC), which alleviates the token-level bottleneck by reasoning over continuous hidden states. Specifically, the training framework consists of two stages: (i) curriculum supervised fine-tuning (SFT), and (ii) information-gain reinforcement learning (RL). First, the curriculum SFT stage gradually converts token-level reasoning into latent reasoning. Subsequently, the information-gain RL stage self-evolves latent reasoning without ground truth supervision. Experiments across diverse text-to-image generation benchmarks show that LARC consistently improves generation quality, prompt following, and compositional fidelity. Notably, LARC achieves state-of-the-art performance on GenEval with a 7%p gain over BAGEL and demonstrates clear test-time scaling behavior, outperforming token-based generation under both parallel and sequential test-time scaling strategies.
View arXiv page View PDF Project page GitHub 2 Add to collection
Get this paper in your agent:
hf papers read 2610.42149
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash
Models citing this paper 0
No model linking this paper
Cite arxiv.org/abs/2610.42149 in a model README.md to link it from this page.
Datasets citing this paper 0
No dataset linking this paper
Cite arxiv.org/abs/2610.42149 in a dataset README.md to link it from this page.
Spaces citing this paper 0
No Space linking this paper
Cite arxiv.org/abs/2610.42149 in a Space README.md to link it from this page.
Collections including this paper 0
No Collection including this paper
Add this paper to a collection to link it from this page.
Source: Hugging Face Daily Papers · huggingface.co