Unified model generates images and text together with improved planning
JUICE: Unified Multimodal Modeling in Pretrained Representation Spaces
Computer Vision and Pattern Recognition
Summary
Most systems that create both images and text do them one after the other, which can limit how well they work together. The authors developed a model called JUICE that creates images and text simultaneously by learning both at the same time in a shared space. This approach lets the model do tasks like making pictures, writing captions, answering questions about images, and even planning routes on maps better than previous methods. Testing shows that this joint creation improves success on complex tasks like maze solving and visual planning.
What this means in practice
- •For multimedia application developers: Create systems that generate coordinated images and text together for enhanced user experiences in storytelling or interactive media.
- •For robotics navigation teams: Use joint image and text modeling to improve visual spatial planning and route description for autonomous robots or assistive devices.
Authors
Jiahan Li, Huangyuan Su, Shengbang Tong, Shusheng Yang, Ruihan Guo, Suozhi Huang, Hexu Zhao, Pratyusha Sharma, Sherry Yang, Jiatao Gu, Rob Fergus, Kyunghyun Cho
Abstract
Most unified multimodal models treat images and text asymmetrically: a pretrained language model generates text, an attached diffusion head generates images, and outputs that contain both are produced in a fixed order, so the first output cannot adapt to the second. We ask whether both modalities can instead be generated by one process that is trained on them jointly from scratch. Pretrained encoder spaces hold both modalities in one continuous form and carry knowledge learned at scale, and recent work finds them effective as generative targets for vision and language. We therefore model the joint distribution directly in these spaces. We introduce JUICE (Joint Unified multimodal modeling In Continuous Encoder spaces), a single flow matching Transformer over frozen image and text representations, in which each modality has its own noise level. Which modalities are clean and which are generated then decides the task: image generation, captioning, visual question answering, or joint generation. Controlled pretraining runs test each design choice. At equal guidance, generating text as continuous states leaves image generation intact (48.41 GenEval, against 47.86 for a model trained only for text-to-image generation), while captions lose some exact wording. Training on joint and conditional denoising together beats either alone, and semantic encoder spaces beat VAE latents and raw pixels as targets. At 2.1B parameters, with no language model initialization and at 224x224, JUICE reaches up to 0.88 on GenEval and 73.1 on VQAv2. By sampling both modalities from the joint distribution (coupled sampling), it can also draw a route on a map while writing out the matching moves. On mazes with several valid solutions, joint success rises from 54.6% with independent sampling to 93.7%, and no modality has to go first. On visual spatial planning, JUICE solves 99.8% of maps at its training sizes.