Semantic Generative Tuning (SGT) for Unified Multimodal Models
This paper introduces Semantic Generative Tuning (SGT), a post-training paradigm for unified multimodal models (UMMs) that bridges the gap between visual understanding and visual generation. The authors find that image segmentation tasks serve as optimal generative proxies, providing structural semantics that improve both perception and generative layout fidelity. SGT aligns representation spaces across understanding and generation objectives, improving feature linear separability and visual-textual attention allocation. Evaluations show consistent gains on multimodal comprehension and generative fidelity benchmarks.
Related guides (3)
Related events (8)
PGT: Procedurally Generated Tasks for Improving Visual Grounding in MLLMs
This paper introduces Procedurally Generated Tasks (PGT), a data-driven framework that overlays geometric primitives on images to create dense supervision signals for fine-grained visual grounding in multimodal large language models. PGT serves both as a training augmentation method and a diagnostic tool to isolate perception failures from semantic priors. Instruction tuning on LLaVA-v1.5-Instruct augmented with PGT data yields gains of up to +20% on the What'sUp benchmark and +13.3% on CV-Bench-2D. The results suggest that spatial reasoning deficits in MLLMs stem primarily from inadequate supervision rather than architectural or resolution constraints.
Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers
Hugging Face published a blog post detailing how to train and finetune multimodal embedding and reranker models using the Sentence Transformers library. The post covers techniques for building models that can jointly embed text and images for retrieval and reranking tasks. This represents an extension of the Sentence Transformers ecosystem into multimodal territory, enabling practitioners to build cross-modal search and ranking systems.
Meta Research Improves Image Generation via Staged Planning and Self-Revision Fine-Tuning
Researchers from Meta and collaborating universities propose a fine-tuning method that teaches image generators to compose images through discrete plan-sketch-inspect-refine cycles rather than generating all at once. Starting from BAGEL-7B, they construct ~62,000 training examples using GPT-4o and FLUX.1 Kontext to supervise each stage, achieving 83% on GenEval versus 77% for the base model and a competing method (PARM) that required 11x more training data and ~8x more inference steps. The approach improves spatial relationship accuracy, object attribute fidelity, and real-world knowledge grounding in generated images.
Joint Energy-Based Models Reveal a Generative-Discriminative Sweet Spot for Human-Aligned Vision
Researchers use Joint Energy-Based Models (JEMs) to isolate the effect of learning objective—independent of architecture, scale, and data—on human alignment in visual representations. By varying a single mixing coefficient between discriminative and generative training, they evaluate models across six human-alignment benchmarks and find that alignment peaks at intermediate points on the generative-discriminative continuum rather than at either extreme. The results suggest that hybrid objectives combining categorical structure from discriminative learning with input-structure sensitivity from generative learning yield the most human-like visual behavior. This challenges the framing of generative vs. discriminative as a binary choice for building human-aligned vision systems.
MAGIC: Multimodal Alignment & Grounding-aware Instruction Coreset for Vision-Language Models
MAGIC is a training-free coreset selection method for multimodal instruction tuning that uses three intrinsic signals—Multimodal Gain, Bridging Relevance, and Skill-Neuron Signatures—to identify compact, behaviorally faithful training subsets without backpropagation. The method operates in a three-stage pipeline: filtering low-gain examples, ranking by a quality objective, and bucket-wise budget allocation over neuron signatures. On LLaVA-665K and Vision-Flan datasets with 20% data budgets, MAGIC matches or slightly exceeds full fine-tuning performance (100.3% and 101.6% relative) while reducing wall-clock training time by 73.7%. Results transfer to LLaVA-1.5-7B and -13B target models.
G2Rec: Scalable framework unifying graph-based user modeling with semantic tokenization for generative recommendation
Researchers propose G2Rec, a framework that combines holistic graph-based user co-engagement modeling with semantic tokenization for industrial-scale generative recommendation systems. The approach addresses limitations of existing methods—scalability issues in graph serialization and lack of supervision in semantic tokenization—by learning user interest prototypes without ground-truth labels. The system has been deployed in production across product surfaces and evaluated on public datasets, showing improvements over prior methods.
Image GPT: Transformer Models Applied to Pixel Sequences for Image Generation and Classification
OpenAI demonstrates that a large transformer model trained autoregressively on pixel sequences can generate coherent image completions and samples, analogous to text generation. The work establishes a correlation between generative sample quality and downstream image classification accuracy. The best generative model achieves features competitive with top convolutional networks in the unsupervised setting, suggesting shared representational principles across modalities.
Multimodal Embedding & Reranker Models with Sentence Transformers
Hugging Face's Sentence Transformers library has added support for multimodal embedding and reranking models, enabling joint text-image (and potentially other modality) representations within a unified framework. The update extends the library's existing text-focused embedding capabilities to handle cross-modal retrieval and reranking tasks. This lowers the barrier for practitioners building multimodal search and RAG pipelines using open-weights models.