Marian Lupaşcu

Marian Lupascu, Sebastian Ripa, Mihai Trascau et al.

Personalizing text-to-image diffusion models to render several specific subjects in a coherent image remains challenging: the model must preserve each subject's identity while keeping the scene spatially and visually coherent. Methods that fuse independently trained concept adapters in a shared weight space (via federated averaging, gradient fusion, or orthogonality constraints) suffer from identity confusion and style bleeding and require joint retraining. In this work, we show that composing concepts as separate image layers, instead of merging their adapters in a shared weight space, avoids parameter-level interference. We introduce LILAC, a framework that composes independently trained low-rank adapters at inference time: each subject is conditioned on the frozen composite of previously placed subjects, with exactly one adapter active at a time, therefore identities never interfere at the parameter level. LILAC composes the adapters without any joint training, scales linearly with the number of concepts, and is backbone-agnostic. Under the Orthogonal Adaptation protocol, LILAC applied on Qwen-Image-Edit reaches an ArcFace detection rate of 0.861, while Orthogonal Adaptation reports 0.745 in its original setting. Adaptation reports 0.745 in its original setting. Code is available at https://github.com/marianlupascu/LILAC.

Marian Lupascu, Ryan Murdock, Ionut Mironica et al.

In this work, we propose a complete framework that generates visual art. Unlike previous stylization methods that are not flexible with style parameters (i.e., they allow stylization with only one style image, a single stylization text or stylization of a content image from a certain domain), our method has no such restriction. In addition, we implement an improved version that can generate a wide range of results with varying degrees of detail, style and structure, with a boost in generation speed. To further enhance the results, we insert an artistic super-resolution module in the generative pipeline. This module will bring additional details such as patterns specific to painters, slight brush marks, and so on.

Marian Lupascu, Nipun Jindal, Ionut Mironica et al.

Typography generation in diffusion models faces a persistent trade-off: enabling precise font control typically degrades text legibility, while maintaining readability often sacrifices typographic fidelity. We present FontFusion, a plug-and-play conditioning framework for Diffusion Transformer (DiT) architectures that resolves this dilemma through three core innovations: (1) a hierarchical token representation establishing explicit text-font relationships at multiple granularities, (2) position-aware embeddings creating spatial bindings between typography and image content, and (3) a multi-level token dropping strategy improving both computational efficiency and generalization to unseen fonts. Our systematic evaluation of font embedding spaces reveals that a dual encoder combining DeepFont and DINOv2 outperforms any single encoder for typography tasks. FontFusion demonstrates 76% relative improvement on challenging decorative fonts over single-encoder baselines and font consistency gains exceeding approximately 68-76% over unconditioned models, while integrating into existing DiT architectures without retraining.

Marian Lupascu, Ionut Mironica, Mihai-Sorin Stupariu

Creating visually appealing composites requires optimizing both text and background for compatibility. Previous methods have focused on simple design strategies, such as changing text color or adding background shapes for contrast. These approaches are often destructive, altering text color or partially obstructing the background image. Another method involves placing design elements in non-salient and contrasting regions, but this isn't always effective, especially with patterned backgrounds. To address these challenges, we propose a generative approach using a diffusion model. This method ensures the altered regions beneath design assets exhibit low saliency while enhancing contrast, thereby improving the visibility of the design asset.

Marian Lupascu, Ana-Cristina Rogoz, Mihai Sorin Stupariu et al.

In this survey, we systematically analyze techniques used to adapt large multimodal models (LMMs) for low-resource (LR) languages, examining approaches ranging from visual enhancement and data creation to cross-modal transfer and fusion strategies. Through a comprehensive analysis of 106 studies across 75 LR languages, we identify key patterns in how researchers tackle the challenges of limited data and computational resources. We find that visual information often serves as a crucial bridge for improving model performance in LR settings, though significant challenges remain in areas such as hallucination mitigation and computational efficiency. We aim to provide researchers with a clear understanding of current approaches and remaining challenges in making LMMs more accessible to speakers of LR (understudied) languages. We complement our survey with an open-source repository available at: https://github.com/marianlupascu/LMM4LRL-Survey.

Marian Lupascu, Mihai-Sorin Stupariu

Image editing in rectified flow models remains challenging due to the fundamental trade-off between reconstruction fidelity and editing flexibility. While inversion-based methods suffer from trajectory deviation, recent inversion-free approaches like FlowEdit offer direct editing pathways but can benefit from additional guidance to improve structure preservation. In this work, we demonstrate that optimal transport theory provides a unified framework for improving both paradigms in rectified flow editing. We introduce a zero-shot transport-guided inversion framework that leverages optimal transport during the reverse diffusion process, and extend optimal transport principles to enhance inversion-free methods through transport-optimized velocity field corrections. Incorporating transport-based guidance can effectively balance reconstruction accuracy and editing controllability across different rectified flow editing approaches. For inversion-based editing, our method achieves high-fidelity reconstruction with LPIPS scores of 0.001 and SSIM of 0.992 on face editing benchmarks, observing 7.8% to 12.9% improvements over RF-Inversion on LSUN datasets. For inversion-free editing with FlowEdit on FLUX and Stable Diffusion 3, we demonstrate consistent improvements in semantic consistency and structure preservation across diverse editing scenarios. Our semantic face editing experiments show an 11.2% improvement in identity preservation and enhanced perceptual quality. The unified optimal transport framework produces visually compelling edits with superior detail preservation across both inversion-based and direct editing paradigms. Code is available for RF-Inversion and FlowEdit at: https://github.com/marianlupascu/OT-RF

Marian Lupascu, Mihai Sorin Stupariu, Ionut Mironica

Score distillation turns a pretrained 2D diffusion model into a 3D generator, but the per-step gradient is estimated from a single randomly chosen view: it is high-variance and blind to global shape consistency. Prior work addresses this by retraining the diffusion prior on multi-view data; this improves consistency but makes the sampling contribution inseparable from prior quality. We instead isolate the sampling axis. The per-step gradient is one noisy sample of an expectation over views; aggregating K samples per step at a fixed total UNet budget reduces variance without touching the prior. We introduce Multi-View Aggregated Score Distillation (MV-SDI), which aggregates gradients from K views per step via gradient accumulation, keeping peak memory unchanged and the 2D prior frozen, and draws views as antithetic antipodal pairs, a prior-independent geometric property, for balanced angular coverage. At a fixed 10,000-UNet-call budget, K=2 raises CLIP R-Precision from 74.8% to 83.8% and CLIP score from 0.297 to 0.312, with consistent gains on HPSv2 and ImageReward and a 0.0% divergence rate on the 43-prompt benchmark; optimization steps halve as a consequence. K=4 gives a fourfold step reduction at R-Precision 86.9% and CLIP 0.307, still well above the single-view baseline on every alignment metric. MV-SDI is compatible with gradient-based score-distillation pipelines, including Score Distillation via Inversion, and requires no retraining and no multi-view data.

Andrei-Marius Avram, Marian Lupaşcu, Dumitru-Clementin Cercel et al.

This paper presents UniBERT, a compact multilingual language model that uses an innovative training framework that integrates three components: masked language modeling, adversarial training, and knowledge distillation. Pre-trained on a meticulously curated Wikipedia corpus spanning 107 languages, UniBERT is designed to reduce the computational demands of large-scale models while maintaining competitive performance across various natural language processing tasks. Comprehensive evaluations on four tasks - named entity recognition, natural language inference, question answering, and semantic textual similarity - demonstrate that our multilingual training strategy enhanced by an adversarial objective significantly improves cross-lingual generalization. Specifically, UniBERT models show an average relative improvement of 7.72% over traditional baselines, which achieved an average relative improvement of only 1.17%, and statistical analysis confirms the significance of these gains (p-value = 0.0181). This work highlights the benefits of combining adversarial training and knowledge distillation to build scalable and robust language models, thus advancing the field of multilingual and cross-lingual natural language processing.