OmniAlpha: A Sequence-to-Sequence Framework for Unified Multi-Task RGBA Generation

2025-11-25

AI Summary

Existing generative models predominantly focus on RGB synthesis and lack unified support for RGBA image generation and editing; single-task alpha models suffer from poor generalization, while mainstream multi-task frameworks omit the transparency (alpha) channel. Method: We propose the first sequence-to-sequence capable unified RGBA multi-task framework, covering 21 diverse generation and editing tasks. Our approach introduces bidirectionally extended MSRoPE-BiL positional encoding, constructs a high-fidelity multi-layer AlphaLayers dataset, and employs a Diffusion Transformer to enable joint multi-input–multi-target modeling. Results: On AIM-500, our method achieves an 84.8% relative reduction in Structural Alpha Distance (SAD); for layer-conditioned alpha completion, it attains over 90% human preference—consistently outperforming task-specific baselines across all evaluated dimensions.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Responsible Web: Algorithmic accountability and transparency on the web
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Generative models have excelled in RGB synthesis, but real-world applications require RGBA manipulation. This has led to a fragmented landscape: specialized, single-task models handle alpha but lack versatility, while unified multi-task frameworks are confined to the RGB domain. To bridge this critical gap, we propose OmniAlpha, the first unified, multi-task generative framework for sequence-to-sequence RGBA image generation and editing. Its architecture features MSRoPE-BiL, a novel RoPE method with a bi-directionally extendable layer axis for its Diffusion Transformer (DiT) backbone, enabling the concurrent processing of multiple input and target RGBA layers. To power this framework, we introduce AlphaLayers, a new dataset of 1,000 high-quality, multi-layer triplets, built via a novel automated synthesis and filter pipeline. Jointly training OmniAlpha on this dataset across a comprehensive suite of 21 diverse tasks, extensive experiments demonstrate that our unified approach consistently outperforms strong, specialized baselines. Most notably, OmniAlpha achieves a dramatic 84.8% relative reduction in SAD for mask-free matting on AIM-500 and wins over 90% of human preferences in layer-conditioned completion. Our work proves that a unified, multi-task model can learn a superior shared representation for RGBA, paving the way for more powerful, layer-aware generative systems.

Problem

• Bridging the gap between specialized RGBA models and unified frameworks
• Enabling concurrent multi-task RGBA image generation and editing
• Creating a shared representation for diverse layer-aware generative tasks

Innovation

• Unified multi-task RGBA sequence-to-sequence generation framework
• MSRoPE-BiL RoPE method with bidirectional extendable layer axis
• AlphaLayers dataset with automated synthesis pipeline for training