OmniAlpha: Aligning Transparency-Aware Generation via Multi-Task Unified Reinforcement Learning

TL;DR

OmniAlpha introduces a unified reinforcement learning framework that models transparency-aware image generation by jointly optimizing RGB and alpha channels across multiple tasks, improving quality and coherence.

Contribution

It presents a novel multi-task RL approach combining an alpha-aware VAE and Diffusion Transformer for high-quality, unified RGBA generation and manipulation.

Findings

Achieves 9.07% reduction in RGB L1 on layer decomposition.

Outperforms specialized models with 74%/68% improvements on SAD/Grad for matting.

Consistently outperforms baseline models across five transparency-aware tasks.

Abstract

Transparency-aware generation requires modeling not only RGB appearance but also alpha-based opacity and cross-layer composition, which are essential for tasks such as image matting, object removal, layer decomposition, and multi-layer content creation. However, existing RGBA-related methods remain largely fragmented, with separate pipelines designed for individual tasks. While a unified model is desirable, supervised fine-tuning alone is insufficient, as localized regression objectives cannot directly optimize the compositional fidelity, alpha-boundary precision, and structural consistency required for high-quality RGBA generation. To address this, we propose OmniAlpha, a unified multi-task reinforcement learning framework for transparency-aware generation and manipulation. OmniAlpha combines an end-to-end alpha-aware VAE and a sequence-to-sequence Diffusion Transformer, with a bi-directional layer axis in positional encoding to jointly model multiple RGBA inputs and outputs within a single forward pass. Built on a multi-task SFT cold start, it further performs GRPO-style post-training with layer-aware rewards defined on decoded RGBA outputs, enabling direct optimization of cross-layer coherence and fine transparency details. Experiments across five categories of transparency-aware tasks show that OmniAlpha consistently outperforms its unified SFT baseline and achieves strong performance against specialized expert models, including a 9.07% relative reduction in RGB L1 on layer decomposition and 74%/68% improvements over conventional matting tools on SAD/Grad for automatic matting.