The Year in AI Papers
The year/Independent research

Paper 2605.25378

CollectionLoRA: Collecting 50 Effects in 1 LoRA via Multi-Teacher On-Policy Distillation

Fangtai Wu, Hailong Guo, Shijie Huang, Jiayi Song, Yubo Huang, Mushui Liu, Zhao Wang, Yunlong Yu, Jiaming Liu, Ruihua Huang

Published
May 2026
Research lab
Independent
Citations
2
GitHub
29 stars

01 In brief

Summary

CollectionLoRA is a multi-teacher on-policy distillation framework that consolidates up to 50 diverse visual effects and few-step generation capabilities into a single LoRA, addressing storage overhead, routing latency, and parameter conflicts in conventional multi-LoRA pipelines.

The method introduces three key components: Probabilistic Dual-Stream Routing (PDSR) for regularization and generalization preservation, Asymmetric Orthogonal Prompting (AOP) for concept isolation in the prompt space, and a Coarse-to-Fine Distillation Objective (C2F-DO) that combines trajectory anchoring with distribution matching to stabilize optimization and restore high-frequency details.

Experiments on EffectBench show that CollectionLoRA achieves superior concept fidelity (CLIP: 0.727, DreamSim: 0.425, EditReward: 1.052) and lower bad case rate (0.087) compared to single-task teachers and naive multi-task baselines, while reducing deployment overhead to 0.5% of conventional paradigms at 180 effects.

The framework also demonstrates zero-shot effect composition and incremental effect extension without retraining.

02 From the paper

Abstract

Customized image editing aims to equip pre-trained diffusion models with specific visual effects using limited paired data, typically via Low-Rank Adaptation (LoRA). As the number of desired effects grows, storing and dynamically loading numerous these effect LoRAs significantly increases deployment overhead. Furthermore, current pipelines typically cascade these effect LoRAs with acceleration modules for fast generation, which triggers severe parameter interference and results in concept bleeding and style degradation. We propose CollectionLoRA, a multi-teacher on-policy distillation framework capable of distilling the concepts of up to 50 different effect LoRAs along with few-step generation capabilities into a single LoRA. This fundamentally resolves the feature interference issue and significantly reduces deployment costs. Specifically, the method introduces (i) a Probabilistic Dual-Stream Routing mechanism that enables the model to randomly switch between data sources during training, effectively enhancing its generalization in unseen scenarios; (ii) an Asymmetric Orthogonal Prompting strategy to achieve concept isolation within the prompt space; (iii) a Coarse-to-Fine Distillation Objective to mitigate the distribution gap between the teacher and student models. Extensive evaluations show that CollectionLoRA distills all customized effects and few-step generation into a single LoRA, reducing deployment overhead while achieving concept fidelity comparable to or better than independently trained teacher models. Code: https://github.com/Qwen-Applications/CollectionLoRA