# ACE++: Instruction-Based Image Creation and Editing via Context-Aware Content Filling

**Authors:** Chaojie Mao, Jingfeng Zhang, Yulin Pan, Zeyinzi Jiang, Zhen Han, Yu Liu, Jingren Zhou  
**arXiv:** [2501.02487](https://arxiv.org/abs/2501.02487)  
**Date:** 2025-01-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2501.02487v1/

## Key points

- ACE++ is a powerful instruction-based diffusion framework for image generation and editing tasks
- Two-stage training scheme maximizes the potential of image generative priors
- Offers comprehensive models for full and lightweight fine-tuning options
- Superior in generating high-quality images and accurately following prompts, as shown through qualitative analysis
- Capable of producing visually appealing and contextually relevant images across various creative tasks, such as foxes, cartoon characters on desktop computers, logos on products, and imaginative portrayals of ducks

## Summary

, , , , 

Introducing ACE++, a powerful instruction-based diffusion framework for image generation and editing tasks. Building upon the Long-context Condition Unit (LCU) from ACE, we draw inspiration from FLUX.1-Fill-dev's input format for inpainting and extend it to cover a wide range of tasks. Our two-stage training scheme maximizes the potential of image generative priors, utilizing pre-training with specialized models like FLUX.1-Fill-dev and fine-tuning for general instructions across all tasks defined in ACE. With comprehensive models for full and lightweight fine-tuning options, ACE++ offers broad applicability in diverse scenarios. Through qualitative analysis, we showcase its superiority in generating high-quality images and accurately following prompts. Visualizations featuring foxes, cartoon characters on desktop computers, logos on products, and imaginative portrayals of ducks highlight the capabilities of ACE++ in producing visually appealing and contextually relevant images across various creative tasks. Overall, our refined summary highlights how ACE++ leverages advanced diffusion techniques to excel in image generation and editing while offering flexibility for customization and adaptation to different use cases.