Ace by Alibaba: Chatbot-Based Image Editing

Ace is Alibaba's new, open-source tool that lets users edit images simply by chatting with a bot.

This tool, built on a diffusion model, allows users to adjust images with prompts rather than manual edits. It covers a wide range of tasks in one platform, from single actions to complex, multi-step edits, by tracking previous instructions.

Key Features

• Versatile Edits: Users can ask the bot to change colors, add text, adjust angles, and more.
• No Manual Steps: Ace doesn’t require manual steps like masking or inpainting.

Ace is designed to handle many editing requests within one model, like a GPT-4 for images. By processing instructions in a unified format, Ace can complete a variety of edits in response to user prompts.

Ace's Availability

Ace's code is available on GitHub (Apache 2.0), and the tool's demo can be tested on Hugging Face.

Ace's Limitations

The paper mentions that:

1. Image Quality: While Ace supports many edits, it doesn't match the image quality of tools focused only on text-to-image creation, like Midjourney.
2. Instruction Limits: Ace sometimes struggles with complex or multi-part instructions, as it can’t handle multiple edits in one go without breaking them down.