原文 · 未翻译
Tech
AI
Report
Adobe’s conversational AI agent is a mediocre design intern
The Firefly AI Assistant isn’t as good as a professional human designer or photo editor, but it’s fun to watch it work.
AI image tools rarely make me feel like I’m part of the creative process. They are, after all, mostly designed so that people with no design experience can type in a few words and get back a usable result. So I was pleasantly surprised by Adobe’s latest take on an AI image assistant: It’s a bot designed to take away some busywork, while still granting you creative control.
Unlike AI generators that are specifically designed to make and edit images or video, Adobe’s Firefly AI Assistant, which I’ve been testing in beta, is more like a multitasking middleman that can operate Adobe’s design apps for you. On its website, Adobe says that you can just “tell Firefly AI Assistant (beta) what you need, and it will use tools from apps like Photoshop, Illustrator, and more to complete multistep projects in moments.”
The user interface looks like a typical chatbot. There’s a text box you can type prompts into, and a plus symbol for uploading media files. It doesn’t use the actual Adobe apps on your computer, but it has access to common capabilities like masking, object detection, and image generation. The AI assistant is designed to be conversational, so you can ask the chatbot to “make this photo more colorful,” and it’ll do so while explaining its actions.
Photo edits and illustrations completed by Adobe’s AI are convincing at a glance. It changed my hair color in one photo and then the background location and lighting in another. The results aren’t perfect: some had colors that were too vivid or alterations that hadn’t been properly blended into their surrounding environments. But I suspect the average person wouldn’t assume that my results were made or manipulated with AI — it just looks like the work of a novice designer.
What actually makes the Firefly AI Assistant intriguing is how it interacts with you. I gave it a picture of my cat by a window and asked it to make the sky cloudless and sunny. It didn’t just go “sure” and give me the edited image. The chatbot described the scene in the pre-edited image with surprising detail (it correctly identified that my cat is a Maine Coon despite the photo mostly just showing his ass), and then explained how it’s going to achieve the results I’d requested. It mentioned specific tools from Photoshop and Lightroom using established editing terminology, explaining the process step by step. You don’t actually get to see the image being edited in real time, but the chatbot will tell you which features it’s using to achieve each result.
The Firefly AI Assistant is also surprisingly forthcoming about its limitations. When I asked it to separate objects from a JPG file into separate layers, Firefly said it couldn’t do so, but offered two different courses of action for splitting the image into separate elements, explaining the pros and cons of both. After I chose one, the bot then described its editing process, including the fact that what it was doing wasn’t working. “I notice the gaussian blur approach isn’t giving me true transparent cutouts — it’s outputting full-image PNGs,” it wrote. The chatbot redirected itself and used masks and Adobe’s image cropping and resizing tool instead.