跳到正文
CNET· Katelyn Chedraoui·· 10 天前AI 評分60

Adobe 三項 AI 研究項目聚焦精準圖像與影片編輯

Adobe’s Cutting-Edge AI Research Projects Are All About Precision Editing

AI 導讀

Adobe 在 Adobe Max 的 Sneaks 專場展示 Deep Brush、Hands-On 和 Scene Sweep 三項 AI 研究項目,分別用於場景感知繪畫、透過動態介面調整物件,以及編輯影片中的角色。

正文

Adobe has been investing in generative AI tech for creative work for years, and now researchers have started using AI models to enable small, precise edits — the very opposite of an AI hallucination, if you will.

These are three of the projects that will be presented at Adobe Max during its Sneaks session on Wednesday. You can follow along with CNET’s live updates, but this is your exclusive sneak peek.

Project Deep Brush for context-aware painting

Project Deep Brush, developed by senior research scientist Iliyan Georgiev, introduces a context-aware digital brush that paints seamlessly across uneven surfaces such as walls, fabric and skin. Instead of applying text, paint or objects as flat, separate layers, the tool automatically bends and conforms them to the scene’s natural geometry and physics. Automating this process eliminates a task that would otherwise require significant time and advanced manual skill.

More from CNET

The tool automatically matches the lighting of an image, so the shadows make logical sense. You can have the painting curve behind existing objects, like a tattoo that disappears under a shirt collar. You can select between different modes, including flat, tilt and warp, to best align with the physical geometry of objects in a project.

Adobe

Project Hands-On for wordless AI prompting

While AI enthusiasts might enjoy writing the perfect AI prompt, Project Hands-On lets creators literally get their hands on the objects in a scene with a new dynamic user interface.

Adobe

Celso Gomes, principal research designer, told me that it’s a short two-step process. The program runs an image analysis on the project, identifying objects and their characteristics in the scene. The AI pulls up a short list of widgets for elements that you can adjust — an analysis of a perfume bottle would include the bottle shape, the material it’s made of, and the color and the photo style, for example. Then, using the dynamic UI placed over objects, you can resize and edit different components of each object until you get it just right.

You can apply multiple edits in one go before running the changes, a nice time saver for professionals who don’t want to wait in between each generation. Behind the program is a hodgepodge of AI models, combining UI and AI image knowledge into a new editing tool.

In these progression photos (left-right) you can see how the intelligent UI understands the instructions to write “MAX” with the juice.Celso Gomes/Adobe

Project Scene Sweep for precise multishot video edits

Project Scene Sweep is one of the most impressive AI video tools I’ve encountered. With this tool, you can upload your raw film, select a character, write a prompt to adjust something (change their hairstyle or jacket, for example), and the AI will identify the character in every shot and apply the change consistently across the entire timeline. And you can make multiple edits.

Hyeonho Jeong, a research engineer presenting the project, told me the video editor isn’t meant to change the narrative of a piece. Instead, it’s meant to help editors make use of the content they have and avoid costly reshoots.

A screenshot of Project Scene Swap for multi-tasking video editing
You can see how the video editor pulls out specific characters and locations that can be edited via prompt.Adobe

Big picture: The future of Adobe research

The throughline across all three of these projects is using AI for precise edits. Generative AI, as a whole, isn’t always the most precise tool — that’s how we end up with things like hallucinations and seven-fingered people in AI images. But new advancements in the tech over the past year, evident in these projects, have helped bring that level of control and precision to creators.

The goal is to create tech that understands the user’s intent, said Hailin Jin, senior principal scientist who oversees image and design research. “This is not just only about getting the best or most interesting adjustments to a particular image,” Jin said. “It’s about streamlining the entire process so that users don’t have to be working on repetitive things… so users can focus on the creative part.”

There’s no guarantee any of the Sneaks projects will be released to the public, though two of these, Deep Brush and Hands-On, are getting an initial public release this week on the relaunched Adobe Labs website. The Adobe Labs playground space will grant early access to select research previews going forward.

According to Gautham Mysore, senior principal scientist overseeing video and audio research, the goal is finding new ways to blend science and art.

“Building out new technical methodology, the science and the art together,” said Mysore. “That’s the path we’re gonna go.”

Katelyn Chedraoui

Katelyn Chedraoui

Reporter 2

Katelyn is a reporter with CNET covering artificial intelligence, including chatbots, image and video generators. Her work explores how new AI technology is infiltrating our lives, shaping the content we consume on social media and affecting the people behind the screens. She graduated from the University of North Carolina at Chapel Hill with a degree in media and journalism. You can reach her at kchedraoui@cnet.com. See full bio

來源:CNET · cnet.com