Meta’s New AI Tool Brings Precision to Generative Content
By Adam Pease
It’s been an interesting year for anybody interested in AI’s potential to generate media. With OpenAI’s DALL E 2 revealing near-photorealistic image generations, many feel we are on the verge of a breakthrough in the way content gets made. Meta recently previewed a new tool it calls Make-A-Scene, which promises to bring new precision to the creation of visual content with AI.
What Is Make-A-Scene?
Meta’s new tool is geared towards artists first and foremost. Since the rise of generative image content—powered by the strength of new diffusion models—many artists have begun to incorporate AI into their creative workflows. Meta has been beta-testing Make-A-Scene with a group of generative artists who are testing the tool in their own artistic practice.
Make-A-Scene operates similarly to other text-to-image generative tools like DALL E—with one important exception. Rather than taking a single input prompt and generating a response render, Make-A-Scene also allows the artist to paint a preliminary scene layout for the image, which guides the image generation process, matching certain shapes to certain painted patterns and giving the artist much more precision about the size, form, and placement of generated elements.
Why Precision Is So Important for Generative Content
While Make-A-Scene is initially pitched towards artists, it has far broader applications in the content creation space more generally. Precision has been sorely lacking in image generation, putting users at the mercy of what the algorithm chooses to generate. Make-A-Scene’s ability to guide the generation process with more intention could dramatically improve the speed and quality of image generation workflows.
By beginning with artists, Meta has tapped into an important pool of knowledge about AI workflow that will hopefully guide it towards releasing a tool that delivers much-needed precision to image generation. Meta has opened up the tool for testing beyond its initial round of artists to select employees of the Meta corporation. Next, it says that it expects to create a publicly available preview to allow more content creators to experiment with the technology.
Meta’s Make-A-Scene tackles a major problem in generative content head-on. If Meta can manage to deliver an effective tool for guiding image generation at scale, it could remove one of the most serious obstacles to the growth of generative content.