As AI graphic technology advances, tools like Microsoft’s Bing Image Creator and NovelAI are enhancing the quality of generated images, elevating the reputation of AI graphic technology. Meta AI’s Imagine has emerged as a new contender in this arena, offering an alternative for individuals restricted from using Bing or lacking the hardware to run Stable Diffusion independently.
Operating akin to Microsoft’s Bing Image Creator, Meta AI’s Imagine allows users to input a word prompt in natural language, click “Generate,” and subsequently receive multiple images reflecting the AI’s response accurately after a processing period. Despite encountering difficulties with certain prompts such as “Chaos V-Ray,” potentially attributed to the term “Chaps,” the quality of the generated images remains notably high. Similar to Bing, Meta implements stringent prompt filtration, rejecting prompts deemed harmful or inappropriate.
Through our experimentation with various prompts, we observed diverse outcomes. This article presents four scenarios comparing Meta’s Imagine with Bing, DALL-E 3 Image Creator, and a local installation of Firm Diffusion XL 1.0 on a GeForce RTX 4080. Bing consistently outshines in delivering superior images across most cases, leveraging the impressive performance of DALL-E 3. Nonetheless, Meta’s Imagine generally excels over SDXL in image quality, a commendable feat considering it is a free service generating just four images.
Generating four images akin to the provided example on a GeForce RTX 4080 requires approximately one minute. While this processing duration is relatively brief, it lags behind Bing and notably trails Imagine in speed. Furthermore, Imagine produces higher-quality images compared to Bing or SDXL, with Imagine presenting images at 1280×1280 resolution in contrast to SDXL’s 1024×1024 resolution. (Although SDXL permits output size customization, its optimal performance is at 1024×1024, aligning with its trained image resolution.)
Nevertheless, nuances merit consideration. Imagine exhibits reluctance in crafting images devoid of figurative content, evident in an instance where it declined to generate an image lacking a person, unlike SDXL, which excelled in a 3D render scenario. Microsoft’s stringent AI constraints extend to avoiding contentious terms, concepts, and even celebrity names, necessitating altered prompts for Bing to navigate these limitations effectively.
While Bing’s Image Creator typically yields optimal outcomes, Imagine presents advantages in speed, image size, and, presently, marginally stricter content policies. Unlike Bing, Imagine does not mandate “boosts” for expedited generation post exhausting free usage. However, its dependence on Meta, Facebook, or Instagram profiles for login may pose a hurdle for certain users.
For enthusiasts keen on image creation with a robust GPU, establishing a local installation is advised. This approach furnishes complete control over model parameters devoid of external constraints, albeit running AIs on personal GPUs may exhibit slower performance compared to cloud-based services by Meta or Microsoft. Users possessing AMD GPUs are urged to explore this avenue, while individuals with Arc GPUs can reference specific guidelines for implementation.
