When compared to state-of-the-art text-to-image models like Google’s Imagen, OpenAI’s DALL-E 2, or Midjourney’s namesake program, TikTok’s method produces quite rudimentary results. It only generates abstract and whirling visuals, which reflects the dreamlike character of TikTok’s recommended suggestions such as “astronaut in the water” and “flower galaxy.” In contrast, some models can generate both photorealistic pictures and intricate and cohesive drawings that seem to have been created or painted by people.
However, the restrictions of TikTok’s paradigm may be deliberate. For starters, more complex models need more computer power, which would be costly and resource-intensive for the organization to adopt. Second, TikTok has over a billion members and giving all of these people the ability to generate lifelike representations of whatever they can conceive will almost surely result in some terrible outcomes.
