Generative AI for Images and Video

Generative AI for images and video is one of the most visible areas where AI is being used today. Tools can produce pictures, edit photographs, and generate short video clips. To see why this area has grown so quickly, it helps to start with a small case that shows how fast attitudes can change. The case involves AI-generated profile pictures.

Creative work is one of the areas where AI has found some of its most visible success. The reason given is AI's probabilistic nature. A fixed-rule tool produces the same kind of result for the same kind of input by following a set procedure. A probabilistic generator does not follow one fixed rule for every output. Its results can differ from one attempt to the next.

This openness matters for creative tasks because many tasks have several acceptable answers rather than one correct answer. A headline, a color palette, or a scene can each have many good versions. A generator whose outputs vary is well matched to a problem where variety is useful, which is why creative tools are a natural fit for it.

Reading the comparison as a column pair makes the contrast clear. The left column describes a tool whose output is set by a fixed rule. The right column describes a generator whose output varies, and whose variety is the property that suits creative work.

The material points to several startups as evidence of how this area is developing. The tools fall into three groups by what they do. Each group is listed below with its named tools.

The material also reports market signals. Midjourney, at one and a half years old in late 2023, had already generated 200 million dollars in annual recurring revenue. As of December 2023, half of the top 10 free Graphics and Design apps on the Apple App Store had AI in their names.

The main lesson of the profile picture case is the change in public attitude, not any single image. In 2019, a platform treated AI-generated profile pictures as a problem and banned the practice. By 2023, platforms had started offering the same kind of image as a feature, and it had become a built-in feature in many social media apps. The image did not change. The attitude did.

AI-generated profile pictures are now common on social media, from LinkedIn to TikTok. One motivation the material explains is that many job candidates believe AI-generated headshots can help them present themselves well and improve their chances of landing a job. Read this motivation alongside the timeline above: the same kind of image moved from restricted to widely accepted.

Advertising and marketing have been quick to adopt AI. The material describes three things AI can do for an enterprise. It can generate promotional images and videos directly. It can help brainstorm ideas. It can also produce first drafts that human experts then refine. The key idea is that AI typically takes a starting role, and people shape the result.

The workflow has a fixed order, and each stage depends on the one before it. AI can create multiple ads and seasonal or location variations. Audience testing then decides which version works best. The chosen version is picked from the test results, not from the first draft. The flowchart below follows one ad from first draft to chosen version.

In the workflow, human experts keep a central role at several points. They refine the AI first draft and iterate on it. The material describes AI as producing first drafts that human experts then refine, and it describes human experts iterating upon what AI produces. The people involved decide what the ad should become.