KRD Creative Studio has recently been given access to Adobe Firefly, the Artificial Intelligence offering from Adobe. This is currently in Beta mode, where we can experiment with how AI can integrate within the current workflow and range of Adobe software in professional use.
The first and most obvious thing to try was text-to-image. Writing text prompts to create imagery is not as simple as it may seem, and initial results are not always as expected. After a short trial, my initial impression is that this is most definitely not a replacement for a stock image library search. And certainly not a replacement for a planned photo shoot. There is an element of surprise which is interesting when experimenting but is not acceptable for professional work.
Prompt writing is, and will be, a skill set of its own, just like mastery of any computer coding. Until the software is developed enough to produce repeatable results from identical commands, it is unlikely that without extensive self-education within any given data set, to become a replacement for traditional media generation. In the interim, it will likely become a sub-genre of image creation, specialised in by some.
For testing, the prompt attempted to render a kitchen and modern living area. Interested to see if we could craft a rendering ourselves for our kitchen appliance clients. After many iterations, Firefly managed to get some that looked ‘nice’ overall, however the detail is lacking. The legs of chairs and whole chairs were disjointed, out of perspective. They pass ‘at a glance’ but any deeper viewing and they fail to capture the real detail of a whole scene.
These initial images could perhaps prove useful for ideation before a project begins, but at this point are unlikely to be suitable for detailed real-life finished art.
Once we can upload our own images into the software to be edited it may prove to be more useful to our professional workflow. Eg: being able to change the colour scheme in a real photograph with a click, instead of painstakingly masking, overlaying and manipulating.
As the software develops, the gap between ideation and real life will diminish, but whether it will diminish equally across all subject matter will determine its usefulness to professional workflows, and its level of disruption to the creative industries.
More useful are the tools to remove and change backgrounds automatically from images, which photoshop already has to a limited degree. But much the same, these tools will become much more useful in a combined workflow when using traditionally made images.
More likely to be disrupted quickly is language-based content. ChatGTP is easy to use, and can generate reasonable content at lightning speed. However, these pieces of generated content still need to be checked for accuracy. As with any database, the content you get out is only as good as the content you put in. ChatGTP appears to write with authority, so this is something to be closely monitored if used in a professional capacity. For an alternate way of wording what you already know, it is a fast effective re-writing tool. But don’t ask a question you don’t already know the answer to, and if you do, make sure to check it against reputable sources.