How to make great AI visuals

More control and better results with scribbles +

prompts.

Kein Englisch? Besuch die deutsche Version...

Without a doubt, AI-generated images have now conquered various markets and are an indispensable tool for generating audiovisual content.

There are numerous examples.
Coca Cola released a commercial that no longer features actors. Some new shows from popular streaming services feature AI-generated intros. And especially in online media, AI ads have now become the standard.
In the latter, you can only recognize the artificially generated images by the fact that the "successful young entrepreneur" shown in the picture, who is supposed to have invented a "revolutionary solar panel," always somehow looks like Elon Musk.

Not only are AI-generated thumbnails increasingly replacing "hand-crafted" ones on YouTube - some of the video content itself is already being completely generated by computers. And even though the overall quality of the digitally conceived images and sounds still leaves much to be desired, positive comments ("Hollywood is cooked!") prevail, while critical voices have become rarer.

This is also because the AI industry still enjoys a great deal of trust when it´s CEOs are promising more wonders. Not only tech bros are convinced that AI development will continue to progress rapidly.
And even if there are still concerns of creators (whose content was used to train the models): let's not kid ourselves, AI tools are here to stay.

There are boundaries that text prompts can´t cross

The question many are asking now is: how can I use AI for myself? In case humanity is  NOT wiped out by a flood of artificially generated disinformation or the long-predicted AGI (Artificial General Intelligence)?

Every creative working in the advertising industry who has ever used AI to work on mood images, storyboards, or even animatics experienced that frustrating feeling at some point:

This already looks great - but why on earth am I not getting exactly what I imagined or prompted?

Aside from obvious errors/inconsistencies (like additional limbs) the AI doesn't always do what you have described to it in words.

Even experienced prompt specialists eventually reach a limit when their seemingly clear directives are misunderstood - or straight up completely ignored by the non-human models.

An algorithm is simply not a human model to which you can say, "Turn your head a bit more towards the camera and let the cape fly nicely behind you!".

The solution: scribbles and prompts

If you need a complex or non-generic visual, there are still ways to achieve a better result.

By adding an illustration, especially a scribble, AI can often be guided more effectively than with words alone.

Of course, it helps if you can draw well.

However, the AI delivers better results when the template is consistent in terms of image composition and details. Therefore, it may make sense to get a professional illustrator on board in your project.

Because by using a sketch, he/she can not only provide precise instructions to the AI, but also give fundamental recommendations for visualizing an idea/concept to the art director or the client.

Especially in the field of test films (narratives, animatics, digimatics, etc.), it can be extremely advantageous to utilize the expertise of an illustrator who specializes in visual storytelling.

Incidentally, such a professional illustrator is also a craftsman who can "repair" unsuitable parts of the visuals, for example when the AI being used has once again "imagined" additional limbs.

Because, just like the "three-body problem" known among physicists, the "six-finger problem" in generative image AI is still not completely solved.