085: Using GenAI in the heat of a branding pitch

They're not a cure-all for strategic insight, or creativity in general—but they are amazing

Apologies for the delayed post. I spent most of this week heads down developing a branding pitch. Also, I’m 85% done with the AI for Artists and Entrepreneurs curriculum; planning to share that next week with you all. And our house is encased in plastic sheeting because of the ongoing construction. Seriously, I could startup a business that comes into your house after the framers, electricians, sheet rockers or whomever wraps for the day and I would add so much value just by cleaning up…a little! And I would pay for that service, too. Hello, general contractors? Anyway, here’s this week’s post arriving on a Saturday. C’est la vie.

Getting Pragmatic with AI in a Branding Pitch

What’s possible now in December 2023 is truly remarkable.

Two people, enabled with and practiced in ChatGPT/Dall-e3, Claude, Bard/Gemini, Seenapse, Midjourney, Firefly, Emu, and Photoshop’s generative fill can run circles around agency teams doing similar work 20 years ago. But you need lots of hours in each of the tools to understand their permutations, limitations and peculiarities. You also need to understand how strategy functions, how brand concepts are developed, how ideas come to life—which is easier to acquire through a career than querying an LLM. The tools are a marvel on the production and content tonnage side of the equation; the metaphoric and lifestyle image elements, or “give me 20 different ways to rephrase this statement” tasks.

Here’s an example: Let’s say we want to present taglines along with a single evocative image, something that adds to the feeling, the mood of the words and helps further convey creative and directional intent. 20 years ago we’d be combing through magazines, possibly querying the Internet (remember, it’s 2003), maybe hiring an illustrator… but wait! Even before we did that research, or that commissioning, we had to develop a thesis, a general sense of what the image we seek will convey. What are we looking for? This specific task, the “figure out an image” job, is akin to a commercial film director’s treatment. You’ve got a script, but what might it all look like?

You could absolutely use text GenAI here. “Given creative direction and tagline X, give me 20 visual metaphors that build on a the core idea.” And I have certainly done that—it is the back-and-forth of a creative team. Sometimes you get lucky with generative outputs. For me, the point isn’t necessarily to seek a clear result and just be done with the exercise; though that would get me to bed earlier. Instead, I view the generative tools as a sparring partner, a tireless collaborator who will spew (that’s the technically correct term) potential. It beats me working all by my lonesome.

Now let’s presume we have our direction honed. We have our prompt. Here are broad observations, having spent over 40+ hours trying to solve visual challenges in Midjourney, Firefly, Dall-e3, and to a lesser extent, Meta’s Emu.

Midjourney—Still the leader for photorealistic output; but still requires the most prompting skill, the most insider/expert understanding of how the tool itself does what it does. If the metaphor you seek is sci-fi(ish) or literal, Midjourney can work well generating concepts. There’s lots to love about the granular controls you can work in prompting, but the Discord interface is tiring. All that said, I’ve learned a ton just scrolling the general feeds and copy/pasting prompt language into my prompt doc. (Yes, I recommend you have single text doc where you keep all your prompt language.)

Firefly—Quickly becoming my preferred go-to. I really appreciate all the effort Adobe’s put into its web-based Text-to-Image generator. The ability to use one image as the basis for three more, and keep that original on screen is appreciated. The camera, lighting and conceptual controls are fantastic. The tooling is robust, and requires you spend the hours using it to gain full advantage. If I owned another agency I’d require my designers and ADs to spend time learning these interfaces. But what about broad conceptual output? I’d rank Firefly third, behind Midjourney and Dall-e3. It’s very subjective, but the results I’m getting in terms of pure “concept” within Firefly tend to be much less successful. Typically too literal, or totally missing the point. But it’s early days, and I’m certain that aspect will improve. If your concept is meant to be photographic, and realistic, then Firefly can suggest decent conceptual results.

Dall-e3 (accessed via ChatGPT)—Surprisingly great, but not for photo realism. As a general conceptual tool, I preferred D3 results over MJ and FF most of the time. Lord knows how the models are coded, but D3 seems to parse conceptual and metaphoric requests “better,” at least for the kinds of outcomes I was seeking for this project. And the fact I could easily respond to a result, i.e. “Add more space to the left” makes using this tool worthwhile. Photorealism is not D3’s strong suit. It prefers to work in an illustrative yet realistic space, at least at this point. And you only get one image at a time in D3, and if you ask it to regenerate, you lose the original. So from a process standpoint, I suggest always downloading a result before adjusting.

Emu—Meta’s tool offers conceptual results similar to D3, it seems to prefer illustrative versus photorealistic outcomes. It only outputs one image, and only in 1:1 aspect ratio. All the other tools enable 16:9 or alternative ratios. In my process this week Emu was, “Well have we tried Door #4 yet?” We’ll see how their coding evolves.

And here’s the thing: Every generative tool requires lots and lots and lots and lots of trial and error—which is entirely the point

I’ll keep saying it—GenAI tools are not search engines. They are not binary results providers, they are conversational tools. The point isn’t a first page result, the point is the iteration and dialogue and the back and forth. The mistake is in confusing a screen based software platform with all previous platforms. Generative tools might look like Search but they don’t act like Search.

In this week’s effort we ended up shipping eight distinct directions. So, eight attempts at querying and generating direction-specific images. Some were easier than others and we got lucky early in the process. Others took hours if not days of iteration and prompt refinement. But is this any different from 20 years ago when we’d gather stacks of potential images, or phone calls with illustrators to refine a pencil sketch? Ideas take time. The advantage today are the tools’ immediacy, flexibility, speed and democracy. What I could accomplish alone is remarkable. But it really helps if you’ve put in the hours to know how GenAI tools can work.

And thank goodness for Photoshop’s Generative Fill.

None of the generative images we used came out perfect the first time. Nearly all of them needed retouching, or layering and blending of multiple images to create a scene. I spent hours in Photoshop. And a lot of that time was spent navigating Generative Fill to remove or replace elements. This is another way of saying, GenAI tools are not a panacea, and while they will save time in some areas they create additional work, too.

But you still need to be able to think, and create, without the tools

I recall talking with an architect changed with building out what would become a new 800 person ad agency office. He asked, “As a creative director, what do you need from the office space?” And honestly, I disappointed him. “Well,” I replied, “all we really need are pens and paper to capture ideas and walls to put them on so we can evaluate and refine.” I suspect he had been hoping I would require a fireman’s pole or similar stereotype of a creative environment. As if the space itself was key to the work. As proven time and again, the space can enable but it ain’t the work.

GenAI tools are the same. They can enable, but they ain’t the work itself. It really helps if you understand how strategy becomes and informs ideas, if you’ve developed ad or brand or design campaigns outside of the tools. If you’ve spent time with humans working through similar circumstances. Then you see the role and the place in which GenAI creates real value.