Comment by canvascritic | Hacker News Reader

Comment by canvascritic | Hacker News Reader

My partner and I run a handful of small internet side businesses. One of our content-driven D2C businesses heavily relied on bespoke illustrations for our display ad creatives. we found that our ctr was decent, pretty average, but the CPC was killing us and ROAS really sucked.

Several months ago we decided to A/B test SD against our usual illustrators. In our case the results were pretty dramatic, we actually found that the ctr shot up by almost 20% and cvr showed a consistent uptick. I don't agree with the blog post's claim that AI generated images work best in businesses where the content doesn't actually matter; this particular venture is a fantastic counter example. In our case the AI-generated images seemed to resonate more with our target audience, as we were able to achieve much more granular personalization at lower cost than before. not only did it reduce the CPA significantly, but the tight control we had over creative variations meant we could optimize in realtime based on audience segmentation.

Not to mention that our time-to-market for launching new campaigns went down by half. no more back-and-forths over design nuances, missed deadlines, or creative blocks.

And I do feel a bit mixed about the diminishing role of human touch in creative processes. But from a purely growth-hacking POV, this was a gamechanger, and we have the numbers to prove it.

Overall I think this is a net win, especially because I don't think this needs to be the end of the road for human illustrators, but this will force them to adapt and bring more sensitivity to the needs of their clients. It makes no sense for even a content business to be subject to so much friction in the procurement of creatives, and this forces more consideration to our needs

Anywho there's efficiency, and then there's soul. Hats off to the robots for (mostly) nailing the former, and sometimes surprising with the latter.

no more back-and-forths over design nuances, missed deadlines, or creative blocks.

This evoked, for me, the "can I get the icon on cornflower blue" scene in Fight Club.

How much of this reduction in back-and-forth is influenced by the immediate/interactive response (dealing with fewer humans) and how much is due to a level of trust-of/delegation-to the machine? "A machine generated this icon based on my description, there's no need for me to question its choice of colors." — really the classic problem of considering machines as infallible and more expert than humans.

It's probably some of both.

I think your usage of "matter" and theirs is different. It's furniture. Furniture "matters" in a restaurant and having the wrong furniture can hurt your business— but compared to the food, it's essentially inconsequential.

There's a spectrum of how much furniture matters in any given place ranging from very short stay waiting areas to architect's offices, and commercial art is no different. If that image was truly inconsequential, you wouldn't need one there. Non-informational graphics on most non-professionally designed power point decks likely matter less. I'd say there's about a zero percent chance of a two page spread opening a feature article in a magazine being ai-generated unless it's an article about ai-generated images, and even then, it probably took professionals longer to massage it into shape than all of the rest of them. Specificity and per-pixel control is just so important in professional graphics workflows and despite what a huge stack of people who aren't professional designers will tell you, they are simply the wrong tool for the job. It's fundamentally the wrong interface. Maybe what Adobe or another player who knows what the industry needs will nail it, but it won't look like Midjourney— that's for sure.

What is your type of business and what kinds of images did you generate? Curious as I was thinking of doing something similar for mine.

> Overall I think this is a net win, especially because I don't think this needs to be the end of the road for human illustrators, but this will force them to adapt and bring more sensitivity to the needs of their clients.

The advantages of AI that you crow over simply can’t be met by any human professional artist. A human can’t do hundreds of revisions profitably. There’s increased “sensitivity” and then there’s needing to read the client’s mind.

If you think this isn’t a death knell for human illustrators in this particular market, you’re deluding yourself.

A professional artist that is proficient in the latest generative image models can increase their ability to attend to client needs.

The client "needs" in question here are low cost overall, low marginal cost for each revision, and a totally-interactive "do what I mean" interface.

Shoving a human artist in the middle is a liability on each front.

Your definition of the needs does not have any requirements on the fitness of the output, nor on time spent on customer side. That does not seem realistic.

They’re not mine; I pulled them directly from the parent comment.

Reply on news.ycombinator.com