Whisk Produces Very Good AI Character Art...and Free

My last post talked about Microsoft's Copilot and how it suddenly gave me the cold shoulder. To summarize, I've been looking for a way to add content to my author site and one element that's been touted is character art. I've seen some excellent AI-generated images online, but I don't have the money or really the need to purchase a subscription to one of those apps. I knew that Copilot was available to me for free, so I assumed it was my only option.

Well, as my last post details, things didn't quite work out. And honestly, I wasn't thrilled with having Flintstones-like depictions of my characters. It was my own fault for not requesting "realistic" images, but even had I known that Copilot possesses that functionality, it still would have frozen up, maybe even sooner than it did.

Out of curiosity, I did a search for the "best" AI applications, but just as I thought, they're all subscription based. But wait ~ what's this? Google has a free program? (Is there anything Google doesn't have?) That led me to Whisk. As with anything free, I didn't hesitate to give it a whirl. 

I started with Leah, naturally. I described her as fully as needed, which Copilot was great training for. AI, at the present time, believe it or not, can't read one's mind. What I neglected to do was insert the phrase, "realistic image", and thus it returned an animation, albeit a better one than Copilot was able to produce. I knew from Whisk's homepage that it was capable of creating lifelike representations, so I tried again. This time I got a "real" person. It was pretty good; not perfect, but I was able to refine it.

I get that no AI program creates actual people out of thin air. It scours its database and pulls from real photos, then tries to modify them to the user's specifications. Sometimes it works well; other times it fails miserably...and in the weirdest of circumstances. Elements that should be easily reproduced are sometimes impossible. At times, Whisk gave up and told me it just couldn't do it.

But as I went along, and I went along for a long time (more about that later), I learned more about Whisk's quirks and capabilities, and I adjusted my inputs accordingly. For images it just can't do, it allows the user to upload a reference photo. For the character of Paula in particular, that was my last resort. Whisk seems to think that a sixty-year-old woman is a rocking chair granny, with all the accompanying wrinkles and sags. So I found a closeup of Reba McEntire and uploaded it. (Spoiler alert: I loosely based Paula's character on her, at least in appearance.) Whisk went with the image prompt and subsequently did much better.

(Yes, I did specify the stage direction.)

My biggest frustration was Burt. I still don't have an accurate Burt, at least per my mind's eye. The first few generated images depicted an old grumpy guy, even though I specifically told Whisk that he "has a twinkle in his eye". And when I said he's a flamboyant dresser, Whisk went way overboard, giving him a multi-color garish neon sports coat. "Flamboyant" was obviously the wrong input term; maybe "colorful" ~ no, scratch that. That probably would have been worse. (I'm imagining a jacket with assorted color blocks, like a quilt.) Plus, Whisk insisted on giving him a mustache, which he most certainly does NOT have. Again, I had to find a photo of a mostly bald man without a mustache and upload that. Repeating, "No mustache!" wasn't getting the job done.

Strangely, my most frequent refinement requests involved backgrounds. AND HERE IS A VERY IMPORTANT NOTE: If Whisk returns an image you like everything about except for the background, it won't simply replace the background. It'll generate a new image with the background change you requested. That's my biggest beef with the app. I'll get an image I really like except for one element, and I either need to accept it as is or gamble on a new image that's inferior.

It's also easily confused, particularly if you are asking for a large quantity of images. Unless you specifically indicate "NEW IMAGE", it will take the last image produced and add the new description to it. That's how I ended up with a tour bus with a picture on the side of it of a man sitting in a recording studio wearing headphones. Or a woman signing a contract next to an oversized "Welcome to Chance" road sign. And Burt kept popping up in the strangest settings.  

As picky as I am about my characters, I asked for a ton of either refinements or outright do-overs. This depleted my ration of free daily images. The app doesn't specify the number of images one can receive daily, and Google search doesn't even seem to know:

"Determining the exact daily image generation limit for the free version of Whisk AI is a bit unclear based on the available information."   

Suffice it to say that I got a lot. I had so much fun with the app, I kept going for hours until Whisk threw up the stop sign. Admittedly, I may have downloaded only about fifteen or so images (which I intend to use to create a photo collage...or "mood board" or whatever the hell people want to call it), but I'll be back at it again today.

                                                              Leah at her lowest point

Users will be able to subscribe to Whisk's paid option once it concludes its experimental phase, and will thus have the ability to generate a much larger number of images, but it's not as if I'll ever use the app once this project is done, so the free choice is quite fine with me. 

Call me cuckoo, but Whisk is really addictive. I kept conjuring up new scene scenarios and didn't want to stop...until I was forced to.

If, like me, you're a penniless author and are looking for content to add either to your website or to social media, I highly recommend Whisk. And you know me; I'm pretty stingy with my recommendations.