šŸ“”

(09.2024) - Review PicStudio.ai

In my last post I talked about how I had watched the Masterclass course about AI and that they had mentioned a few different AI tools during the course.

I thought Iā€™d try one out called PicStudio.ai

If you want to try it out and support me at the same time, hereā€™s my referral link: https://www.picstudio.ai/?via=dino

So the way it works is that you

  1. Select a ā€œpackā€
  2. Additional Options (depends on pack)
  3. Create a model

Select A Pack

The Pack is essentially the style of the photo. These are preset stylings like the lighting and concept for the background.

image

Additional Options

Depending on the pack you choose, there are sometimes additional options to select.

In my case, I selected the ā€œField Biologistā€ Pack. Once I selected it, I then needed to select from a list of Clothing options and a list of location options.

For the clothing, I picked ā€œLong-Sleeve Moisture-Wicking Shirt With Roll-Up Sleevesā€

For the location, I picked ā€œWooded Area With Deciduous Trees And Flowing Streamā€

šŸŖ²Bug Interlude:

I did this on my iPhone, but ran into what I can only describe is a bug. I assume theyā€™ve optimized their website for Chrome on mobile because Safari didnā€™t seem to work.

After selecting your pack and options, you need to add it to cart, and this simply wasnā€™t working for my.

I switched to using Chrome and then was able to complete the process there.

image
image
image

Create your model

Once you have selected the styling of your photo, (And paid for it), you need to provide information about the model (person) upon which the images are based.

This includes answering a bunch of question like age, gender, body type and so on. These were my details.

Model details

  • Ethnicity: White
  • Eye color: Brown
  • Body type: Athletic
  • Age: 39
  • Gender: Male

You then need to upload ā€œat least 8 picturesā€.

šŸŖ²Bug interlude:

Here again there was some strangeness in the process. The copy and tips all say that you need to upload ā€œat leastā€ 8 pictures of yourself. And there is even a tip that says ā€œfor better results, upload more than 8 picturesā€. But when it came to actually uploading the pictures, as soon as I uploaded the 8th photo I got a ā€œphoto limit reachedā€ alert and was not able to upload any additional pictures. šŸ¤·

anyway, these were the pictures I uploaded:

image

I can also see on my ā€œmodelā€™sā€ details, that these images were automatically reinterpreted and cropped to the following:

image

Essentially, it looks like itā€™s mainly focussing on my face, and cropping out the rest. In some cases it seems to have kept some of my upper body, while in other pics it cropped that part of my body out.

Results:

So once youā€™ve filled out the question and provided your 8 pictures, and confirmed, it will take roughly 30 minutes for your pictures to be generated.

Each pack generates 40 images based on the same model.

Pros:

Overall I was blown away with how good the results were, it looks incredibly real and even generates images with my looking in various directions which is very cool.

My hair was very accurate, including how thick it is and the curls, the individual follicles were crazy. Most incredible was the beard hair. I have a mostly dark hair, but a few grey/white hairs in my beard, and thatā€™s exactly what came out, Iā€™m still in shock.

Another thing that I found really good was my nose. Iā€™ve tried generating ā€œcleaned upā€ photos of myself with ChatGPT in the past and they were complete failures. It always spits out an Ć¼ber smooth fake looking person, my nose is always turned into this weird long , thin nose, and my hair is always something that feels almost random.

Wrinkles and smile creases and even how my eyes close when Iā€™m smiling were all 100% believable.

Cons:

On the face, I have very little to comment on, it was really good.

After really interrogating the photos, there were two things on the face that I noticed being ever so slightly off.

  • Teeth: My two front teeth are slightly too big and flat, while the spaces between my teeth are too pronounced, leaving a lot of dark shadowing between them than in reality. Although I will say it got my one canine on the right side of my mouth really on point.
  • Eye Wrinkles: So when I smile, I have a lot of wrinkles on the side of my eyes. It got the gist of it right, but because the pictures are generated to look like they are super hi-res, it also over-defines those wrinkles in many cases, especially the ones below my eyes. but like I said, I really had to interrogate the images to notice that this looks slightly innaccurate

So what about the rest of the body?

  • Arm muscles: I found this to be over-pronounced in almost all the images. As much as it would be nice to look like this ā€œnaturallyā€, I think I might be able to pull this off if I was flexing throughout the whole photoshoot, after having just done 50 bicep curls, and was wearing a shirt one size too small. The pics arenā€™t completely crazily out of the question, but they are just beyond the limit of believability in about 80% of the cases.
  • Body length: I found that many of the images make my torso look way too long. I think itā€™s interesting that none of the questions ask about height and all the pictures I uploaded were all automatically cropped below the chest area. Again, the outputs were not totally off, but there were definitely a few that made my look taller than I actually am. For reference, Iā€™m about 178cm // 5f9 (5ā€™10 if Iā€™m standing up really straight and craning my neck a bit šŸ˜›
  • Hands: I know AI hands are really bad generally, so in comparison, these are excellent, theyā€™re just not my hands though. I canā€™t explain it, I just look at them and know Iā€™m not looking at me.

Here are some of the outputs that I found to be both good and as close to accurate as possible.

image
image
image
image
image