PhotoGPT AI Reviews

I really wanted to like PhotoGPT, but after spending significant time testing, I can’t recommend it. The core problem is that the software simply does not follow prompts or respect user direction in any reliable way.

I created a trained model of myself with 15+ reference photos. Despite that, PhotoGPT consistently produces:

Incorrect body type – I am 6’7” and 340 lbs, yet the software keeps generating slim, generic versions of me no matter how many times I reinforce the prompt.

Wardrobe drift – Even when I explicitly state no suits, no ties, no jackets, it repeatedly puts me in exactly that.

Overpopulated scenes – I ask for 3–4 people, but it generates 7–10.

Clones – My likeness keeps being duplicated across other characters in group scenes, resulting in multiple “me’s” in one image.

Unrealistic group shots – Instead of natural-looking startup culture images, it defaults to stiff stock-photo corporate boardrooms.

I’ve tried tightening prompts, adding negative prompts, and following best practices, but the results are inconsistent at best and flat-out wrong at worst. For personal portraits it’s passable, but for any scenario that involves more than one person or specific instructions, the tool falls apart.

Bottom line: This is not ready for prime time if you want professional-grade, brand-consistent images. I wasted far too many hours trying to coax usable results and came away frustrated. Unless they make serious improvements in respecting prompts, handling diversity, and preventing duplicates, I can’t recommend it.

I waited a long time before leaving a review and I have to say this is one of the best deals because of the fact that the photos have realism.

The only main concern I have with photo GPT is sometimes it doesn't follow the prompt. So I hope that that improves in the future. Despite that I still give the photo quality five stars because it does look good.

I bought the max tier because after my first tries I discovered an incredible tool! I created my profile very quickly and now I can generate any image with the real me, no strange faces or people who just look similar. It is really me in every image generated! Sometimes the result is not perfect, but that is rare. I am truly impressed with the quality. If this offer returns on AppSumo, do not hesitate to try PhotoGPT, you will love it!

I purchased Tier 2 mainly to save time creating images for my social media. While the tool can produce nice results, I was disappointed by some limitations that make it less useful for my needs.

First, there’s no ratio option for platforms like LinkedIn cover images. I bought this thinking it would help with social media graphics, but without a wide banner option (e.g., 6:1 ratio), I’ve had to move my workflow back into Canva to resize and adjust everything.

Second, when I tried to get an image of just the ladies at work so I could create my own background in Canva, Photo GPT still generated a full background — it didn’t give me a clean subject image.

Lastly, I asked for my face on one of the people and different ladies for the others, but often all three ended up looking like me (or at least one of them did). This can make group images look a bit strange.

Overall, the product is OK for general image creation, but for something marketed toward social media use, the lack of flexible aspect ratios and the challenges with subject variation limit its practicality for my purposes.

First of all, I played a lot with local models, also trained my own model and generated many free images, which work "ok". But locally its more effort, more difficult and also image generation is slow (Geforce 4060). But thats normal, you need high end hardware to be faster. Also you don't have the same (but also very good) quality like some online models (but thats get better each week).

So I gave PhotoGPT a try. And overall summary:
It's not much better in case of quality, like anything you can do locally for free. Most of the time you can have the same results locally. But, and thats why I give 3 stars, its much more easy and faster. But to be a good product (there are tons of others in this niche), the quailty must be improved.

Some good things:
1) very easy to use, everybody could use it in seconds
2) nice UI
3) can generate really nice images with your face, some of them can look nearly exactly like you (but you need to generate a few for best results)
4) really nice presets, which can give you many ideas

Some bad things:
1) Many times, when the AI puts my head on a body, my head is much larger and it looks weird. I never had this when playing with open source models.
2) My age is 40, which I put into the model, but I don't have a ton of wrinkles yet. The AI actually added many wrinkles to many of the outputs, making me look much older (50+). Also if you force the output, to be close to original (which in most cases doesn't make much difference at all). I also never had this with other models.
3) There is no way to ask the AI to touch up the photos.
4) When it comes to the body, you should not expect it to be close to your real body. Sometimes the AI makes you really really skinny. It should not only use your face, also your body. At least a bit more than now.
5) When it comes to tattoos, it's totally over... it detects them, but give you some random ones, also on places, where you don't have. So better to avoid that, by telling the AI not to use tattoos, but this way, it't not you sometimes. But thats a problem of all current AI models.
6) On many presets you have the typical "eary" AI image look, like plastic or 3D Render. This has nothing to do with "photo".
7) Sometimes the generated person is totally different. It's not even kind of close to your trained model. For example, you generate 4 pictures with "dating" preset, 2 of them it's not your face, 1 other is ok, and the last one your head is much bigger. If I use my model, which was trained on 18 high quality and good images, I expect more. Its totally fine, that not all images are good and usable, but they should respect the given model.
8) The upscaler is really bad, it makes you images worse most of the time. It looks like its not a real upscaller, because your images often look different and not upscalled.

So 3 of 5 tacos. I would say it's 3.5
But quality needs to be improved.
I generated like 50 images so far, but only like 5-6 are good and usable (at least from my quality expectations). So keep that in mind.