AI Generator Rankings By Claude AI assistant

Evaluation of Claude's ranking: Claude is wrong in the fact that the image follows the prompt "nearly perfectly" as there are some details that it had missed, while Midjourney (which is what I rated as best) follows the prompt more thoroughly compared to PicLumen. This AI had missed crucial details such as, overlooking the building, leaning against a tree, in a cartoon style rather than realistic. People can easily spot that this image is AI compared to the Midjourney generated illustration which can be mistaken for a human made creation. From my findings, I believe that Claude AI cannot spot distortions within images (in this example, the eyes of the subject are distorted) and Claude mostly focused on ranking the images from most realistic looking, rather than, prompt following, realism and aesthetic all together in overall scores like what I've done to try to give each generator a fair score. Claude also can't really have their own opinion on what's aesthetically pleasing since Claude cannot have certain opinions like humans which requires human emotions/feelings. Additionally, I the prompt asked for a building, not specifically an abandoned castle, perhaps Claude thinks the castle building is more interesting or unique than any other kind of random plain buildings other generators composed?

Evaluation of Claude's ranking: I like how Claude tries to evaluate what kind of building it is rather than just "some building" I agree that it does look more like an apartment building than any other building.  I also ranked google labs as rank number two so its cool to see that our ranks align with each other. I also like how Claude uses multiple adjective such as "Excellent, Perfect, Nice etc." to describe what it feels about certain details. Claude had also spotted that the art style was specifically in anime style and not in Disney's style. It's also interesting to see that Claude had complimented this image's details a lot more than the one it ranked at number one, showing that it appreciates the artistry of the illustration more than a "photograph". I've also notice that it doesn't specifically say if it looks AI-made or human-made, I think Claude got confused and interested as "realistic" instead which is why Piclumen was ranked 1st because it seems more like a realistic photograph.

Evaluation of Claude's ranking: It's interesting how Claude refers to it as "grey building" rather than "abandoned grey building" so they picked up on the fast that the building doesn't indicate that it's abandoned necessarily. I'm also starting to believe Claude cannot pick up on things that are anatomically incorrect, since the girl in the image isn't leaning against anything in particular, the pose she's in would realistically make her fall to the ground. Claude did not note that she wasn't necessarily overlooking the building, rather in the opposite direction. Claude and I have agreed with the same ranking placement once again.

Evaluation of Claude's ranking: I noticed Claude didn't mention any negatives about this image, yet he ranks it low, maybe I should've had Claude go more in-depth or into detail since it was probably dispositioned/programmed to not go into so much detail unless the user requests it specifically. Claude notices that the lighting is off from the prompt but can still recognize that it's still good lighting. Reading back at the other style descriptions, Claude doesn't precisely state in a definite way if the style is either in anime or Disney, it seems to be unsure but it leans in the right direction regardless.

Evaluation of Claude's ranking: Claude was able to differentiate school uniform from hiking attire. I really like how Claude describes what kind of building it is, it gives a sense of what the viewer may interpret without knowing the context of the prompt. It's getting harder and harder to see why Claude ranked the way it did, because for the past few rankings from above, they point out a single negative detail in each but it's not clear why one is better than the other exactly. I feel that my ranking system is way more comprehensive, because I give more in-depth explanations and score. The viewers can see exactly how many negatives the image has and how much it weighs with the point system I've created. 

Evaluation of Claude's ranking: Claude observed the building's design which is very fascinating. I do not agree with when it thinks that it doesn't look abandoned and that the characters outfit does not look hiking specific. It's kind of cool to think about how one AI model is correct while the other says it's incorrect. Since Claude cannot generate images at this time, I wonder that if it was able to now, it would make mistakes and be hypocritical by judging another generator's image "mistakes" just to make worse mistakes! Like it's self aware of mistakes after showing what not to do yet does it anyways.  

Evaluation of Claude's ranking: Claude states more negatives in this image than the ranking below it which doesn't make sense since they ranked it above but I guess it prioritizes "realism" over no matter how badly it follows the prompt. Claude didn't pick up that it seems to be a mixture of both forest and urban instead of just urban. 

Evaluation of Claude's ranking: Claude notices how detailed or less detailed a composition is. Claude describes the building as "modern" just because it's white but a "modern" building can still look abandoned so the logic falls there. Claude also doesn't specify what exactly isn't "detailed" such as the leaves.

Evaluation of Claude's ranking: I disagree with Claude when it thinks it's "beautiful autumn colors" because to me the colors are dull so I have a feeling it cant interpret colors all too well. I really like how it notices that the building is way too far in the distance which is a very interesting observation that I didn't think of. We both agree that this image belongs in the bottom of the rankings. 

Regarding AI detection:

- Piclumen (Rank 1) and Hugging face (Rank 9) appear to be actual photographs

- Ranks 2-8 show various signs of AI generation (consistent lighting, perfect symmetry, similar character designs)

- The artistic styles are very consistent within each image, which is often a hallmark of AI generation

- The backgrounds and characters have that characteristic AI smoothness in textures

The most "human-made" looking ones would be Rank 1 and Rank 9 (as they appear to be actual photographs), followed by Rank 1 which has the most consistent and intentional-looking artistic style.

Evaluation of Claude's observations: It's weird how it ranked the top rank and the lowest rank to be the most "realistic" but didn't mention Art Breeder since that one was also in realistic or photograph style. I disagree with Claude stating that the character designs seem to be similar, I think they're all pretty different since they all look like different girls. I also think the art styles are pretty different and not too similar, similar to artists in real life. Maybe Claude is getting general information about how AI art differentiates from real art rather than looking at the specific images I submitted myself. I think the rankings are mixed up in general and that Claude hadn't properly ranked as I asked it to. 

Prompt 2 Rankings:

Evaluation of Claude's ranking: We both ranked DALL-E on top. Claude notices that the way the illustration drawing technique such as the linework and cross-hatching. I also like the way that Claude is so sure in it's answer of "Definitely appears human-made." It also points out the perfection in the lamp and wall picture placement. Claude rates this image out of 10 points claiming it's a perfect image, although I spotted 3 inconsistencies such as the lighting shape curving around the left characters head, the card designs and that the two characters are the same exact design. This rating made Claude seem more human-like in his response, but it almost seems like Claude didn't take the time or "thinking power" to find the smallest of details which were the inconsistencies I pointed out. Since it said it was a perfect 10/10 I wonder how similar Claude would generate it's images, if it had an image generation function. 

Evaluation of Claude's ranking: Claude cannot pick up on lighting logic, since the lamp's lighting is inconsistent. Claude does not mention whether or not they look like they're playing cards and in this image they aren't, but I still agree with Claude in the fact that the art style and detail was generated very well and cleanly.  

Evaluation of Claude's ranking: It's interesting that Claude thinks this image is "more detailed than requested" but the image he ranked above this one is just as detailed, if not more. I think Claude may see more cartoony styles as "less detailed" and realistic looking style as "detailed." Claude also says that this image is "convincingly human-made" rather than the one ranked above it which is weird. 

Evaluation of Claude's ranking: Claude doesn't mention any negatives which feels like it negates the rankings, the viewer may think "so what makes this image worse than the image in rank 2?" Claude was also succesfully able to detect that this image was less detailed as requested in the prompt. 

Evaluation of Claude's ranking: Claude was able to interpret the lighting settings perfectly by describing it as "dramatic." Claude didn't explain what exactly the "texture handling" was. Able to detect every lacking quality except the fact that they're the same exact characters and the cards are over saturated which makes it lack detail. 

Evaluation of Claude's ranking: Claude said everything right except for pointing out the same character design.

Evaluation of Claude's ranking: Claude said everything right except for pointing out the same character design.

Evaluation of Claude's ranking: I'm surprised that Claude pointed out that the hair styles did not match with the timeline of the prompt. Claude also mentions that it's 'obviously AI" which I like. Claude can spot what's the modern style and what's the past style well.

Evaluation of Claude's ranking: I disagree in the fact that Claude rated Art Breeder's image higher than this image. Art Breeder's image seems like it didn't follow the prompt at all while this image, at least attempted to be more accurate. I believe that Art Breeder's image is way off from the prompt and aesthetically looks way worse than this image, but I do agree with the other aspects it stated above.

Prompt 3 Rankings:

Claude: I'll analyze each image based on the prompt requirements and rank them according to both prompt accuracy and photorealism. Let me break this down systematically:

Key prompt elements:

1. Aquarium setting with marine life (sharks, tropical fish, dolphins)

2. Soft blue lighting

3. Child viewing from left side

4. More vibrant right side with shimmering effects

5. Cool, soothing color palette

6. Natural, human-made appearance

Evaluation of Claude's ranking: Claude misinterpreted this image as a photograph when really, it's obviously fictional and seems to be drawn which is interesting. Not only is the art style unrealistic but the logic of the tanks on the left is also unrealistic yet Claude believes it to be "passable as a professional photograph." I like how in-depth and descriptive Claude is when describing the color balance and lighting effects, which I totally agree with it on. I find it really good that Claude was able to determine that this image has natural posture and scaling, it describes the image way better than me. 

Evaluation of Claude's ranking: Claude finds this image to be "emotional" which I find interesting since I don't really find it any more "emotional" compared to the other images. Claude doesn't find the distorted or unrealistic aquarium to be that odd, but still describes it as a "human photograph." Claude also didn't notice the lack of sharks which was requested in the prompt.

Evaluation of Claude's ranking: Didn't mention missing dolphins. I liked how Claude noticed that it looks like an actual realistic aquarium setting compared to the other images.  

Evaluation of Claude's ranking:  I find it odd that Claude mentions that the child's proportions aren't natural, since I don't know what exactly it's referring to since I think the proportions are fine. It was also smart enough to detect that it was in the style of photorealism but not perfect which I agree o.

Evaluation of Claude's ranking:  I disagree on Claude about the child's proportions, I feel like the head is too big for the body deeming it a bit unrealistic. I also don't understand what it means when it says "Limited right-side vibrancy contrast" since the image seems consistent on all sides in terms of lighting. 

Evaluation of Claude's ranking: Not sure what "Less natural child positioning" means but I'm guessing the angle of the limps make it look like the child's position is awkward but to me seems natural because toddlers can stand in weird ways especially since their limbs are a bit different from a grown person. I wish Claude would go more in detail as to what makes it "More obviously AI-generated" so I could have more to comment on and know exactly what it's thinking.  

Evaluation of Claude's ranking: This image very similar to Microsoft Copilot's image yet Claude ranks it a few tiers lower which is interesting since they use the same model. This child looks even smaller then the Microsoft Copilot one yet apparently this one is has even worse child proportions. At this point I feel like Claude is barely trying or getting more and more inaccurate.

Evaluation of Claude's ranking: Another image that is "emotional" to Claude which I find interesting but still not sure what exactly makes it emotional compared to the top ranked images.

Evaluation of Claude's ranking: Claude is wrong about the "Minimal right/left contrast" because this was one of the only generators that listened to the proper positioning of the prompt instructions while the other ranked images that are higher on this list put the child in the middle of the image which is not what was instructed, unless Claude somehow means the lighting contrast and not the child's positioning.

Evaluation of Claude's ranking: It's interesting to see Claude view this image as the "most stylized appearance" even though the top few images had a better art style in my opinion, unless Claude is talking about the composition of the image where it finds it unique that the girl is inside the aquarium and viewing the fish. 

Midjourney and Piclumen stand out as the most successful, achieving both prompt accuracy and photorealistic quality. They demonstrate superior technical execution in lighting, composition, and natural human representation. The lower-ranked images tend to show more obvious AI artifacts or deviate significantly from the prompt specifications.