DPChallenge: A Digital Photography Contest You are not logged in. (log in or register) 
 

DPChallenge Forums >> Side Challenges and Tournaments >> The ongoing AI generated imaging Side Challenge
Pages:  
Showing posts 26 - 50 of 68, (reverse)
AuthorThread
 01/20/2025 12:53:26 PM · #26
An attempt to mimic the Backside II blue ribbon winner which ended up quite short



I like to see that one though:


Initial prompt: I would like you to help create an image of a bird shot from behind and to the right using a 300mm lens at f6.3. The bird is standing on a brick fence covered with yesterday's snow. The bird has a nice little cheeky Mohawk it is proud of. It looks back at the camera and appears to be saying something like "Like what y' see?! PERV!".

Full session in the comments.

I think the problem with Text-to-Image is that it removes the mystery factor. Once everything has been verbalized, nothing is left for us viewers to work with and imagine. Compare that with an Image-to-Text process, that the viewer is actively involved in. Even other people's text/thoughts/critics of an image doesn't shut you off.
 01/20/2025 01:32:42 PM · #27
Originally posted by MargaretNet:
Originally posted by posthumous:
[quote=MargaretNet] I am not interested in looking at shortcomings of AI, I could not paint Mona Lisa from the back either, what kind of test is this? What does it prove? Why not just have fun playing with ideas??


...
Doesn't anyone else want to contribute and play with AI?


Willem was here.

But I am not interested in extensive play with AI.
 01/20/2025 06:07:04 PM · #28
Originally posted by MargaretNet:
Doesn't anyone else want to contribute and play with AI?

I'll play along as soon as I get familiar with an AI site. I've done one already.
 01/20/2025 06:08:47 PM · #29
Originally posted by MargaretNet:
Originally posted by posthumous:
I am having fun. Why don't you start playing and contribute AI images to this thread?

To do that I would need to become a member again as I would need to upload images, putting them on my Website and linking is a nuisance. I would be very interested to see output of multiple AI generation sites with the same prompts but if it is just the two of us contributing it will be very limited.

It's not worth $2.15/mo to play with us again? Just askin'...
 01/20/2025 06:44:52 PM · #30
I just made this one for you in ImageFX, Margaret :-) Nothing special, ("Modern kitchen with a view of the Matterhorn")

01/21/2025 04:38:17 AM · #31
Originally posted by posthumous:
backside challenge

Create a photograph of Mona Lisa from behind

ChatGPT said:
Here is an artistic interpretation of the Mona Lisa from behind, showcasing her seated posture and the Renaissance-style details. Let me know if you'd like any adjustments!

You said:
you are not behind her. the point of view should be from behind her looking at her back

ChatGPT said:
Here is the revised depiction of the Mona Lisa from behind, showing her back as she sits in her chair. Let me know if further adjustments are needed!



btw, i spent a lot of effort trying to make this. i was (and still am) fairly certain that it's not really a training data problem (there'll've been many pictures of women's backs sitting down) so took the two-stage strategy: first try to compose a picture of the pose that I want, and then to transfer the style of the Mona Lisa to it. The main problem is that I wasn't able to get the style transfer quite right, and in the end none of them were really recognisable as her. But that being said, I'm not an expert at this kind of pompting and guidance. I think it should be possible but I think I have a tooling gap.

Here are some of the outputs. The first one was fun (albeit unsuccessful) output of it though. I like it a lot stylistically, but it is so far away from what I wanted :-D The second one is the image that I tried to make not as the mona lisa, but to do a second-step of style transfer on it. You can do some nice things with more controllable tools like doing edge detection and using that as a control layer, then you can set the model to a high level of creativity (with your prompt). So it would keep the edges, and fill in the gaps a bit differently. Kind of like painting by numbers, where the numbers are defined by the style you pick. Maybe you'll have more luck with the style transfer than me!


Message edited by author 2025-01-21 04:38:51.
 01/21/2025 05:21:06 AM · #32
You are probably right. You can achieve it if you describe the style elements step by step and as long as you don't mention "Mona Lisa".
As soon as you mention "Mona Lisa" the image of the real Mona Lisa takes precedence because it is the only image of Mona Lisa available.
I did also try that example of a watch showing 33 minutes past 6 and it is true, it will show a watch with the hands 10 past 10 and additionally show a "digital" minute counter with 33.
When doing this kind of work I think we have to keep in mind it does not have "imagination" (not yet anyway) , it can only reproduce, but in a very extensive and impressive way.
 01/21/2025 05:26:19 AM · #33
This is fun. I first asked for an image of the Matterhorn from the north and then from the south. The direction of the light and the vegetation and foreground has changed, but the rest is the same view.


Message edited by author 2025-01-21 05:27:57.
 01/21/2025 05:50:42 AM · #34
Originally posted by willem:
I did also try that example of a watch showing 33 minutes past 6 and it is true, it will show a watch with the hands 10 past 10 and additionally show a "digital" minute counter with 33.


This is almost certainly a case of training data like you speculated earlier. The slightly subtle difference between this case and the mona lisa one is that in this one there is a definitive and objective aim (hands being at 6:33). A simple and quick check by anybody can verify whether the generated image is right or wrong. There is no subjectivity in the task. On the other hand, whether the mona lisa one is "correct" (for some definition of correct) is a question of judgement. Slight (and even major) changes in the resultant image could all reasonably be marked as good generations, meaning that assessment of task completeness is far more subjective with mona lisa's back than the watch example. There are traits that we would identify (style consistency, identifiability, etc) but those req's are much softer than the watch example.

Both the watch and mona lisa are great examples I think.
 01/21/2025 09:07:11 AM · #35
Originally posted by willem:
This is fun. I first asked for an image of the Matterhorn from the north and then from the south. The direction of the light and the vegetation and foreground has changed, but the rest is the same view.

Hah, that's very funny :-)
 01/21/2025 10:28:03 AM · #36
Originally posted by NiallOTuama:
The slightly subtle difference between this case and the mona lisa one is that in this one there is a definitive and objective aim (hands being at 6:33). A simple and quick check by anybody can verify whether the generated image is right or wrong. There is no subjectivity in the task. On the other hand, whether the mona lisa one is "correct" (for some definition of correct) is a question of judgement.


The difference isn't subjectivity. Both cases require understanding (not "imagination"). The AI simply cannot do it without an additional program add-on. A program that calculates hand position based on time is probably easier to write than a program that distinguishes back images from front images. (I say probably, I don't know all the challenges involved in either one)

A great example is chess. Try to place chess with a chatbot. Not only is it not good at chess, it will eventually play an illegal move. UNLESS someone added a chess-playing program to it. Then it will beat you just like any other chess program.
 01/21/2025 12:22:08 PM · #37
I was going to upgrade to member but seeing the most recent posts I lost interest. Proving that AI is stupid is not my idea of fun.
 01/21/2025 12:53:00 PM · #38
Originally posted by MargaretNet:
I was going to upgrade to member but seeing the most recent posts I lost interest. Proving that AI is stupid is not my idea of fun.

I don't think we are trying to prove AI is stupid. We are just playing around and in doing so also run into its limitations.

I did not expect the Matterhorn to come out the same from both directions. I was really curious how it would look.

At least we are trying.

Tomorrow I will post an AI version of my image that is currently in voting.

Message edited by author 2025-01-21 12:57:16.
 01/21/2025 09:47:08 PM · #39
Originally posted by MargaretNet:
I was going to upgrade to member but seeing the most recent posts I lost interest. Proving that AI is stupid is not my idea of fun.

If some of us actually keep posting AI we can get the thread back on track. I'll make sure of it. JUmp back in, Margaret!
 01/21/2025 11:08:22 PM · #40
Stop fighting
Anachronism
ImageFX
2 photographers duelling each other like gunfighters with cameras


Message edited by author 2025-01-21 23:10:32.
 01/22/2025 02:02:52 AM · #41
Entry of the extended free study

AI generated "image of a blue tit on a birdfeeder with a sunflower seed in its beek"


Message edited by author 2025-01-22 02:03:31.
 01/22/2025 03:26:40 AM · #42
I used Adobe Firefly
Prompt: Mountain Piatra Craiului seen from Rasnov Fortress, in black and white, with snow caps on top and a sea of clouds. landscape aspect ratio 16X10

Entry

AI
01/22/2025 05:02:38 AM · #43
And using a different image generator within chatgpt

and then with the additional prompt to get a closer view and make it more photorealistic

I think it does not have the proportion of the seed version the bird right and also it is better at generating humans than birds, more photorealistic.
The colors and patterns of the bird are quite good.

Message edited by author 2025-01-22 05:03:32.
 01/22/2025 05:12:17 AM · #44
Originally posted by willem:
Entry of the extended free study

AI generated "image of a blue tit on a birdfeeder with a sunflower seed in its beek"


Just out of curiosity I experimented with your prompt using Photoshop. It refused to work with the prompt until I changed "tit" to "bird".
 01/22/2025 06:10:39 AM · #45
Originally posted by GinaRothfels:
[quote=willem] Entry of the extended free study
Just out of curiosity I experimented with your prompt using Photoshop. It refused to work with the prompt until I changed "tit" to "bird".

Censored. LOL
 01/22/2025 09:55:15 AM · #46
OK, I am in. As usual DPC costs me money! ;) 25$ DPC plus 10$ for one month of Midjourney as ImageFX is not as yet available in Switzerland. Anyway, this is nothing compared to all the cameras and lenses I bought over the years just to enter challenges! :)

I noticed that most posters to this Side Challenge ignored the OPs rules of the game, I am reposting them here:

• Pick a Challenge: Select any completed challenge as your reference.
• Create with AI: Use any generative AI tool to produce an image for that challenge.
• Upload Your Image: Share your creation with a fitting title.
• Include Key Details: In the image comments, provide:
1. The reference challenge
2. The AI engine used
3. Your prompt (including any custom options and links to reference images if applicable)
• Share Here: Post a thumbnail of your image in this thread, along with the challenge name and image title.


I will post something later Swiss time.

PS Bear, I am in Zermatt at the moment so your image was a good guess :)
01/22/2025 01:50:35 PM · #47
An attempt to mimic my entry from Feast VII in November 2023.

I ended up with the prompt "pigeon standing on a half-eaten loaf of bread. another pigeon is standing to his left watching him eat".



I like it better than my photo:
01/22/2025 02:39:17 PM · #48
I'm looking fairly closely at the pigeons and I'm finding no obvious hints that it is AI... It's a very nice "photo" isn't it?
 01/22/2025 02:50:46 PM · #49
Originally posted by Bear_Music:
I'm looking fairly closely at the pigeons and I'm finding no obvious hints that it is AI... It's a very nice "photo" isn't it?

Indeed much more photorealistic than the blue tit that I generated using two different chatGPT engines. Those look like drawings. The pigeon looks real.
01/22/2025 03:16:42 PM · #50
I'm not using any AI stuff right now, but maybe one of you bird fanciers might want to have a go at this picture of mine ...

Pages:  
Current Server Time: 08/02/2025 03:22:48 PM

Please log in or register to post to the forums.


Home - Challenges - Community - League - Photos - Cameras - Lenses - Learn - Help - Terms of Use - Privacy - Top ^
DPChallenge, and website content and design, Copyright © 2001-2025 Challenging Technologies, LLC.
All digital photo copyrights belong to the photographers and may not be used without permission.
Current Server Time: 08/02/2025 03:22:48 PM EDT.