Rules Winners of Fake Contest images missing ??

[paperscarf]

The list of winners including the First Place is missing in your post.
Also it would be better if you could post the images of the winners when the announcement is made.

One suggestion would it not be better to bring the minimum fakes for the contest to be 2 rather than 4.
Almost all fakes posted is AI generated low resolution patch up job with little effort in quality.
The the difference between fakers who put in the effort to make it look real and others who post a dozen mediocre fakes is obvious.

Hope somebody reads this and is not drowned in the messages.
Thank you.

 

[MPKMNF]

correctly pointed out, it takes hours to create a fake , and in secods these AI engines renders some loose shit , and these should not be considered for even contest.

Its like I am not against the idea and concept of fakes creation using AI, but there needs to be separate threads and contest , so that the original work gets the due recognition.


Hope the admins reads these forums!!!

 

[paperscarf]

Totally agree.
I have been playing with SD models from Day 1. You can make really cool fakes.
But it takes time to push it to 4K resolution without artifacts.
Here users are just copy pasting a 512px on to a 3k resolution image with no zero compositing.

Just for the last contest, some body has posted around 20+ images.

That's why I feel the minimum images should be 2 or even just one image.
One fantastic fake image rather than a dozen garbage fakes.

Hope the admins are reading this.

I sometimes wonder that I could take all these fakes of one actress and train it on SD 1.5.
Then you can pump out infinite fakes.

 

[MrGhostRider]

I sometimes wonder that I could take all these fakes of one actress and train it on SD 1.5.
Then you can pump out infinite fakes.

your idea is good but not quite right. You don't need to feed actress fakes to train... just the normal pics enough... And i am doing exactly that... you can checkout my thread... you can generate normal pics as well as fake for 'trained' actress.

 

[paperscarf]

Training with just actress fakes you might get close face representation but the body shapes will be generic and you would go about ploughing prompts to get appropriate body shapes.
Assuming the input fake images are made with similar body proportions of the actress's, the trained model will churn out similar outputs for the body shapes.

Checkout a few samples trained on SD 1.5

 

[paperscarf]

On my earlier point, could the minimum entry count be reduced to two for competing in the contest.

On each contest, you find one or two really excellent fakes that really standout made with PS I guess, compared to the rest of the lot.

Even the team at Desfakes would agree to have two great fakes rather than many average fakes.

Thanks.

 

[Beinghuman]

All suggestions are welcome for season 5. We are ending season 4 next week. Can't do much changes in season in between. When season start we locked the format. AI fakes are new and came in between season that's why there is conflict between AI and Photoshop fakes. We will try to sort out all queries in next season. Thank you.

 

[paperscarf]

Thank you responding.

Let me make it clear, I have no issues people using SD to create fakes.
Fake is just a fake, be it created in SD, PS or any 3d modeling tool.
Different workflows SHOULD push the quality of the fakes created.

Also with you guys locked up with server migration, any site assets creation can be put to your users,
who I assume are seriously skilled in many graphical software.

Please do consider reducing the image count for the contests and still be eligible to win.

Cheers.

 

[paperscarf]

With the season 4 coming to an end, hope you would consider brining the minimum fakes required to compete to win the contest from 4 to 2.

Hope you could internally have a discussion among the site administrators.

Thanks.