The purpose about that rule is to do with quality, since historically the technology didn't exist to just magically upscale an image without the quality really suffering. Obviously the last few years has seen a lot of amazing things happen in this space.

I don't think it matters how someone goes from 300x450 to 600x900 so long as the quality is good. It's only ever been about quality.

To be super specific, I am talking about upscales. I think there's a different position we may need to think about when it comes to purely AI generated content.

@travisbell said:

The purpose about that rule is to do with quality, since historically the technology didn't exist to just magically upscale an image without the quality really suffering. Obviously the last few years has seen a lot of amazing things happen in this space.

I don't think it matters how someone goes from 300x450 to 600x900 so long as the quality is good. It's only ever been about quality.

To be super specific, I am talking about upscales. I think there's a different position we may need to think about when it comes to purely AI generated content.

But what if it looks strange though? imo there are definitely instances where an AI upscale increases the quality but ultimately changes the way an image looks. Like "AI" is almost an aesthetic/artistic-style (like as the name goes, an artificial look)- changing the intended aesthetic of the official poster/still produced by the makers/distributers of the movie/show. There are instances where this isnt the case. Although I feel like if you allow to make exceptions about AI then it will become a difficult issue of determining what looks passable as non-AI and what looks like it was processed through AI. In any case I think there should be information about this in the contribution bible- there currently isnt as far as I can tell anyway.

There are some examples of pictures that I believe were AI upscaled that I have reported and those images have been since deleted by moderators. Also I feel like you didnt answer Kampfkeks question, if AI upscales increase the quality of an image, are all successful AI upscales (that increase quality) allowed, or should the ones that look like AI be not allowed?

Off the top off my head this is one example I can remember: https://www.themoviedb.org/movie/1213-the-talented-mr-ripley/discuss/668c45fae1cfd4da2379ec15

(I could be wrong, maybes its not an AI upscale)

Edit: Here are some relevant reports, some still open some closed:

https://www.themoviedb.org/talk/64e67d1006f98400ae47f4a0

https://www.themoviedb.org/talk/669e575d9b3d153df8805387

https://www.themoviedb.org/talk/65fb0b5c0bc529017cae98f4

https://www.themoviedb.org/talk/65c7853398f1f100c9d5da62

https://www.themoviedb.org/talk/65d74435992fe6017cf9555e

https://www.themoviedb.org/movie/220289-coherence/discuss/669f615c69cf97f37a70362d

There are more examples of reports but yeah.

(I personally am against AI upscales as imo its essentially the same thing as fanart since its basically third-party editing imo- its not a simple crop or a downscale which are categorically a different type of edit that simplifies information in an image down, wheras AI upscales imo adds information to an image- although I guess maybe it would be impossible to enforce a no-ai image policy since im sure some images that are ai upscaled simply dont look ai upscaled)

Most of the 2000x3000 Netflix posters are upscaled (AI or not) because the common resolution of the Netflix posters is 1500x2222 (with rare exceptions). Most of them look ok though, when done by someone who knows what they are doing. Most of the "clean" posters are third-party edited, sometimes with the help of AI tools. IMO as long as the image looks good, it doesn't matter how it got there. AI is not the problem. Humans, on the other hand, usually are.

@shotfirer said:

Most of the 2000x3000 Netflix posters are upscaled (AI or not) because the common resolution of the Netflix posters is 1500x2222 (with rare exceptions).

...is that a good thing/should they be?

@softpillow said:

@shotfirer said:

Most of the 2000x3000 Netflix posters are upscaled (AI or not) because the common resolution of the Netflix posters is 1500x2222 (with rare exceptions).

...is that a good thing/should they be?

In my opinion, they shouldn't. But they already are. It looks like most of the contributors who "clean" the posters consider the upscale a part of the process, so the "clean" ones don't exist in their original resolution. But as long as it's done properly, and they look good I don't see any reason to specifically hunt and delete them. You might notice the difference when directly comparing (e.g. this to this), but when judging the properly upscaled posters on their own in 9/10 cases they are decent enough.

@shotfirer said:

Most of the 2000x3000 Netflix posters are upscaled (AI or not) because the common resolution of the Netflix posters is 1500x2222 (with rare exceptions). Most of them look ok though, when done by someone who knows what they are doing. Most of the "clean" posters are third-party edited, sometimes with the help of AI tools. IMO as long as the image looks good, it doesn't matter how it got there. AI is not the problem. Humans, on the other hand, usually are.

Oh there are AI tools to clean posters? damn is that relatively new or is that how all the clean posters on tmdb get done for a long while now?

@softpillow said: Oh there are AI tools to clean posters? damn is that relatively new or is that how all the clean posters on tmdb get done for a long while now?

Content-aware and generative fill algorithms that involve neural net computing have been available for quite some time. Sometimes this is enough to clean the poster, depending on where the unwanted text elements are placed.

@shotfirer said:

@softpillow said: Oh there are AI tools to clean posters? damn is that relatively new or is that how all the clean posters on tmdb get done for a long while now?

Content-aware and generative fill algorithms that involve neural net computing have been available for quite some time. Sometimes this is enough to clean the poster, depending on where the unwanted text elements are placed.

Oh you mean like the resynthesize and heal selection tools on photo editing programs like photoshop and gimp.. I didnt realise that was considered AI... huh

@softpillow said:

Oh you mean like the resynthesize and heal selection tools on photo editing programs like photoshop and gimp.. I didnt realise that was considered AI... huh

AI or cloud-based algorithm or whatever, it doesn't matter. It is still third-party editing beyond "a simple crop or a downscale" even when done purely old-school, manually, with brush and stamp. So what's the difference? What matters is the quality of the result. I've seen enough manually cleaned posters that looked awful and cleaned a fair bit myself, sometimes getting surprisingly good results when using only AI tools.

Anyway, my point is - if the image looks bad, it doesn't really matter why, just report it and it will be gone eventually :)

@shotfirer said:

Most of the 2000x3000 Netflix posters are upscaled (AI or not) because the common resolution of the Netflix posters is 1500x2222 (with rare exceptions). Most of them look ok though, when done by someone who knows what they are doing. Most of the "clean" posters are third-party edited, sometimes with the help of AI tools. IMO as long as the image looks good, it doesn't matter how it got there. AI is not the problem. Humans, on the other hand, usually are.

Does anyone know what those ai tools being used to upscale the posters are? If they are actually successfully upscaling images (increasing quality) without creating a strange AI effect, then I feel like everyone might as well learn to use them (depending on how complicated it is, etc.)

Anyway my concern is just that I have noticed some images on the site that I believe to have been upscaled by AI. They have increased in quality, but have changed in style. So even though the quality is higher its still to me equivalent to someone manually adding filters/special effects to images (i.e. the same as fanart).

just sharing some more of my thoughts here:

I've been experimenting with some of the free ai-upscaling websites and what I have found/concluded, at least for me anyway, is that it can work pretty well for upscaling images that are already "fake" or "digital", for example some posters are just made up of digital graphics (text/shapes, etc.). Here are two examples that I think look fine:

https://image.tmdb.org/t/p/original/xSB0IQUn0a0WQa3UltwcR2a66f.jpg

https://image.tmdb.org/t/p/original/mbO5YlXWpVVxNgsy3aOv72qsttg.jpg

They look fine right?

But when it comes to upscaling with ai, on images containing people thats when it it runs into problems- I guess the textures, shapes and colors/shades in human features/skin is much trickier to upscale. Or at least it becomes much more noticeable that the image has been manipulated with ai. It becomes more of a grey-area/subjective thing whether the results of the upscaling are seemless/effective (sometimes obviously bad, and sometimes more iffy)

Although the issue might be that I'm not using more advanced ai or am not using it correctly, I will still stand by that there are images that are uploaded to this site (images with real people) that are badly upscaled with ai (I guess they are being uploaded by users that are also not using more advanced or using ai with enough expertise)

The main problem with profile images is that mostly the images are scaled up from 300x450 (or even less) to 2000x3000. The result is a monster like the one in my first post.

I have also experimented a bit with this in the past and found that it can lead to fairly acceptable results if you downscale again after upscaling. One user, for example, turned this image into this one - which is quite good. At the minimum size, I probably wouldn't even have noticed that anything was done with AI here.

Personally, however, I would wait. In the future, this technology will certainly become even more sophisticated and at some point it could really be applicable. At the moment, the results are so-so. But I'm also more interested in older films/shows where there are usually only low quality images, that could benefit from such an upscale. With the latest blockbuster, this is of course completely superfluous.

Here are some recent examples, I thought I'd just put these here to show what I mean:

https://www.themoviedb.org/tv/62688-supergirl/discuss/670c73a74df59a608c636e12

https://www.themoviedb.org/movie/930094-red-white-royal-blue/discuss/670c7229f58a9206aa4133ea

Im not 100% sure. but just to illustrate what I mean by the images that Ive been seeing recently occasionally that I think are probably ai-ified (and not in a good way).