PP Enhanced (Portraits Portraits Everywhere Overhaul)

[Shipwreck Jones]

To Mods: I couldn't create a topic in Unreleased Projects, so that's why I put it here.  Please move as necessary.

Wut Iz?

I really liked the idea of the PPE (mod that gave portraits to every character with a speaking role in BG 1 & 2), but found the portraits rather disappointing.  Now, to be clear, I have zero doubt that the mod took a tremendous amount of time and effort to put together, and certainly appreciate all the work that all the contributors put into it.  I just want to see the mod improved, and this is my attempt to do so.  As I see it, the PPE suffers from four main problems:

1. Portraits inconsistent with character models and/or personality
2. Inconsistent art quality
3. Inconsistent picture quality
4. Inconsistent art style

I aim to rectify this by creating new portraits using a combination of A.I. and photoshop.  I've already created about 275 portraits, but will probably want to revisit the first 50 or so, as my quality standards kept going up as I worked.

Why am I posting now?

I'm reaching out to the community now for two reasons.  First, I was wondering if anyone wanted to contribute.  PPE has approximately 3,600 portraits in it, and figure it would take me about 4 years to finish if I did it all by myself.  Naturally, I would welcome any help.  Many hands make light work, and all that.

And since I was going to post anyway, I figured I would ask if anyone had any input.  I'm thinking about switching A.I. image generators, and was wondering if anyone had any advice.  I've been using Dream by Wombo, but have been experiencing errors sporadically over the past two weeks, which will temporarily bring progress to a standstill.  On the plus side, the free version doesn't have any image generation limits, is easy to use, gives good results, and allows the use of reference images that influence output.  However, it also doesn't allow any sort of masking (allowing you to preserve parts of an image you like), the interface is mildly annoying, and, as mentioned previously, it has been experiencing periods of downtime recently.

Of course, I'll hop on their discord to find out what I can about Dream, but was wondering if anyone had any suggestions for alternatives.  Requirements:

1. Relatively Free or Completely Free (in order to maximize potential contributors)
2. References Images OR Masking (in order to reduce the amount of work that needs to be done in an image editor.)
3. High Quality Final Product
4. Reasonable Image Generation Times

Nice to have:

1. Completely Free
2. Reference images AND Masking
3. Easy to setup and use (but I will make tutorial videos if I have to)
4. Offline (so we don't have to worry about changes to monetization, changes to image generation limits, changes to the program or models that we have no control over, and times of heavy usage)
5. Fast Image Generation Times

I was thinking about trying Stable Diffiusion, but saw how complicated it was and decided to ask around first.  And, yes, there are plenty of articles about A.I. image generation, but a lot of them are either trying to sell you on something and/or written by A.I.  Moreover, the landscape is changing rapidly.  And, if you're wondering why I didn't do a bunch of research before starting such a large project, it's because the scope of the project gradually grew from me just wanting to make 6 custom portraits for a party for my own playthrough.

Thinking About Contributing?

You don't need any paid software, high end hardware, or specialized skills.  I do highly recommend the use of a photo editor, as it will make life much easier.  However, A.I. handles all the heavy lifting, and I will probably make some video tutorials in order to make getting into it as easy as possible.  Thus, the only things you really need are some free time and a desire to help.

Coordination

I have already made a spreadsheet to keep track of my work.  If people are actually interested in helping, I'll convert it to a Google Doc and create a Discord.

Image Comparisons

I have uploaded some before and after comparisons to this post, but am limited on file size.  Thus, I am also providing a link to additional comparisons.  Hopefully, they'll help get people interested in helping.

And, yes, I’m aware that my images are in the wrong format.  I will use a Photoshop Action to batch convert all the images at once.

[ktchong]

I tried to use AI — my favorite is AI Perchance Character Generator but there are better ones out there — to create my own replacement portraits for "Portrait Portrait Everywhere", but it was too much work for one person.  I tried to match the portraits to their onscreen avatars (appearance and color schemes) as well as any relevant DnD lore, (e.g., elves do not have facial hair; Phlydia is Halruaan = Arab/Middle Eastern/Persian; werewolves in human forms have glowing red eyes, a gnoll that looks like a proper gnoll and a half-ogre that looks a proper half-ogre, etc.)    All these portraits are "original" (i.e., not copyrighted) as they were created by AI that interpreted my detailed descriptions of what I wanted. 

The few exceptions are what have already existed online.  For example, there are lots of Drizzt portaits available, so there was no need for me to generate my own.  I grabbed Volo's portrait from a painting that appears in the DnD movie "Honor Among Thieves".  So those were not generated by AI.

I am not using any portrait mods in my game because, frankly, the quality of portraits in those mods are inconsistent and horrid.  The two popular portrait mods just grabbed a bunch of different existing portraits (many of which were copyrighted) from the Internet and packaged them.  Of course, those portrait mods were made before the emergence of AI.  Now we can use AI to generate and create images that are NOT copyrighted.

Anyway, here are what I had created with AI before I gave up:

Edited March 15 by ktchong

[Shipwreck Jones]

Yeah, I've been trying to match:

1. Head (i.e. Hood, Helmet, Hat, and Hair Color)
2. Race (i.e. Ears, Skin Color, Body Size, Body Proportions (which can be easier said than done))
3. Attire Color and Type (i.e. Plate, Chain, Leather, Noble, Peasant, Wench, etcetera)
4. Expression (i.e. At least when I knew the personality or situation of the NPC in question)

I didn't worry about making them match better than that (i.e. the specific look of the armor and hair length) because there isn't a huge variety in the character models.  So, if I tried to make them match exactly, the portraits would end up looking pretty samey.  Speaking of which, I could get good facial variety when using a reference image, but my faces tarted all looking the same when generating from scratch.  So, I started using celebrities in my prompts to try and give some variety.  And, whenever a character's skin color looked ambiguous, I would make them non-white (there are a crap ton of white humans, and white human males in particular, in the game).  Of course, I probably know a lot less than you when it comes to how different peoples are supposed to look.  So, I probably made mistakes in that regard.

1 hour ago, ktchong said:

Anyway, here are what I had created with AI before I gave up. 

I don't suppose I could entice you into continuing to work on the project.  Since you already know what you're doing, you can help me to help others to get into it, and with so few people working on it (so far just me), each person that decides to sign on can make a huge difference to the amount of work everyone else has to do.  Not only that, seeing that other people are getting involved might help others to feel like the project is actually feasible, which might encourage them to contribute.

I have no expectations.  I know it's your free time, so you can help as little or as much as you like, and I'd still be appreciative.  You can also work on any portraits that interest you (I'll happily pick up the portraits people don't want to do if it means more hands helping).  And if you don't want to help, that's fine.  I still appreciate you sharing.

P.S. Can I have the portraits you've already made?  You have around 50 that I'm missing, and I'd love to use them to fill in some gaps.

[Isewein]

Isn't it such a shame to use AI when there's so much wonderful real art out there? You know, the sad thing is that there was a portrait mod for BGT which had (what I think are) much better-fitting portraits than PPE overall, but it has been discontinued due to copyright concerns. At least the old way of picking art from across the internet willy-nilly gave some visibility to real people behind it! And maybe for a new version it would even be feasible to try and contact some of the artists behind it? I'm sure most would be delighted to allow use of their work. Now, I understand that copyright is important for artists and all, but I do think that as a community we should ask ourselves whether AI isn't far, far worse than copyright infringements for the future of artists. It seems to me a supreme irony that due to the bureaucratic fears surrounding copyright we can't use work from real artists, but we can use the anonymised Frankenstein creations AI cobbles up from countless samples of their work.

Edited March 15 by Isewein

[jastey]

4 hours ago, Isewein said:

At least the old way of picking art from across the internet willy-nilly gave some visibility to real people behind it!

I don't even know what to say to this. Both portrait mods - the BGT one and PPE - uses IP and copyrighted portraits from non-mentioned sources. This is the opposite of giving visibility to real people, it's just making use of their work without giving anything back.

@Shipwreck Jones If I understood correctly, you are planning on replacing all portraits from PPE? Because, please be aware that PPE uses copyrighted portraits without permission and without even giving proper credits. We do not need another adaption of a portrait mods that does that.

[Isewein]

Well, if I was an artist, I'd still much rather have my work included by someone in a non-profit mod, even without prior request or referencing, than have it scrounged up, mangled by AI and thus effectively stolen as well, just without the chance of recognition. Of course the gold standard would be a mod in which every artist received proper recognition, but the threat of AI seems far more grave to me, and it would be supremely ironic if, on copyright grounds, an AI portrait mod was permitted whilst one using real art isn't.

Edited March 15 by Isewein

[ktchong]

18 hours ago, Isewein said:

Isn't it such a shame to use AI when there's so much wonderful real art out there? You know, the sad thing is that there was a portrait mod for BGT which had (what I think are) much better-fitting portraits than PPE overall, but it has been discontinued due to copyright concerns. At least the old way of picking art from across the internet willy-nilly gave some visibility to real people behind it! And maybe for a new version it would even be feasible to try and contact some of the artists behind it? I'm sure most would be delighted to allow use of their work. Now, I understand that copyright is important for artists and all, but I do think that as a community we should ask ourselves whether AI isn't far, far worse than copyright infringements for the future of artists. It seems to me a supreme irony that due to the bureaucratic fears surrounding copyright we can't use work from real artists, but we can use the anonymised Frankenstein creations AI cobbles up from countless samples of their work.

 

PRACTICAL and PRAGMATIC considerations:

 

1. AI-generated arts are NOT copyrighted — because each AI-generated art/portrait is somewhat unique and there is no way to traces their sources. 

I have used Google Image Search and TinyEye to search for the "sources" of my portraits that were created by AI.  Google Image Search and TinyEye were NOT able to find anything even close to those AI portraits.  Those "artists" or sources of the AI portraits were untraceable.  The portraits were unique.  In fact, when I used AI to create portraits using the same exact descriptions and the same exact parameters, the AI never created the same art twice.  The portraits and results were always different.  If I did not immediately save an AI-created portrait when it was created, then I would never be able to use any AI to recreate the exact same image again (even when I kept reusing the exact same descriptions and parameters.) 

That means: there is no way to trace the sources or copyrights of portraits created by AI.   Portrait mods that include only AI-created arts would NEVER have to deal with copyright issues — or be ordered to take down due to copyright complaints.  I understand that AI does not actually create.  AI merely alters, combines and mixes pre-existing arts into new ones, but the alternation, combination and mixing are so drastic that the AI-created arts are essentially new ones that are often very different from their original components.

Even if, let say new copyright laws pass, and AI-generated arts become copyrighted... you will still NOT be able to trace the sources of AI arts, because no two AI-generated arts are completely the same.  They are always different and somewhat unique.  How do you even enforce copyrights on AI arts if you cannot even trace their sources or origins?

Given that copyright has been the BIGGEST problem with portrait mods that caused them to be taken down, I would say using AI arts to replace all the copyrighted portraits (which more or less violated copyright laws) is a GOOD THING.  Portrait mods will NOT ever have to deal with copyright complaints and takedown orders again.

 

2. Even if you can somehow reach the artists of the portrait and persuade them to let you use their arts, the portrait mods include hundreds if not thousands of images.  You are NOT going to reach all of the artists whose images are included in the mods, and certainly NOT all of them would agree to let you use their arts for free.

 

3. Also, different artists have very different styles.  My biggest problems with portraits mods: the portraits came from so many different artists, so many different sources, that they are very inconsistent.  Their styles and qualities differ drastically.   So, a portrait mod includes so many its portraits that look so different and have drastically different styles.

You are NOT going to get one or two artists to produce the hundreds of portraits for a mod, (to keep a consistent style for all portraits.)  However, you can use AI to create hundreds and even thousands of portraits that all have the same style.   Just look at the AI portraits that I have created that used the same parameters for their style, (i.e., "fantasy, realistic oil-painting, etc.")  They all came out very similar, as if they were produced by the same artist. 

By the way, I actually used three different AI generators, but they all produced images that follow very similar style... as if they were produced by the same artist  You are NOT gonna get that result with real artists, AND you are NOT gonna get the same one artist to produce hundreds of portraits for you to use in a mod.

 

Sure, you can complain about "let's use real artists," "let's get their permission," "let's get them proper credit," etc.  Let's be realistic here: you know that is NOT practical NOR pragmatic.  Let's not suggest to people to do something that is NOT gonna happen NOR produce any result. 

Edited March 15 by ktchong

[ktchong]

17 hours ago, Shipwreck Jones said:

I don't suppose I could entice you into continuing to work on the project.  Since you already know what you're doing, you can help me to help others to get into it, and with so few people working on it (so far just me), each person that decides to sign on can make a huge difference to the amount of work everyone else has to do.  Not only that, seeing that other people are getting involved might help others to feel like the project is actually feasible, which might encourage them to contribute.

...

P.S. Can I have the portraits you've already made?  You have around 50 that I'm missing, and I'd love to use them to fill in some gaps.

 

Not gonna do it unless you are able to get a dozen people to share the loads, with everyone agreeing to use AI to produce portraits with very similar parameters and styles, so all the portraits look consistent.   There are like hundreds if not thousands of portraits in Portraits Portraits Everywhere. 

BTW, I looked at your portraits... they still look very inconsistent.  Using AI, it only takes a short time to create a group of images from one set of descriptions and parameters.  However, most of my time went into researching and coming up with the descriptions, because I wanted to  match the portraits to their avatars and BG/DnD/Forgotten Realm lore.  

You can have the portraits, but I have not been able to update the zip files here because the sizes were too big.  You do not even have to give credits to me because I certainly did not create them.

Edited March 15 by ktchong

[Isewein]

Sorry, Chaotic Good here. My concern wasn't about the letter but the spirit of the law, and I still think it's absurd that we as a community (because let's face it, that's the ultimate instance here: nobody is going to launch an IP lawsuit against an obscure, non-profit mod for a 25 year old game) have apparently decided that collecting real art from the internet is bad but using AI to do it is somehow alright. That's the last I'll say here - sorry for derailing your thread. I just wanted to put in my two pennorth.

[ktchong]

51 minutes ago, Isewein said:

Sorry, Chaotic Good here. My concern wasn't about the letter but the spirit of the law, and I still think it's absurd that we as a community (because let's face it, that's the ultimate instance here: nobody is going to launch an IP lawsuit against an obscure, non-profit mod for a 25 year old game) have apparently decided that collecting real art from the internet is bad but using AI to do it is somehow alright. That's the last I'll say here - sorry for derailing your thread. I just wanted to put in my two pennorth.

Okay, let's leave copyright and laws out of this discussion.  Let's not even mention that there have been portrait mods there were removed or taken down due to copyright complaints.

Here is the one reason alone for why using AI is far superior than collecting "real" arts from the Internet:

The AI-generated portraits are simply of higher and more consistent quality than those that were collected from the Internet.   The AI artworks have higher and consistent quality, higher resolutions, AND MORE IMPORTANTLY, AI portraits can have a consistent uniform style to match the one used in the games.  You can NOT get that kind of quality and consistency from cobbling together a bunch of random artworks made by tens if not hundreds of different artists.

My game is heavily modded, but I opted NOT to install any of the portrait mods after testing them.  Why?  I certainly do NOT really care about copyrights in my own private game.  The reason why I chose NOT to use a portrait mod is because the quality of the portraits SUCK.  The qualities an styles of the portraits were all over the places.   Different portraits had different qualities, different resolutions, different styles.  Many of the portraits did NOT even fit well with the rest of the game NOR remotely match the portrait style in the game.  The styles were all over the places, and they break immersion.

I am just gonna be frank here: the two most popular portrait mods, Portraits Portraits Everywhere and  the Picture Standard, they both SUCKS.  The mods themselves work fine, but it was the portraits that suck because of the differed qualities, varied resolutions, AND different styles, which just killed those mods for me.

I do NOT care about if an art is "real", fake, AI, or whatever.  I care about QUALITY and CONSISTENCY of the portrait in my game.  I want all the portraits in my game to have a CONSISTENT, UNIFORM style.  And your so-called "real" arts can NOT give me the QUALITY and CONSISTENCY I want from a portrait mods. 

On the other hand, AI arts CAN produce hundreds of portraits that all have CONSISTENT and HIGH quality, resolution and style.  Good luck getting all that from "real" artists for a  portrait mod.

Edited March 16 by ktchong

[Shipwreck Jones]

@Isewein

No need to apologize.  Doesn't look like this project is going to get much attention anyway.

Honestly, I'm not particularly interested in having a debate -especially since I'm struggling to condense my beliefs concerning the issue down into a concise argument.  So instead, I'll pose some questions to you.

Carlos Mencia once received a ton of backlash for stealing a joke from Bill Cosby.  Yet, regular people repeat jokes created by comedians all the time, without permission or credit, and no one cares.  Why is one bad, and the other okay?

Why is it okay to create memes from other people’s “content”?  Why is it okay for people to repost those memes without permission or credit?  Why is it okay for people to create variations of the meme?  Can memes not be considered a form of (low) art?

When is it okay for creators to “take inspiration” from other creators?  If you believe it’s permissible when the resultant work is transformative, how transformative must it be before it’s okay?  Can generative A.I. tools achieve the same level of transformation?  And if so, why is okay to do it the old fashioned way, but not okay to do it with tools that make the process faster and easier?

Now, don’t get it twisted, I do think there are some serious ethical questions concerning generative A.I. (i.e. companies using it as a means to pay creators less than they deserve).  Moreover, I am greatly concerned about how it is becoming easier to spread more convincing misinformation (which can result in harms up to and including the loss of life).  So, I am not saying that it’s not important to consider the ethics of A.I., or the potential ramifications.  I do, however, disagree with the people who seem to think it’s inherently wrong in all cases, as well as with the A.I. bros who dismiss critics as modern day luddites.

@jastey

Yes, I was planning on replacing all the portraits with ones created using A.I.  Of course, many people seem to take issue with that as well.

 

[Sanderson]

I think having a replacement of PPE that is either completely copyright-free or creative commons with attribution is a worthy goal. I don't care about the use of AI. No artist will make 4000 consistent portraits. That debate belongs somewhere else I believe. Using official TSR or wotc art is a different debate but still should probably be off-limits within the scope of a new, guilt-free PPE.

I was actually using the same portrait generator last week that ktchong recommended, just doing portraits of people you meet on Candlekeep. It is hard to get variety in facial structures and demeanor without a lot of equivocating using terms like: fat, ugly, smiling, evil, sweaty, hungry, greedy, crazy, angry, etc or various nationalities. Including phrases like "in a dungeon" or "in a forest" is helpful, as well as props like "with a book" "holding a sword" or "with beer."  The term "crusader" good for getting knights, as long as you disclude Christian imagery. To get adult women, I generally had to put the term "mom" in there! It does it's best to not make NSFW portraits but I was making some ogres, and let's just say a lot of them were female

it's a simple generator and I haven't used more complex ones but it's a fine place to start. Some Photoshop would probably be required for turning people into ghosts etc.

 

[Incrementis]

You want the result of the images to be consistent.

Additionally, given the massive amount of portraits this mod requires, this is challenging.

My sympathy for that.

Although I can't help you with your attempt as I don't have time, I would still like to offer some suggestions that might help you. If you have already done or considered these things, then please don't bother with what I write.

1. I believe that some community members' suggestion to ask artists for help can still help speed up the process. As for consistency, a skilled artist could create portraits if given examples of consistency.
2. Asking on other platforms, such as the G3 Discord server or the Beamdog forum, could help reach and “recruit” more modders.
3. Since this mod requires a lot of modders due to the size of the images, try to offer anonymity to those who want anonymity, as AI-generated images could be viewed as something offensive in communities, which can prevent open help.
4. Note that the more people contribute to the mod, the harder it becomes to ensure that there are no copyright issues, since you won't know for sure whether the images were created the way you suggested.

Personally, I'd like to see this out of sheer curiosity and as learning material:

On 3/14/2024 at 11:55 PM, Shipwreck Jones said:

  I do highly recommend the use of a photo editor, as it will make life much easier.  However, A.I. handles all the heavy lifting, and I will probably make some video tutorials in order to make getting into it as easy as possible.

Cheers!

[Shipwreck Jones]

@Incrementis

Well, people don't exactly seem to be eager to help out on the project, so I'm probably going to be going at this alone.  I don't mind doing the work, as it's something I can do while watching youtube videos and relaxing at the end of the day.  I was just hoping that it could be finished in a more reasonable time period.  And, I already tried posting elsewhere.  I got zero replies on the beamdog forums, and the Baldur's Gate subreddit mods rejected my post because they have a blanket ban on A.I. generated content.  That one hurt, as the subreddit is probably the largest community of OG BG fans by a rather wide margin.

Anyway, I'm probably not going to make the videos since no one seems interested in helping.  But who knows, maybe putting some out there will help attract attention the project.  So, we'll see.  And, of course, if I do, I'll post a link here.

[Frenzgyn]

While AI generated content is not exactly my cup of tea, I appreciate your intentions.

Don't take it personally, Beamdog boards are scarcely active lately.