|
Where should I go to get a foundation in writing good Midjourney prompts? The GBS thread, despite its name, stopped posting its prompts a while ago, and it was never terribly disciplined about it. I've fiddled around with it enough to have some rough ideas, but I don't get anywhere near the results that other people are posting. It's only gotten worse recently, too: I'm having a devil of a time getting it to generate a person without a horribly mangled face, for some reason.
|
#
¿
Mar 29, 2023 20:11
|
|
|
|
| # ¿ May 22, 2024 11:04 |
|
|
World Famous W posted:just cut it to the last paragraph While I agree with the sentiment you're expressing, the purpose of the proposed rules isn't for the posters in this thread, it's to appease four extremely vocal posters (one of which is a mod) in the feedback thread. Which is futile, because nothing short of a complete ban will actually appease them, but that's where things stand.
|
|
#
¿
Jun 12, 2023 02:55
|
|
|
BrainDance posted:But, there are a handful of people who are not even making arguments, they're just very extreme and yelling stuff as fact. And that's the "four extremely vocal posters (one of which is a mod) in the feedback thread" Kestral is talking about, I think. This is correct, although I actually was referring to PurpleXVI. I don't know what side Nuns is on, I have them on ignore which somehow still works despite their being a mod - maybe it carries over if you add them to the list before they get their mod star. BrainDance posted:It's a trend though. Like I mentioned earlier we see it a lot in the DnD thread (and there I keep thinking, this is DnD, there are rules for how you argue so maybe more probations need to be handed out. Not cuz people disagree with me but because, come on, if I just come into a thread shouting but I wont actually put forth an argument that's not how it's supposed to work there.) And it's the same thing every time, and when pushed they end up coming to absurd arguments that take down everything else with it. TradGames and D&D will probably always need to have different moderation styles, just the nature of the beast. My suggestion when these things come up is to just make your point as best you can and then Cease To Engage. Nobody's obligated to have a fight on the internet, and it's important to remember that you won't change the minds of the people you're arguing with, only the people who are undecided and looking on to see who makes their point well. As a fellow D&D lurker I find it useful to remind myself that I can probably count on one hand the number of times I've seen someone actually change their mind because of an internet argument, and still have fingers to spare. For some actual on-topic posting, I'll just state for the record / Leperflesh's purposes that World Famous W sums up my position pretty well, for whatever that's worth: the thread for arguing over the ethics of AI (and god do I hate that term) already exists and everyone knows where to find it. If the TradGames policy is to have this be a quarantine thread*, then it will also have to be protected from the Loud Angry Posters who will do their level best to poo poo it up, and that's going to mean actual probes being handed out. * Leperflesh: this is not a policy that can endure in the long-term, just from a sheer enforceability point of view if nothing else, you've said you're down for revising things later on. Just be ready for this subject to get revisited later, because this is all going to happen again as this stuff increasingly permeates the culture.
|
|
#
¿
Jun 12, 2023 06:09
|
|
|
Leperflesh posted:I don't love the word "containment" and "quarantine" isn't much better. IMO the best thing is to just have topical threads - eg the warhammer 40k thread isn't and shouldn't be the 40k containment thread, but if people constantly want to talk about fascism vs. parody in warhammer 40k and that was an issue around TG I might workshop a rule insisting that go in a specific place, too, and probably that place would not be in the chat thread, or the 40k thread, where people should get to post about their war dollies with some semblance of peace and take it as a given that we're not just harboring a bunch of fascists on the basis that they still collect space marines. Today, the target of ire is AI stuff, I'm sure there will be new things to be mad about in the future and if we develop a tolerable model for dealing with that, that'd be cool. Agreed on all counts, except that containment/quarantine is what's being set up. Unless it's made mod-voice explicit that this thread (or its successor) isn't a containment thread, but is rather a designated discussion thread "with some semblance of peace" for a topic that isn't banned elsewhere in the forum as long as it doesn't take over, then we're going to get the same confrontations the feedback thread just went through, again and again. If the intent is to make this thread equivalent to other dedicated-topic threads, that's actually ideal IMO, as long as it's receiving the same protections - and it's a model that benefits the chat thread, too. I'd certainly support that approach.
|
|
#
¿
Jun 12, 2023 06:48
|
|
|
It's funny, I'd never considered using ChatGPT as an adjunct to Ironsworn oracles, since the oracles are so spooky-good. Using it for dialogue with NPCs, though, that would save me some time. This is the first I've heard of Character.ai, what's your experience with it been like?
|
|
#
¿
Jun 12, 2023 08:40
|
|
|
KwegiboHB posted:Working with Raenir Salazar now. I only work with Stable Diffusion and image generation but follow all of this closely so I know of and can provide links to a large number of different tools. Sadly I don't play Trad Games so I don't know what will fit best here. It does sound like fun to play it's just going to be quite awhile until I have enough free time again. Awesome, glad to see people taking an interest in the hobby, even if you're not able to play just yet! In answer to your question, there's hundreds of roleplaying games and dozens of wargames - I'll leave the wargaming side to folks who are better acquainted with it. For RPGs, the big dogs are Dungeons & Dragons and Pathfinder, but some other ones with relatively large player-bases include Vampire: the Masquerade, Call of Cthulhu, Blades in the Dark, Exalted, Stars Without Number, Apocalypse World, Shadowrun, Lancer, and the various Warhammer Fantasy and Warhammer 40K RPGs. For ways people play, there's two different environments people tend to play in: either physically around a table, or on a virtual tabletop that is essentially a shared whiteboard with game-specific functions like dice rollers. Around a table, people tend to just have the papers and book(s) associated with their role in the game, possibly minis, maybe some reference art. Depending on the table, they might have tablets or laptops and a group Discord to view "handouts" (reference material / inspirational art) from the person running the game. Whoever's running the game will usually have much, much more art to reference in addition to the above, and yeah, maps or depictions of landscapes are quite common. On a virtual tabletop, things tend to get more media-rich since it's much easier to drop images in front of people: more elaborate maps (there's a whole cottage industry of mapmaking for these games), tokens to replace physical minis (these are often an image placed inside a circular border), backgrounds or collages to set the mood or show a scene, that sort of thing. There's also a growing number of games which can be played solo. These seem to get a lot of mileage out of ChatGPT's offerings: "things that can pretend to be a person" and "things that can offer creative suggestions" and "things that can cite reference materials (ideally with a minimum of hallucination)." Inspirational or reference art. I can't tell you how valuable this is for me as a person with aphantasia, and for everyone who uses a virtual tabletop to play RPGs. We often have concepts that simply aren't represented in any art, anywhere, and the ability to create something that will get the gist across to fellow players is incredibly useful. This can be anything from weird creatures, to landscapes, to vehicles, to specific scenes like "the wreckage of a massive cylindrical freighter-starship, surrounded by its own debris and leaking antimatter." Character portraits. Oh my god we have so many characters running around. Every player has at least one, the person coordinating the game often has dozens, and many of them will change over time. We love character art, especially if it can capture the specific visual signatures of a beloved character. Icons and symbols. This is a bit niche, but I've often found it useful to be able to designate things on a map with icons and symbols rather than a specific image. Also, I'm pretty sure every fantasy game will have a scene in which some weird glyphs are important, and there's always a need for corporate logos or the calling-card symbol left behind by a gang of daring scoundrels. I'm sure other people will have more to say on this, I know my tired brain is missing stuff, but hopefully that's a good start! BrainDance posted:I was gonna type up a very quick, rough thing about local LLMs but my community has no electricity for maintenance tonight so, I guess, just leave some space and I'll get to it when I have power? Quite looking forward to this. This seems like an area with enormous potential for the way I run games, since I tend to generate enormous quantities of notes for the sheer pleasure of writing, and I'm curious what a local model could make of them.
|
|
#
¿
Jun 13, 2023 09:03
|
|
|
KwegiboHB posted:I have to run and do some grocery shopping now, I'll be back later today to work on this some more. If anyone else has any other ideas I'm all ears. This will help me work on the GBS thread as well Thank you for the effortpost! Glad it's useful for GBS too  Something I'm curious about, related to the following: KwegiboHB posted:Where I think things will truly start to shine is in some of the custom training options, such as making a specific character LoRA. You can take the custom artwork you have of a character and train a smaller model which you feed back into the generator which allows you to change specific things about the character, like outfits, action sequences, or battle wounds, etc. Megaman's Jockstrap posted:I think I might tackle the Stable Diffusion section a bit, because one of the biggest things about it is that you can host it yourself - which gives you a ton of options that using it as a hosted service just doesn't - and you can apply a conditioning interface (called ControlNet) to give you quite a bit of control over the output. Very much looking forward to learning more about this usage. Campaigns build up their own aesthetic over time, and the idea of having a tool that can grow with the campaign, essentially be bonsai'd into its aesthetic, would be incredibly useful. Something I keep coming back to about RPG use of these tools is how incredibly specific our image needs can be. So much of what we're trying to describe is, if not unique to our specific campaign, then so rare or specifically-detailed that it's impossible to find representations of it. This can actually be a problem in play! If I'm describing something, and the other players have a mental image that doesn't line up with mine, then they'll make decisions based on incorrect information, and that hampers play. If I spent forever describing a weird thing, that might get us closer to a shared concept, but I'm also monopolizing the spotlight and at a certain point people will just tune out. ()I actually have a story about this that just played out at my table, on the intersection of RPGs, commissioned art, and AI, that I'll post later when I have time to reflect on it a bit.) Which brings me to the actual question, which I think Megaman's Jockstrap may be intending to answer already: what's your current best option for getting an output with a high degree of specificity out of a weird description? Like, off the top of my head, if I wanted to generate an angel whose many wings were made of old televisions and their spear is a TV aerial, is that something achievable with the current tools, or should you be looking at combining multiple outputs in Photoshop and such?
|
|
#
¿
Jun 13, 2023 21:06
|
|
|
Humbug Scoolbus posted:For the maps, midjourney kicks out really good battlemaps actually since v4. You are doing the lord’s work in that class, Humbug, hell yeah On the Midjourney maps, any particular prompts necessary to get it to generate coherent battle maps, or is it as straightforward as “battle map for d&d 5th edition, lakeside village, densely forested” and go? And do you know how well it handles weirder map prompts, where you want something that isn’t a generic wilderness scene?
|
|
#
¿
Jul 4, 2023 20:55
|
|
|
Humbug Scoolbus posted:Trying out the Claude chatbot. works pretty well... Iiiinteresting - I wonder what the differences are on the backend between this and vanilla ChatGPT. I'm seeing commentary that Claude's real advantage over ChatGPT 4 is in maintaining lengthy context, like reading entire PDFs and answering questions about them. That's the sort of AI assistant I'd find useful for RPGs: "chatbot, please display every creature that can be conjured with Summon Nature's Ally which has a flying speed of at least 80." ***** Been sitting on some thoughts about AI art in the RPG space for a while now. I finally took the sage advice of KwegiboHB and Megaman's Jockstrap and got Automatic1111 running locally. After getting my bearings, I added ControlNet and some must-have extensions (After Detailer and Inpaint Anything are unbelievably good, jesus), grabbed some checkpoints - and even made my own with checkpoint merge, which was shockingly easy! - and after a couple nights of playing around with it and doing some very basic studying, I'm getting results that are just... Astonishing. The specificity that ControlNet and Inpaint Anything give you is remarkable, and solves my biggest problem with the output I was getting from Midjourney. Human artists can obviously outperform what I'm getting, but - well, storytime. One of the players in my long-running D&D 5e game loves the big bad, the corrupted nature-lich Harrower: conceptually, aesthetically, my portrayal of him, he just digs this weird lich A Lot. It's very flattering! Over the years he's done four or five HeroForge versions of Harrower, and cited this as part of his drive to continually up his game with HeroForge, to the point where he's now showing up on the weekly best-of model list for that site whenever he feels sufficiently inspired. None of them have been quite right though, because HeroForge's toolset is limited, and Harrower looks bizarre. Recently, this player got his first job and decided to celebrate by commissioning some art of Harrower, and asked me for as complete a description as I could provide. I said sure, since it was an interesting exercise, and generated a Google doc as if I was commissioning the art myself, complete with visual references, color palettes, clarifying commentary, etc. Weeks go by, there's back-and-forth with the artist, and finally we get the finished product. And it's just... Wrong. Wrong in basic ways that defy the very clear, several-times repeated and clarified and agreed-upon descriptions of the commission. And it's frustrating, because no matter how many times we try to get the artist to understand what we mean, they either can't comprehend it, or they just refuse to do it that way for reasons of their own. Anyone remember that article or tumblr post or something from years back, where it described someone's trouble getting a piece of art for either MtG or an RPG book? The one where the artist was supposed to make a druid in practical forest garb, and got Big Tiddy Elf GF over and over, no matter how many conversations they had with the artist? Same vibe, but less porny. At some point we both just gave up on getting it right and accepted a subpar result that didn't live up to expectations or, IMO, earn the commission, because my player didn't want to fight with this person and I didn't feel it was my place to do so. After it was over, I remember thinking, "Midjourney can't do Harrower justice either, the visual is too complex - but it would have gotten this result in much less than a month, and for free." Years ago, I backed an RPG at a level high enough to have a piece of character art included in the book, with a brief of: "Make a character who fits in this setting, here's some guidelines, follow those and we'll commission it." The process was almost identical, excepting only that I was less specific in my commission because I wanted to make sure the game's author got a book that looked the way they wanted it. I ended up communicating primarily with the artist, and there was no getting through to them. The piece that ended up in the book is nice, but I can't help feeling frustated by it whenever I come across it. Over the years, I've seen other friends commission art of their characters maybe nine or ten times. It's come out correctly perhaps three times. Every other result has been an expensive, time-consuming compromise that left them only vaguely satisfied. One of my other D&D players has had three separate pieces done of her long-running PC, and only one of them came out right, and it cost her something like $300 for that one piece. My sample size is small, but one in three is what we've got, and that's just not good enough. And that's where I'm at with AI art at this point. Human artists can outperform what my customized install of Stable Diffusion can do, no question. But if I'm going to get a result that is "good enough," I would rather get that "good enough" for free, in under an hour, and possibly in under 5 minutes. It's an unbeatable value proposition, and it's only going to get better. I'm very, very glad I'm getting familiar with this technology now, because it's going to be absolutely everywhere in five years, max.
|
|
#
¿
Jul 14, 2023 07:28
|
|
|
For the Stable Diffusion crowd, has anyone else run into a problem with After Detailer where it will try to turn people's hands into faces?  I had great luck with ADetailer on my first couple of serious "learn how to SD" projects, but now it's just going berserk and I can't figure out why.
|
|
#
¿
Jul 16, 2023 19:35
|
|
|
feverish and oversexed posted:No but can I see this please lol I deleted the original batch because they were unsalvageable, but here's another set showing the effect:   BE NOT AFRAID (of the facehands) What's really funny (and infuriating) is that you can see it happening in the step preview. It'll get through the initial generation, then After Details will take over and zoom in on each hand, and you get to watch it transform them into this:   It's actually much less nightmarish this time around for whatever reason: last night I was getting bulging fractal clusters of faces where the hands should be, instead of... Whatever the hell this is. Don't mind the very haunted faces, I haven't even started working on fixing that yet. This image is actually worth talking about to highlight some of the difficulties - and opportunities - in Stable Diffusion that I'm encountering as a new user. Hopefully it'll help others getting into this. My "Learn To Stable Diffusion" project is creating the PCs and major NPCs for a long-running campaign from some years back, since I have clear mental images and descriptions of them, and enough emotional attachment to them to want to get it right instead of just settling for "good enough." I did what I figured would be the easy ones first and learned a lot in the process, so now it's time to try a character who needs a specific pose, one that is iconic to the character and which no portrait of him would be complete without: glaring at the viewer while calling lightning from the sky like an Old Testament prophet. I'd used Controlnet's pose function a bit on the earlier ones, just to see what it was like, but Ihsan here was going to be my first attempt at getting exactly the pose I'm looking for. Now, for whatever reason, SD really did not want to cooperate for poor Ihsan. To pull off what I had in mind, I needed four elements: 1) A young man, ~18-19, with medium-tan skin, any expression between "grim and determined" and "outright hostile," and the build of an artisan: no crazy muscles, which a lot of SD models desperately want to put on men. 2) A savannah background, with lightning-riven stormclouds overhead, and bolts of lightning coming down from on high. 3) White robes (or white with a bit of gold) in a reasonably practical style for traveling long distances by foot. 4) A pose with one hand outstretched. Exactly how the hand is positioned doesn't matter much: a clenched fist, a palm upturned and either flat or with fingers curled, anything like that. And folks, let me tell you, aside from the savannah and the lightning, SD wanted none of it. Fought me every step of the way. It was especially cantanerkous about skin tone (it super, super wanted him to be either pasty caucasian or black), and on clothes that didn't show rows of chiseled abs and were shades of white. Fun fact, in several models I tried, telling it to make him wear "all white clothes" or "pure white clothes" turned Ihsan into a white girl I tried five different models, using the X/Y/Z Plot script to compare them side-by-side, in a matrix with different sampling methods. Often I was able to lock in one major element, but at the expense of another. The more reading I do on this, the more it confirms what I'd noticed in my own experimentations: a lot of these models are overwhelmingly trained on photos/paintings/drawings of white and asian women, and if you want something other than that, you may need to either find a model that specifically gets around it, or add LORAs. ControlNet consistently gave me a workable pose based on this:  ... but I found that the OpenPose preprocessor would often pick up the trousers, giving the character modern clothes instead of robes. After much fiddling, I managed to produce this:  ... which would have been nearly goddamn perfect, except I'd screwed up a ControlNet setting and accidentally turned the Resize Mode to "Crop and Resize" instead of Just Resize, and it cut off the top of his head!  Quite a few more rolls of the dice and tweaking the prompt later and I had this:  And that I can work with! The flaming horizon is actually what sold me on it, because this character's defining moment from the first arc of the campaign takes place while the plains are burning in the background behind him. That showing up was a total accident and I knew I wasn't going to get that effect again without a lot more work, and while the face and outstretched hand (and to a lesser extent the feet) are messed up, they're within workable range. I threw it over to IMG2IMG and started employing the various techniques I've seen people discuss for After Detailing hands and faces. People regard AD as nigh on miraculous, and you can watch it perform some incredible feats: https://www.youtube.com/watch?v=ZNcz4k5JCCo (Skip to 7 minutes) I'm fond of that one because the method he uses makes it impossible to stack the cards in his favor to make it look easy for youtube: he randomly generates his prompt and throws it straight into SD, then does the whole correction process with After Detailer in one take. The results are amazing, and in my limited experience with it I was seriously impressed. Unfortunately, now it's going horribly wrong and I have no idea why! None of the techniques I've used up 'til now have been able to salvage that outstretched hand, which either becomes a gnarled lump of flesh and knuckles, or turns into faces. It's very haunted and I'm nearly at my wit's end. Ah well, time for more research. Kestral fucked around with this message at 06:06 on Jul 17, 2023 |
|
#
¿
Jul 17, 2023 06:04
|
|
|
KwegiboHB posted:Thanks for the write-up, that'll be handy. It's late, I'll watch the video tomorrow and try to wrap my head around what could be going wrong, offhand I can't quite put my finger on it. Face it, it's hard enough to get ahead of these problems. I want to make sure I can help and not just give out second-hand information. Any insight would be appreciated, but no worries if it ends up being a project! I don't want to assign anyone here homework I'm not actually salty about the experience, honestly, it's just part of learning a new set of tools, and I felt like documenting it here so that other folks would know how that learning process can go - it can be frustrating to try to get into something complex or difficult and only see other people posting their Ws, IMO. Megazver posted:This is really cool. I haven't paid attention to SD for a few months and turns out there's an entirely new wondertool available. It's amazing, isn't it? Literally every time I do more research I stumble across at least one new thing that meaningfully adds to its capabilities. Like, Segment Anything! Holy poo poo, you can just... choose areas for inpainting, already neatly divided up for you like it was a paint-by-numbers. This was enormously helpful in one of my earlier projects, where I'd managed to get a fairly complex composition down (a man riding an unusually large Clydesdale-esque horse, turned toward and looking down at the camera), but it kept generating the character as a pinup calendar version of himself, shirtless and abs-out. I could have just inpainted it, but it had also generated this neat necklace that hung down to his chest, and my hands aren't steady enough to preserve that detail while inpainting. But Segment Anything immediately knew what was chest, necklace, horse, etc., blocked it out in colored regions, and let me select just the exposed skin of the chest to iterate on that until he finally put some drat clothes and armor on.
|
|
#
¿
Jul 17, 2023 21:23
|
|
|
I recall reading somewhere that SDXL was also going to substantially increase the system requirements for generating images, to something in the 30XX range. Hopefully that doesn’t prove to be the case, but it wouldn’t surprise me.
|
|
#
¿
Jul 20, 2023 23:35
|
|
|
BrainDance posted:It is definitely slower though, substantially slower. Yeeeeep, this is what I was afraid of. Looks like those of us on pre-30XX cards are going to be sticking with 1.5 for a while.
|
|
#
¿
Jul 21, 2023 04:41
|
|
|
That is straight-up perfect Dark Sun art. Could you post the settings and prompt for that? As soon as the D&D LoRAs / models are out for SDXL, I’m gonna use it to make so many goddamn thri-kreen.
|
|
#
¿
Jul 29, 2023 22:15
|
|
|
Thanks for the rundown, Mega! Speaking of which, turns out ComfyUI handles SDXL enormously better than A1111, to the point where I can run an image in under a minute instead of Considerably More Than That, so I gave it a shot. The interface is clumsy, making changes to your settings is slow and awkward compared to A1111, and my results are nowhere near as good as Megaman's - that VAE misconfiguration was some secret sauce - but it does run, so there's hope for those of us on lower-end video cards. The downside is that ComfyUI appears to load the entire goddamn base model and refiner into RAM to accomplish this feat, so if you don't have at least 16GB, you're out of luck with that method. For folks wondering about battlemaps: Yes. SDXL does battlemaps, although not quite as well as I'd wish:   These are not pretty battlemaps, but they are literally the first two I generated, with an extremely basic prompt of "a battlemap of a field of crystals and small pools" for the first one, and, for the second: "a battlemap of a field of crystals and small pools, in the style of (Gerald Brom:1.1), on old parchment paper." If I were running a session on Roll20 and I needed a map right now for an encounter I hadn't planned on, I'd be quite happy with the second one. Now, when I tried to make them actually look good, that's when it started taking a while, and in fact I still haven't managed one that actually does what I want it to. Adding aesthetic keys really throws it for a loop. It looooves generating cartographic maps and isometric maps though, and when you insist that it not do those things it gets rather cross. Which is really Stable Diffusion in a nutshell: it's very, very easy to get either mediocre results that are in the ballpark of what you're looking for, or great results when you have nothing in mind but a vague concept and you let the machine dream wildly, and it's quite time-consuming to execute on a specific vision. There's a lot of potential here though, looking forward to Megaman's next writeup. Edit: Not sure I agree with the "you don't need long prompts" idea. Long negative prompts, yeah, not as much, but if you've got specific concepts to execute, it definitely still needs you to know how to render instructions to it in a way it understands. Kestral fucked around with this message at 05:51 on Jul 30, 2023 |
|
#
¿
Jul 30, 2023 05:19
|
|
|
Anyone here figured out how SDXL has changed the way prompt weighting works? The old (term:1.5) or (term:0.7) etc. technique is... Finicky, now, and I strongly suspect I'm missing something here.
|
|
#
¿
Jul 31, 2023 04:33
|
|
|
The Eyes Have It posted:Oh hey that's pretty neat. Oh my god I feel this so hard. My first real SDXL project involves a character “wielding a sword” and out of the hundreds of images generated, not a single one was anywhere close to correct. Swords emerge from your forearm, right? Or float next to you, but in a derpy way and not in a cool Alucard way, yeah, that’s the ticket! Someone should shove a bunch of German and Italian fighting manuals into a LoRA and see what happens.
|
|
#
¿
Aug 1, 2023 00:25
|
|
|
|
| # ¿ May 22, 2024 11:04 |
|
|
Has anyone found especially good checkpoints for landscapes in an illustrated/painted style? The "discussion" in the Industry thread gave me an idea for how to solve an especially tricky conversion of an ink sketch into a photo/painting using Regional Prompting, but I'm finding that my checkpoint collection has a real weak spot in the landscapes department. I did attempt a photorealistic style with Photon, which looks gorgeous when it works, but the ControlNet references make it choke up badly. Still, highly recommended for photorealistic landscapes when you're not trying to do something silly like this. Also stumbled across this: https://www.youtube.com/watch?v=ckyhwy0Aui4 Lovely comparison of a ton of artists' styles as recognized by SDXL. For someone like me who has an idea of what they want something to look like, but often struggles with describing it in a way the model will understand, this is a phenomenal reference.
|
|
#
¿
Aug 6, 2023 04:07
|
|