It has been a while, maybe 3 years, since the last time I published a question of the week.
Anyways, besides what I said, what AI program(s) do you use to create characters and/or builds for Wolfery?
For me, while I rely on adopting characters through YCH commishies, for my build, I initially used OpenAIās DALL-E before switching to MidJourney upon release for many of the interior and scenery images. It is easy, cost-effective, and honestly the very first time I have ever gotten my hands on AI stuff, despite some reliability issues.
Iām going to assume this isnāt ragebait. 
I donāt use any AI programs to make my characters at all! In fact, I avoid using them even for NPCs in D&D games, or other similar one-offs. Character design can be tough, but having a connection to a character is important if you donāt want them to feel hollow. Building not just a character I like, but one that I understand how they were made, makes the character more special to me. I highly recommend, particularly if people are new to role play or character design, that you make your characters by hand. Write stories about them. Doodle them even if you canāt draw. Imagine what theyād do in a show or movie youāre watching. Introduce your friends to them and see what they think. Or just dive into role play and let them evolve as you interact with others and explore what and who they are. Keep notes! If you arenāt interested enough in your character to dote on them, how do you expect to embody them, or interest anyone else in them? If thereās absolutely one thing that must be bespoke to exist, itās personality.
They say a pictureās worth a thousand words. But I assure you, your character can be worth a LOT more than a thousand words!
4 Likes
Ahah! I can assure you 100% it isnāt ragebait. Iām just curious because Iām looking to switch to a new software after Sora belly up.
I find it goes both ways. Especially for someone who canāt draw more than stick figures, itās a helpful tool for early on in a characterās life. Being able to rapidly iterate the characterās look and ātry it onā, so to speak, really helps me refine who the character is and how they present themselves.
Obviously, nothing compares to having an actual artist breathe life into a character and get all those tiny details you really want! But getting in on commissions can be difficult, are an expense that not everyone can afford, and constant revisions if you change your mind about things is frustrating for both commissioner and artist. Better to hash those out first, then connect with a real artist.
To answer the original question, Iāve been using a local-run copy of Stable Diffusion, using the automatic1111 WebUI interface. Itās been abandoned for newer forks recently, but Iām still using it because I have it set up and works well enough.
Iāve looked into switching to ForgeNEO for newer features, but I just havenāt felt the need. Also considered learning ComfyUI butā¦ see previous sentence.
All those run locally and use your computerās GPU. I have an RTX 4070 Ti SUPER which ploughs through whatever I throw at it very quickly. If you donāt have a dedicated GPU or an older card, it might not be as effective an option.
1 Like
I only have the integrated AMD card. I do plan to switch to an RTX laptop with an AMD processor and 16GB upgradable RAM. I feel like thatās a perfect combo.
Oh, and welcome to the forums! Itās a pleasure to have you here.
Thank you!
What you have should still work, it will just be slower (and might slow down everything else happening on the computer). Iām not an expert with it and it does require a bit of setup (like installing Python), but if you do give it a try and run into issues, let me know and Iāll do what I can!
No problem! Even though my build still uses Midjourney, I do plan to switch to OpenAIās GPT Images to keep things up-to-date, or use an older ai model that works with my laptop.
I recently played a short scene in a store created via AI by another player for a topic-specific RP that didnāt already have a setting. Perhaps it was lacking a good prompt, but I noticed it sort of cheating by describing some things in generic terms (think āall you could need for golfingā rather than talking about putters, irons and wedges). And while it did use a specific name for the store, which sounded good, it turned out that was also used by a topic-related chat group - which in all likelihood was where it came from.
So Iād second the suggestion to do as much legwork as possible yourself, to create a proper foundation for your build. In most cases this will involve writing and editing, but in some cases it makes sense to start looking for a picture first and then use it as the seed for a description. if drawing isnāt a realistic option, then if you can get out there and take a photo, try that - or find one or the other under a free license on Wikimedia Commons or Flickr, and use that with appropriate credit. Or maybe someone has done a piece of art thatās perfect, but itās not under a free license - try to reach out and ask if you can use it, maybe throw them a Ko-Fi for the trouble if they reply positively.
Failing all that, sure, go with an AI toolchain you run locally, if possible, so at least you learn something - if the ones mentioned above donāt work for you I think the big three GPU manufacturers all have packages for it now for their hardware. At least if youāve done the above you might have a better idea of what you want to get out of it.
1 Like
Something has been bothering me for a bit. While Iām looking to buy a new laptop or a new e-bike to support my local delivery service, AMD offers an AI variant of the Ryzen series. I want to go for the non-AI series as I wanted raw power, but does AI labeling mean itās designed for AI stuff?.
Pretty sure everything is sold as āRyzen AIā since about mid-2024. Itās not like they put extra non-AI compute units in the area that would have had AI. Rather they add a few AI-oriented special compute functions to their existing APUs and call it a day.
In short, for āRyzen AIā think āZen 5 on mobileā, only slightly cut down because a full AVX-512 pathway would be too power-hungry, so they use a physical 256-bit data path like Zen 4.
1 Like
Okay. I donāt want to wade way too deep into it, because thereās plenty of guides on the internet,
But some notes from the AI about AMD and image gen:
Prerequisites for AMD
- ROCm Installation: Ensure you have the ROCm stack installed on your Linux system.
- PyTorch ROCm Version: Standard PyTorch wheels from PyPI are often CUDA-only. Install the ROCm-specific version from the PyTorch Get Started page.
- Supported Hardware: While many Radeon GPUs work, official support is prioritized for RDNA 2/3 (e.g., RX 6000/7000 series) and Instinct (MI series) accelerators. [1, 2, 3, 4, 5]
What is important to understand here is that nVidia uses a protocol called CUDA and AMD uses one called ROCm. CUDA is the lingua franca of the AI world. ROCm is still playing catchup, and honestly isnāt as well supported.
1 Like
Hmm, wellā¦ I donāt want any AI stuff shoved into my laptop unless the task is offshored to a random data center, so I think I know what I want.
Iād experimented with Stable Diffusion using Automatic1111 and have always found the end results to be distracting were I to look at them the wrong way, even after spending much time refining.
And refining them is such a time-consuming process, I eventually figured Iād rather work an extra day or two and use that money to pay somebody to do it for me.
I suppose free services like Perchance donāt have to be locally run, but you only get to pick a couple passable images out of several hundred generated, and they are indeed merely passable.
In my opinion, AI is adequate for throwaway characters. A good number of artists on FA will do you decent headshots for 10 - 20$.
Itās worth keeping in mind, your choice of reference affects who will be interested in your character. Though I personally donāt like AI, I canāt deny using a generated reference ālowers the entry barrierā and more people reach out! Even if less of them follow roleplay ethics.
1 Like