Navigating the AI Frontier: My Journey with Stable Diffusion and Beyond
Questions from Myself to Myself Based on SD3:
Q: How has your experience been trying to work with Stable Diffusion models like SDXL and SD 1.5 on your current setup?
A: Oh boy, it's been a ride! My trusty 2019 iMac does its best, but it's not exactly AI's best friend, especially with its nerfed AMD card. Trying ComfyUI to run SD 1.5 even made my computer throw a tantrum and restart itself—it's a comedy of errors sometimes. This is why I have to rent servers from Vast AI and Runpod. These platforms have become my go-to lifelines, allowing me to push the boundaries of what I can create without having to overhaul my entire setup.
Q: Given the challenges, are you considering upgrading or trying out SD3 Medium?
A: As much as I'd love to dive into SD3, my current setup's like a vintage car trying to run Formula 1. SD3 demands more than my iMac can give, especially with its commercial license requirements and hefty hardware needs. However, I'm not entirely ruling out an upgrade in the future—perhaps when the cost of GPU servers becomes more manageable or when I can justify the investment in light of my projects and income.
Q: Could you share your thoughts on the direction of AI models like SD3, based on your previous experiences and current limitations?
A: Back in April, when I first peeked at SD3, I had my reservations. It feels like SDAI's aiming for the corporate crowd, which is cool, but I'm still adjusting my expectations. Plus, the tech still needs some serious TLC—I've heard tales of LoRAs and ComfyUI mishaps that make me cackle, fall over laughing, and cringe. It’s clear that while AI technology is advancing rapidly, it’s also hitting some hilarious and frustrating snags along the way.
Q: How does your work with AI tie into your everyday life and challenges?
A: Balancing my AI adventures with everyday life isn't easy. I rely on the Civitai creator program to cover my bills and give me a bit of spending money, especially since traditional work's been tough with my CPTSD. It's a juggling act, for sure. AI has provided a creative and financial outlet, giving me a sense of purpose and a way to contribute despite the challenges I face with mental health.
Q: People often inquire about your "crew" or team. How do you respond to those questions?
A: Oh, that's a classic! When folks ask about my "crew," I have to clarify that I'm actually part of a DID system—each number they hear is just another alter. It's a quirky side of my life that sometimes throws people off. And no, having more alters doesn't mean faster AI training—it's a unique journey for each of us, including managing mental health along the way. Each alter brings a different perspective and set of skills, which can be both a blessing and a challenge in our AI endeavors.
Q: Looking ahead, what's your plan with AI development, given these hurdles?
A: Right now, I'm taking it day by day. Exploring PixArt may be an option for me, but the training aspect is expensive. I'm dabbling in SDXL's quirks and further screaming at the clouds when it duplicates noses and buttcheeks. Maybe, just maybe, renting a GPU server down the line will allow me to see what more I can do to learn about SD3. But for now, the cost of a creator's license and those image generation limits? They're a bit steep for my wallet and my understanding. The creators' license fee in USD and the 6k image per month generation limit are concerns that many of us in the community share.
Q: What are your thoughts on the community at large's response to SD3?
A: It depends. There are some communities seeking to undermine and cause rife due to unforeseen disagreements, and there are many that maybe still don't agree with others—but at least have the rights to go, "It's me, that's my feeling, you're valid in your thoughts." I wrote a whole article about this both on Civitai and this website. It's a mixed bag, but the ability to have open discussions, even when there's disagreement, is vital for progress.
Q: Laying on grass.
A: Do you want me to get censored by SD3 Medium? I like my neck, thank you very much. (This refers to the fact that Stable Diffusion 3 (2b) Medium has a censorship issue where if you put anything laying on the grass, the image errors out like a bad birth defect crossed over with things you've seen in horror movies.)
Q: Towards the future: You're in the creator program, are you planning on finetuning anything that's not SD3?
A: I don't really have the funds myself. I'm aware there's talk of crowdfunding options—but it would depend on how much a community could pull together. I don't have the stamina to write 3 million captions even if they were auto-tagged. Not only that, but I'm trying to get my mental health in order and just got an offer for a therapy appointment. Balancing the technical demands of AI with personal well-being is crucial, and while I'm open to new projects, they need to be manageable within my current capacity.
Q: Any final thoughts on the future of AI models like SD3 in your work?
A: It's a wild ride, no doubt. We're still figuring out SDXL, debating the merits of SD 2 versus SD 1.5—there's never a dull moment. The evolution of AI models presents both exciting opportunities and significant challenges. So, here's to the journey ahead, learning the ropes, and maybe one day, leveling up my setup to tackle the big leagues. The road is long and winding, but the potential rewards make every twist and turn worth it.
Comments