ken1171

~SD3 and Llama3 Just Released~ After a long wait, the much-anticipated Stable Diffusion 3.0 was finally released, but it wasn’t exactly what people expected. This time, Stability AI is keeping this new version behind a paywall. They released it from behind an API, which requires an account to reveal a key, and credits you have to buy to use it. However, they say a free version should be available in about 2 weeks, so we will have to wait and see. But what is SD3 all about? Like SDXL, it’s a whole new architecture that brings new features, improved results, and it comes in 2 versions: Standard and Turbo. For those following SDXL, the “Turbo” version is much faster, like almost real-time results, at the cost of lower quality images. I assume SD3 Turbo will be in these lines, though I haven’t tested anything yet. I will wait for the locally installed version to try it out - but keep in mind the initial models have not been finetuned, so the real goodies should come from the community,

~AI or not AI?~ Artificial Intelligence (AI) is here, and it’s starting to make an impact on people’s lives. After the initial legal dispute concerning how its models were trained, it seems like the subject has been settled in the justice courts, and now companies started releasing their AI software, tools, and plugins. It’s on! Even though Stable Diffusion (SD) remains the only free and open-source AI image generator with all bells and whistles unlocked, other companies are now pushing their own versions that do exactly the same things, often the same way. The big difference is that what they push is paid, which makes me wonder why would people pay for it? They can do the exact same things with SD for free, so why pay? After some reflection, I believe the main reasons are: The regular Joe may not know how to install SD locally in their computer. It’s a 3 parts process, and even though we now have 1-click installers, I have seen people failing to get it running for one reason

~Learn LoRA Training!~ Did you like the AI character images I have been showcasing in my gallery since September 2022? How do I do that? In some cases, when the character is famous and popular, a checkpoint model itself may be able to do the job on its own. However, a common case is when we need a specialized model to do it that was trained specifically on the character I want to depict. AI image generation has a special kind of model just for that - the LoRAs (Low-Ranking Adaptations)! Want to learn how to train your own LoRAs? Check my live event HERE. You get a 10% early bird discount for the 2 sessions event if you sign up by Sunday (Feb 25th) using the coupon code “lora10”. There is also a FREE ticket to assist the first 30 minutes. 🙂 In general, LoRAs are add-on models that can be plugged into a checkpoint model to give it a “specialization” it couldn’t handle on its own. In a way, we could say they are “plug-ins” for checkpoints. That’s one way to put it. The most common