A new extension for stable diffusion is now in town: control net promises a lot and delivers. Now we have the power to perfectly match both composition and visual elements by feeding in sketches and references during generation. This is just the last of many changes over the past few months:
- SD 2 was released, albeit with mild interest in the community due to its perceivable nerfing.
- A variety of alt models providing good quality results in specialized domains arose.
- Inpainting, pix2pix, the list goes on and on.
- We got LORA and Textual Embeddings, lightweight training processes to integrate new concepts.
- I got a proper tesla p40. They got dirt cheap on ebay and with 24GB of ram they provide support for virtually all workloads.
All this together led to more tools and possible applications. To be continued