I’m sorry, but what is ill informed or opinion about it? Fact is it can do things no other image generator can do, open source or not. It can also effortlessly do things that would require a lot of tinkering with controlnet in comfyui, or even making custom lora’s. It’s a multimodal model that can do image and text both input and output, and does it well. All other useful image generators are diffusion based, which doesn’t read a prompt in the same way, and is more about weighting patterns based on keywords rather than any real understanding of the prompt. That’s why they’re struggling with relatively simple things like “a full glass of wine” or “a horse riding an astronaut on the moon”. If I’m wrong about this, please prove me wrong. Nothing would make me happier than finding an open source model that can do what openai’s new image model can do, really. I already run llama.cpp servers and comfyui locally, I have my own AI server in the basement with graphics cards. Please, please prove me wrong here.
I love open models, and been running them locally since first llama model, but that doesn’t mean I willfully ignore and pretend what claude and openai and google develops doesn’t exist. Rather I want awareness about it, that it does exist, and I want an open source version of it.
theterrasque@infosec.pub 5 days ago
I’m sorry, but what is ill informed or opinion about it? Fact is it can do things no other image generator can do, open source or not. It can also effortlessly do things that would require a lot of tinkering with controlnet in comfyui, or even making custom lora’s. It’s a multimodal model that can do image and text both input and output, and does it well. All other useful image generators are diffusion based, which doesn’t read a prompt in the same way, and is more about weighting patterns based on keywords rather than any real understanding of the prompt. That’s why they’re struggling with relatively simple things like “a full glass of wine” or “a horse riding an astronaut on the moon”. If I’m wrong about this, please prove me wrong. Nothing would make me happier than finding an open source model that can do what openai’s new image model can do, really. I already run llama.cpp servers and comfyui locally, I have my own AI server in the basement with graphics cards. Please, please prove me wrong here.
I love open models, and been running them locally since first llama model, but that doesn’t mean I willfully ignore and pretend what claude and openai and google develops doesn’t exist. Rather I want awareness about it, that it does exist, and I want an open source version of it.