Using Anima as base model, with controlUI, trying loras, IMG2IMG with flux.
It’s very clear that the generation is limited to a dataset.
Even more clear when you use SDXL that basically works by tags.
But even with models which process natural languages you see the limitations imposed by the training data.
If you want to make a generic image that’s been done 10000 times it’s very easy. But if you want to do something very specific that the thing have not been trained to do… you are not going to make it. Unless you literally paint want you want and train a lora with it, which kind of defeats the purpose.
Hold on a second. If what you’re saying is true, which you have no way of proving to me that it is, I’ll take it at face value nevertheless. But if that’s the case, then you clearly do have some understanding of how these technologies work.
You’re telling me you’re upset that an LLM, which has been trained on pre-existing data, can’t produce something completely novel without referencing some kind of source?
That’s simply how the technology functions. What a lot of people actually do is come up with their own original ideas and put them on paper or into a digital drawing program. Then they take that original picture or concept and put it into AI tools, where the AI can manipulate, refine, or transform that original work and save them significant amounts of time on tasks that artists traditionally had to do manually.
So your gripe with AI is essentially describing a limitation of the technology. That’s not really an argument against its use. It’s a limitation that can be worked around depending on how you use the technology.
As said I have use it. Both for image generation and llm for text generation. But knowing what they can and cannot do.
I mostly use it for roleplay. And generating images of places and characters on the go is fun and easy. But it’s not a tool that allow for a very high degree of creativity. Neither have the ability to put the “image” that someone has in their head onto canvas.
I have tried.
I spent several days on a comfyui workflow.
Using Anima as base model, with controlUI, trying loras, IMG2IMG with flux.
It’s very clear that the generation is limited to a dataset.
Even more clear when you use SDXL that basically works by tags.
But even with models which process natural languages you see the limitations imposed by the training data.
If you want to make a generic image that’s been done 10000 times it’s very easy. But if you want to do something very specific that the thing have not been trained to do… you are not going to make it. Unless you literally paint want you want and train a lora with it, which kind of defeats the purpose.
Hold on a second. If what you’re saying is true, which you have no way of proving to me that it is, I’ll take it at face value nevertheless. But if that’s the case, then you clearly do have some understanding of how these technologies work.
You’re telling me you’re upset that an LLM, which has been trained on pre-existing data, can’t produce something completely novel without referencing some kind of source?
That’s simply how the technology functions. What a lot of people actually do is come up with their own original ideas and put them on paper or into a digital drawing program. Then they take that original picture or concept and put it into AI tools, where the AI can manipulate, refine, or transform that original work and save them significant amounts of time on tasks that artists traditionally had to do manually.
So your gripe with AI is essentially describing a limitation of the technology. That’s not really an argument against its use. It’s a limitation that can be worked around depending on how you use the technology.
I’m not upset.
As said I have use it. Both for image generation and llm for text generation. But knowing what they can and cannot do.
I mostly use it for roleplay. And generating images of places and characters on the go is fun and easy. But it’s not a tool that allow for a very high degree of creativity. Neither have the ability to put the “image” that someone has in their head onto canvas.