This is part of a series of markets about things I could do in 2024. See: /Mira/which-of-mira-s-cool-ideas-will-mir
Summary
Right now the killer app that AI image models are missing is better control over the outputs.
AI models won't replace all comic artists, illustrators, animators, etc. until you can show it a character and a pose and a background, and get that character in that pose in that room. Or draw a 5-minute sketch, and turn that sketch into the $2000 version of that 5-minute sketch.
Right now you can't use it for illustrations or comics because the outputs, while awesome, are not consistent with your vision or with each other.
Some AI art tools let you inpaint/reroll parts of the image you dislike, right? That sounds like it would already be pretty useful but I haven't tried working with any of those tools yet.Examples of techniques that could improve prompt control might be:
Finetuning Llama 2 or Mistral to parse the prompts for the Stable Diffusion model
Adding segmentation models, GPT-4V, and inpainting the model
Adding specialized text-generation models
If I spend time studying how to improve AI art models and some money training them, how much money will the project net?
Market Mechanics
Trigger condition: I train a prototype model or model workflow and indicate that it seems promising enough to expand on.
Resolves NA if the trigger condition is not met. Otherwise, resolves to the logarithmic interpolation between nearest bounding answers(or to $1 if I start this but don't make any positive profit for the entire year). Units are "profit in USD". I'll add new options at multiples of 10 as needed before resolving.
Example: I make $500 in annual profit. The nearest options are $100 and $1000. Then this market resolves (ln(500) - ln(100)) / (ln(1000) - ln(100)) = 70% $1000 and 30% 100.
"Profit" is net expenses. However, money spent training a model is not necessarily counted as money lost, if the model itself seems valuable. I would value the model at $0 if nobody uses it, but otherwise I will probably depreciate it over 3 years so only 33% of its cost would count against it. If I purchase equipment like GPUs, those would be depreciated over 5 years, so only 20% of their cost would count against them.
Mira will not trade on this market(or will market sell if I accidentally do).