ViviX12
Engaged Member
- Jan 5, 2019
- 3,238
- 5,075
- 698
Also vast majority is ugly and uninspired. Lack of any consistency is easy to point out, but also because the people who reach for algorithm generated images (there is no actual artifical intelligence involved) are not the type who would correct those mistakes.I'd go with plus solely for for the combat mechanics, although a future update that might change.
Half of these are fixed by running a model locally on a mid-teir gaming rig and ethically sourcing input (read: getting permission to use input data). Meanwhile some my actual concerns are:
- Images -- The tendency for ai models to blur images conflicts with the game's pixel art style.
- Getting the same result even with the same prompt is highly unlikely. This can cause serious issues such as-
- Animation -- Triggering motion sickness in more sensitive people like myself.
- Animation -- AI animation is primitive and has not improved much since it's conception. Yes occasionally a decent one comes out, but these ones have limited movement or the previous issue.
- Ai is notoriously hard to correct, meaning a good chunk of time would be spent micro managing or completely redoing output.
- This also makes hallucinations still relevant as models tend to double down on these false details.
Either way the solution wouldn't work at all because when you need to account for many variables, having a separate asset for every combination is the worst way to go about it, as it both takes tons of time to create (regardless of method) and makes the total weight of assets explode.
The generated images wouldn't give you a solution that could use layers as should be done (and already is in case of DoL).