There are many "80s anime" and "vaporwave" style loRAs, but I wanted one that specifically does this. It's trained on 30 screenshots (which is every major cut) of Kouji Morimoto's opening sequence for the 1986 movie Dirty Pair: Project Eden and in case you're not familiar with Safari Eyes, <a target="_blank" rel="ugc" href="https://www.youtube.com/watch?v=91oqZjZTw7Q">allow me to show you the coolest thing you're going to see today</a>. I hope it can replicate even a tiny bit of that style.It works best with more abstract imagery, such as:<ul><li>smoke/clouds/fluids</li><li>silhouettes</li><li>specific bold colors</li><li>patterns</li><li>close-ups / detail shots</li><li>or just letting it do its own thing (but see below)</li></ul>Known issues:<ul><li>it likes to add black borders, which is a dataset defect (easy enough to crop out if you care)</li><li>you need to prompt at least something it can grab on to, just the trigger is not enough to make a big difference - when in doubt use either "1980s (style)" or "flat color" or both.</li></ul>

『SAFARI EYES』

Use more strength (up to 1.5) for distilled models. Omit the trigger word if the model wants to spell it out, it's optional.

Files with SHA256: c976c70ffdefd5a9787df5c0485e5743eb5f8bf19433c135724acb6fb526e008