Hi~!Here I am again. I plan to slow down with merging for a while — I want to come back to it a bit later, after doing some research on how it actually works and experimenting more with small tweaks to better understand the process. And this merge is a great place to put a comma (or a bookmark), as it's quite different from everything I've done before.In its nature, it's pretty much the same — just a simple roughly 50-50 merge of two models, without any LoRAs and with CLIP layers left unmerged. But the result turned out to be surprisingly good. It's half @EarthboundAI's Icarus and half of my Black_Bloom. Yet, somehow, it became almost like Icarus — clean, detailed, and beautiful — but with a subtle touch from my merge that makes the results a bit more colorful (and, surprisingly, I think a bit crisper too. Don't ask me how; I have no idea XD).I didn't experience any issues during generation — everything went smoothly. As always, here's a quick rundown of my settings:First Pass (Base Generation):<ul><li>CFG Scale: 5.</li><li>Sampler/Scheduler: I use <code>res_3m</code> and <code>beta57</code>, but standard <code>euler</code> with <code>normal</code> or <code>beta</code> should also work fine if you prefer simplicity.</li></ul>Upscaling (I use Latent Iterative):<ul><li>CFG Scale: 4.</li><li>Sampler/Scheduler: <code>dpmpp_2m</code> with <code>beta</code> worked well for me. I haven't tested other combinations.</li></ul>The outputs look pretty clean to my eye. They're finely detailed, with sharp lines... and yes, broken fingers are still there — can't escape that. Maybe 2026 will finally teach us how to avoid them, lol. But honestly, I get decent hands pretty often, so I can't complain. The model is still quite colorful and pleasantly wild. I tested it with my old, well-known prompts and sometimes got very different results, which I absolutely loved.Please, have fun with this little creation. And, as always, have a good day :3

Files with SHA256: 17089531d5095fba918a132bedc1320c0a28d0caa39b837609b83cd1fef2b40f