works well with natural language with some tags at the end. use updated v1.5 vae (or whatever vae you want, i'm not your mom).

trigger words aren't need imo. I also recommend a low cfg value like 3.5 or 5. 512x768, 512x512, or 768x768 likely works best, then use hi-res or img2img to get a higher resolution. model was not created with ?????? in mind, but it can be unintentionally horny sometimes, so prompt accordingly (simpler prompts get hit harder by this, so be descriptive in what you want).

images generated with DEIS using onnxdiffusersui with max attention slicing, so you probably won't be able to recreate them with auto1111 due to how it handles seeds. quality should be comparable though.