Try size: 768 x 1024 or 768 x 1280

Weights: 0.5~1.0

Pictures from realisticBlendXL_v50 and calicomixPrisma_epsV10 and ZImageTurbo

Note: Generally, "grey" and "silver" have a higher failure rate. If it fails, you can try adding "fishnets, mesh, net," to improve it. My prompt is to ensure that "halterneck" works; if you just want "crochet" material, you can simplify it. If the asian base model shows a "profile" shot, please adding "facing viewer" or delete "sideboob". ([Illustrious)

For realistic models, version "1.0 ILXL" would be more suitable; however, for anime base model, version "1.0 plus ILXL" would be better because higher weights are more stable. ([Illustrious)

Without considering image quality, speed, or nipple issues, I personally prefer Flux.1D. However, ZImageTurbo seems to be a new trend, so I trained a its LoRa. Also, if you're a die-hard fan of danbooru, I still recommend Illustrious. (ZImageTurbo)

Sampling method: Euler or Euler / Schedule type: Beta or Simple or DDIM / In fact, Flux.1 D is like the Toyota AE86 Fujiwara tofu shop car. You can continue to take advantage of its performance for downhill turn on mountain roads, but its importance will gradually be relegated to second or third place, which is inevitable. (Flux.1 D)