<h1 id="first-and-foremost-i-just-started-doing-lora-training-work-and-would-appreciate-likes-and-encouragement.-thanks">First and foremost, I just started doing LoRa Training work and would appreciate likes and encouragement. Thanks 🙏</h1><h1 id="lara-croft-atlantis-2026-krea-2">Lara Croft Atlantis 2026 — Krea 2</h1>A LoRa for the new Lara Croft of the Atlantis/Catalyst games. It took me several days to generate a body and face and equipment accurate v1 dataset of over 200 images to train. I’ll be updating and refining in the upcoming weeks especially once new images of the new Lara are widely available on the internet Trigger: <code>laracroft26, Lara Croft, Lara Croft Atlantis</code>Settings: strength 0.9 · <code>LoraLoaderModelOnly</code> · Krea 2 Turbo at 8 steps, CFG 1.0, euler/simple · ~1.0–1.6 MP. CFG 1.0 means negative prompts do nothing — say it all in the positive.Prompt in plain sentences, not tags. It was captioned in prose:<blockquote>A full-body shot of laracroft26, Lara Croft, at a low camera angle. She is kneeling in an obsidian stair well, wearing a dirty teal tank top, brown leather shorts and a jade pendant. Her expression is focused, her hair in a single long braid. Lit by hard rim light. Photorealistic render.</blockquote>Face, build, hair and freckles are baked into the trigger — you don't need to describe her. Wardrobe is free: classic kit, gym, gala, swim and casual all work.Three things worth knowing<ul><li>Never write <code>three-quarter view</code> — it makes duplicates of her. Say <code>facing the camera</code>, <code>in profile</code>, or <code>low camera angle</code> instead.</li><li><code>wet skin</code> + warm light turns her oily. Use <code>matte skin</code> if you want it clean.</li><li>Likeness needs pixels. In a wide shot with a tiny figure the face will drift.</li></ul>Two epochs — 7 and 10. Try both at 0.9 on the same seed and pick. And yes: Model is NSFW trained.

Files with SHA256: f43ded245254151da24e495b5e882f07648ee0878b70cd17d32f83903c3d976f