Tomb Raider: Legacy of Atlantis – Lara Croft (trl_lara) | Krea 2 LoRA

Description

This LoRA is trained on about 97 images in the style of Tomb Raider: Legacy of Atlantis. It teaches Krea 2 the photorealistic 3D video game render look, Lara's face and build, and the game's world.

• Character: Lara Croft with the teal tank top, black fingerless gloves, backpack, holsters and cargo shorts. It also covers the burgundy quilted snow jacket and her braid, ponytail and short bob hairstyles.

• Style: the game's cutscene, screenshot and key-art look, with warm cinematic lighting and shallow depth of field.

• Environments: jungle ruins, Egyptian tombs, Incan gear temples, and snow and ice caves. You can also generate environments without a character.

• Graphics and text: game-style magazine covers, title logos and character profile cards.

Trigger words

TriggerUsetrl_laraThe character. Use it for any shot with Lara.trl_styleThe game's render look. Use it on its own for environments, logos and UI.

Start every prompt with trl_lara, trl_style, for Lara or trl_style, for scenes without her.

Prompting tips

1. Follow the training caption structure. Write long, descriptive natural-language prompts:
   trl_lara, trl_style, a photorealistic 3D video game [cutscene screenshot / screenshot / key art render] of trl_lara, [face], [shot type and camera angle], [expression], [hair], [skin], [build], [outfit], [setting], [lighting], [depth of field], no text.

2. Choose the render type. "Cutscene screenshot" gives close-ups and portraits. "Screenshot" gives action and gameplay. "Key art render" gives polished promo art.

3. Describe the outfit explicitly. If you leave it out, the LoRA falls back to the teal tank top. Name the clothes if you want something different, for example a dress, a leather jacket or the snow outfit.

4. Back-view and face-hidden shots are well covered. Write "seen from behind", "her face is hidden" and "braid over her shoulder".

5. Shot and camera terms that work well: "low-angle", "three-quarter view", "medium shot from the thighs up", "full-body shot" and "close-up from the chest up".

6. Lighting: the training data favors warm, low-key cinematic light with deep shadows. Add "shallow depth of field" for portraits and "deep depth of field" for wide shots.

7. Add no text at the end if you don't want any text in the image.

8. Text in the image:

   • Put the exact text in quotes and say where it sits and what it is made of, for example "chiseled into a stone lintel" or "white serif capitals at the bottom of the frame".

   • Keep it short and write it only once.

   • Text on a clean, free-standing area, like a title card, works better than text on a surface.

9. Strength: start at 0.8–1.0. If the style overpowers a new setting or outfit, lower it to about 0.6–0.8.

10. Settings outside the dataset (cities, deserts, ships, underwater) work, but long prompts help the look carry over.

Example prompt

Copy

trl_lara, trl_style, a photorealistic 3D video game cutscene screenshot of trl_lara, a young woman with an oval face, high cheekbones, defined dark brown eyebrows, almond-shaped dark brown eyes and full lips, shown in a low-angle three-quarter view from the waist up, looking up and away from the camera with an awed, curious expression. Her dark brown hair is pulled back into a long braid, and her tan skin has light freckles and dirt smudges. She has a slim, athletic build and wears a fitted teal tank top, a cord necklace with a green jade pendant and brown leather backpack straps. The setting is an overgrown ancient jungle temple with sunbeams breaking through mist. Warm cinematic lighting, shallow depth of field, no text.

Fan-made, non-commercial. Tomb Raider and Lara Croft are trademarks of their respective owners.

Fill in the base model version, training details and recommended sampler settings yourself. I don't have them from the dataset, so I haven't guessed.