✨ Vizio's Lustre 3D ✨
a glossy, polished 3D game-character look, baked straight into Krea 2

🌟 What is this?

Hey! This is my own look for Krea 2, and it is a full checkpoint, not a LoRA. The style is baked right into the weights, so there is no trigger word and no extra LoRA to load. Load it, write your prompt, done.

Everything comes out like a polished 3D game character or a collectible figure: smooth doll-like skin with a soft shine, glossy leather and hair, clean saturated colours and big clear natural eyes. I trained it on about 4,100 curated images on purpose, so it is not just girls in portrait: men, couples, creatures, scenery and objects all get the same finish.

🎯 What it is good for

• 👩‍🎤 Game-character women: portraits, cowboy shots and full-body shots, in armor, robes, ninja gear, evening gowns, anything with a costume

• 🧙 Fantasy and sci-fi characters: knights, mages, witches, elves, pilots

• 👫 Men and couples, they work too

• 🐉 Creatures and scenery: dragons, castles, harbor towns, cozy rooms

• 🍜 Objects and food: glossy cars, ramen, robots, cats, anything that benefits from a shiny 3D finish

📥 Which file should I download?

Two versions, same look. Pick the one that fits your graphics card:

•  bf16  ~24 GB, full quality. Best if you have 24 GB VRAM or more (I run it on a 32 GB card).

•  fp8  ~12.6 GB, half the size, and in my side-by-side tests it gave almost the same images as the bf16 file. Best for 12 to 16 GB VRAM.

💡 Short on VRAM but have lots of system RAM? You can also use the bf16 file and set weight_dtype to fp8_e4m3fn in the Load Diffusion Model node. That is exactly how I made the example images. These VRAM numbers are rough guides from the file sizes, your text encoder, resolution and offload settings change what fits.

⚙️ Settings I use

• Load it with Load Diffusion Model, and use the same text encoder and VAE as regular Krea 2

• The Krea 2 Turbo LoRA (rank 64) at 1.0, 8 steps, CFG 1, sampler euler, scheduler beta

• CFG 1 means a negative prompt does nothing, so you can leave it empty

• I start around 0.5 MP (for example 544x960) and then run a 2x upscale pass for the final image

✍️ Prompting tips

Krea 2 likes plain English sentences, not tag soup. This order works well: framing → who → hair, eyes, outfit → place and light.

• 🎥 Framing words that work: close-up portrait, cowboy shot from the front, full-body shot from a low angle, three-quarter angle

• 🧑 For people I always write young adult woman or adult man

• 👁️ Name natural eye colours (grey, hazel, green, brown, blue)

• 👗 Describe the outfit fully. If you leave it vague it can drift toward revealing, so say things like long sleeves, closed up to the neck, fully covered when you want it covered

• 💪 If you do not want abs, skip words like athletic, toned and fit

• 🍑 The model leans curvy on its own. I use slim hourglass body with a very narrow waist and wide hips, and you can say slim or petite if you want less

🔥 Want the full glossy 3D look? The style is already in the model, but this is the recipe I used for most of my covers. Put it at the start of the prompt:

A 3D character render made in Blender, a CGI game character model with anime-style proportions, large expressive eyes, a small delicate nose, smooth doll-like skin with subsurface scattering, soft rim lighting and a clean glossy finish.

and this at the end:

High-quality 3D model render, vibrant saturated colours.

Other words that push the shine: glossy doll-like skin, smooth rubbery highlights, high gloss finish, collectible figure look, studio-quality CGI. Leave all of it out for a softer, more natural finish. Both work fine.

🧪 Example prompts

Crimson hair, cathedral

A 3D character render made in Blender, a CGI game character model with anime-style proportions, large expressive eyes, a small delicate nose, smooth doll-like skin with subsurface scattering, soft rim lighting and a clean glossy finish. a dutch-angle shot looking down at a young adult woman with long wavy crimson hair and dark brown eyes, deep red lips, in a black off-the-shoulder dress, lying back on the steps of a gothic cathedral with one hand reaching up toward the camera and a sly smile, hundreds of candles and floating embers around her. Fine detail in the skin, a soft glossy glow on the cheeks and lips. High-quality 3D model render, vibrant saturated colours.

Aurora ridge, low-angle leap

A 3D character render made in Blender, a CGI game character model with anime-style proportions, large expressive eyes, a small delicate nose, smooth doll-like skin with subsurface scattering, soft rim lighting and a clean glossy finish. a dramatic low-angle shot, a young adult woman with long white hair and ice blue eyes, in a fitted silver cropped jacket and tight silver pants, leaping off a snowy cliff edge with her hair and a long scarf streaming behind her, a huge green and violet aurora filling the sky, sparkling snow bursting around her. Fine detail in the skin, a soft glossy glow on the cheeks and lips. High-quality 3D model render, vibrant saturated colours.

Every image on this page was made with this checkpoint and the Turbo LoRA only. No other LoRAs, no inpainting, no editing.

🧩 Using it with other LoRAs

Krea 2 character LoRAs work at strength 1.0 (I tested with my Miss Fortune LoRA). Style LoRAs on top change very little at low strength, because the look is already in the checkpoint.

⚠️ Things to know

• 🔤 Text on clothes, logos and neon signs usually comes out as gibberish

• 🏰 Very dark, high-contrast scenes (like night castles) can get heavy, so add some light to the prompt

• 👙 Without an outfit description it likes small, revealing clothes

• 📏 Big bust and wide hips are the default, that is my taste, so say what you want if you want something else

Thanks for downloading, have fun with it! 💖
If you make something you like, post it. I would love to see what you do with it.