OzzyGT/Krea_2_Turbo_nunchaku_lite_nvfp4

Krea 2 Turbo Nunchaku Lite NVFP4

Left: original bf16  ·  Right: this NVFP4 model (same prompt and seed).

This is a 4-bit quantized version of krea/Krea-2-Turbo:

• Transformer — NVFP4 (W4A4 SVDQuant) via Nunchaku Lite, run through optimized CUDA kernels loaded with the Hugging Face kernels package.
• Text encoder (Qwen3VLModel) — NF4 (4-bit NormalFloat) via bitsandbytes.

Requirements:

• A diffusers build with Nunchaku Lite support (PR #14100, until merged).
• pip install -U kernels bitsandbytes
• NVFP4 kernels require a Blackwell (RTX 50 / RTX PRO) GPU.

Sample image

The quantized (right) image above was generated with the following prompt and settings (seed 7):

import torch
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained("OzzyGT/Krea_2_Turbo_nunchaku_lite_nvfp4", torch_dtype=torch.bfloat16)
pipe.to("cuda")

prompt = (
    "A cozy corner bookstore-cafe on a rainy evening, cinematic wide shot. "
    'A large hand-lettered chalkboard sign in the window reads "FRESH COFFEE & OLD BOOKS" '
    "and below it in smaller chalk letters \"open 'til late\". "
    "Warm golden light spills onto wet cobblestones that mirror pink and blue neon reflections. "
    "Inside, tall mahogany shelves are packed with hundreds of colorful book spines with tiny legible titles, "
    "a barista in a striped apron pours delicate latte art, steam curling upward, "
    "a tabby cat sleeps on a windowsill beside a stack of paperbacks. "
    "Intricate detail, sharp focus, shallow depth of field, photorealistic, rich color grading."
)

image = pipe(
    prompt,
    num_inference_steps=8,
    guidance_scale=0.0,
    height=1024,
    width=1024,
    generator=torch.Generator("cuda").manual_seed(7),
).images[0]
image.save("sample.png")