Kokoboyaw's picture
Upload folder using huggingface_hub
9dccfb7 verified
metadata
license: other
license_name: newbie-nc-1.0
license_link: https://huggingface.co/NewBie-AI/NewBie-image-Exp0.1/blob/main/LICENSE.md
base_model: NewBie-AI/NewBie-image-Exp0.1
tags:
  - text-to-image
  - anime
  - dit
  - fp16
library_name: diffusers

NewBie-image-Exp0.1 (FP16)

FP16 quantized version of NewBie-AI/NewBie-image-Exp0.1.

Model Details

  • Original Model: NewBie-image-Exp0.1 (3.5B DiT)
  • Architecture: Next-DiT (based on Lumina research)
  • Quantization Format: FP16
  • Original Precision: BFloat16
  • Size Reduction: ~0%

Usage

With Diffusers (Python)

import torch
from diffusers import NewbiePipeline

pipe = NewbiePipeline.from_pretrained(
    "NewBie-AI/NewBie-image-Exp0.1-fp16",
    torch_dtype=torch.float16,
).to("cuda")

image = pipe(
    prompt="1girl, solo, blue eyes, white hair",
    height=1024,
    width=1024,
    num_inference_steps=28,
).images[0]

With ComfyUI

Use the single-file version with standard Diffusion Model Loader or the dedicated NewBie nodes.

Comparison

Format VRAM Usage Speed Quality
BF16 (original) ~14GB Baseline Best
FP16 ~7GB ~Same Excellent
FP8 E4M3 ~3.5GB Faster* Very Good

*Faster on RTX 40xx/Ada GPUs with FP8 tensor cores

License

This model follows the Newbie Non-Commercial Community License (Newbie-NC-1.0). For non-commercial use only.

Credits

Original model by NewBie-AI. Quantization by NewBieAI-Lab.