You can deploy this model and start earning money today!

anonymous-cvpr2026/
handcontrolnet

Views

No views yet

HandControlNet

Data-Efficient ControlNet Adapter for Anime Hand Anomaly Mitigation
CVPR 2026 Anonymous Submission

A lightweight ControlNet adapter trained on only 677 high-quality anime hand images, achieving state-of-the-art hand fidelity (FID 16.3, HSA 89.3%) with CPU-optimized 1024×1024 training.

Model Details

Item	Description
Base Model	Stable Diffusion v1.5
Adapter Type	ControlNet (zero-conv)
Training Data	677 hand crops from 5 games
Input	Bounding box → Gaussian heatmap
Resolution	1024×1024
Training Steps	10k
Hardware	i7-12700 CPU (47.5h)
Weights	`advanced_controlnet.pth`

Usage (Diffusers)

python

from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
import torch
from PIL import Image

# Load model
controlnet = ControlNetModel.from_pretrained(
    "anonymous-cvpr2026/handcontrolnet", 
    torch_dtype=torch.float16
)
pipe = StableDiffusionControlNetPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5",
    controlnet=controlnet,
    torch_dtype=torch.float16
).to("cuda")

# Generate
image = pipe(
    prompt="anime girl holding apple, detailed hands",
    controlnet_conditioning_scale=1.0,
    # Your bbox → heatmap logic here
).images[0]

image.save("output.png")