tensorart/Bokeh_Depth_Controlnet

Bokeh_Depth_Controlnet

Description

• Input Image: Depth information map generated by depth estimation model
• Output Image: Base model integrates depth information for control, generating more realistic depth-of-field effects and spatial structure control.

The depth map enables the model to better understand spatial relationships within images, achieving more precise depth-of-field control and three-dimensional effect generation.

Example

input	output	Prompt
		a old man
		A lone astronaut in a white suit and helmet, with a backpack, strolls on a rocky, red landscape under a star galaxy
		a eagle , black background
		A crystal heart sits on a gnarly, moss-covered log, with a galaxy background

Use

We recommend using ComfyUI for local inference

With Bokeh

import torch
from diffusers import StableDiffusion3ControlNetPipeline
from diffusers import SD3ControlNetModel
from diffusers.utils import load_image

controlnet = SD3ControlNetModel.from_pretrained("tensorart/Bokeh_Depth_Controlnet")
pipe = StableDiffusion3ControlNetPipeline.from_pretrained(
    "tensorart/bokeh_3.5_medium",
    controlnet=controlnet
)
pipe.to("cuda", torch.float16)

control_image = load_image("https://huggingface.co/tensorart/Bokeh_Depth_Controlnet/resolve/main/images/001_depth.png")
prompt = "A old man talking"
negative_prompt ="anime,render,cartoon,3d,bad hands,extra finger"
negative_prompt_3=""

image = pipe(
    prompt, 
    num_inference_steps=28,
    negative_prompt=negative_prompt, 
    negative_prompt_3=negative_prompt_3,
    control_image=control_image, 
    height=1440,
    width=1440,
    guidance_scale=4,
    controlnet_conditioning_scale=0.85
).images[0]
image.save('image.jpg')

Contact

• Website: https://tensor.art https://tusiart.com
• Developed by: TensorArt
• Api: https://tams.tensor.art/