fountai's picture
push
fca8815
raw
history blame
6.84 kB
import gradio as gr
from PIL import Image
import os
from src.flux.xflux_pipeline import XFluxPipeline
import random
import spaces
def run_xflux_pipeline(
prompt, image, repo_id, name, device,
model_type, width, height, timestep_to_start_cfg, num_steps, true_gs, guidance,
neg_prompt="",
negative_image=None,
save_path='results', control_type='depth', use_controlnet=False, seed=None, num_images_per_prompt=1, use_lora=False, lora_weight=0.7, lora_repo_id="XLabs-AI/flux-lora-collection", lora_name="realism_lora.safetensors", use_ip=False
):
# Montando os argumentos simulando a linha de comando
class Args:
def __init__(self):
self.prompt = prompt
self.image = image
self.control_type = control_type
self.repo_id = repo_id
self.name = name
self.device = device
self.use_controlnet = use_controlnet
self.model_type = model_type
self.width = width
self.height = height
self.timestep_to_start_cfg = timestep_to_start_cfg
self.num_steps = num_steps
self.true_gs = true_gs
self.guidance = guidance
self.num_images_per_prompt = num_images_per_prompt
self.seed = seed if seed else 123456789
self.neg_prompt = neg_prompt
self.img_prompt = Image.open(image)
self.neg_img_prompt = Image.open(negative_image) if negative_image else None
self.ip_scale = 1.0
self.neg_ip_scale = 1.0
self.local_path = None
self.ip_repo_id = "XLabs-AI/flux-ip-adapter"
self.ip_name = "flux-ip-adapter.safetensors"
self.ip_local_path = None
self.lora_repo_id = lora_repo_id
self.lora_name = lora_name
self.lora_local_path = None
self.offload = False
self.use_ip = use_ip
self.use_lora = use_lora
self.lora_weight = lora_weight
self.save_path = save_path
args = Args()
# Carregar a imagem se fornecida
if args.image:
image = Image.open(args.image)
else:
image = None
# Inicializar o pipeline com os parâmetros necessários
xflux_pipeline = XFluxPipeline(args.model_type, args.device, args.offload)
# Configurar ControlNet se necessário
if args.use_controlnet:
print('Loading ControlNet:', args.local_path, args.repo_id, args.name)
xflux_pipeline.set_controlnet(args.control_type, args.local_path, args.repo_id, args.name)
if args.use_ip:
print('load ip-adapter:', args.ip_local_path, args.ip_repo_id, args.ip_name)
xflux_pipeline.set_ip(args.ip_local_path, args.ip_repo_id, args.ip_name)
if args.use_lora:
print('load lora:', args.lora_local_path, args.lora_repo_id, args.lora_name)
xflux_pipeline.set_lora(args.lora_local_path, args.lora_repo_id, args.lora_name, args.lora_weight)
# Laço para gerar imagens
images = []
for _ in range(args.num_images_per_prompt):
seed = random.randint(0, 2147483647)
result = xflux_pipeline(
prompt=args.prompt,
controlnet_image=image,
width=args.width,
height=args.height,
guidance=args.guidance,
num_steps=args.num_steps,
seed=seed,
true_gs=args.true_gs,
neg_prompt=args.neg_prompt,
timestep_to_start_cfg=args.timestep_to_start_cfg,
image_prompt=args.img_prompt,
neg_image_prompt=args.neg_img_prompt,
ip_scale=args.ip_scale,
neg_ip_scale=args.neg_ip_scale,
)
images.append(result)
return images
@spaces.GPU(duration=500)
def process_image(image, prompt, steps, use_lora, use_controlnet, use_depth, use_hed, use_ip, lora_name, lora_path, lora_weight, negative_image, neg_prompt, true_gs, guidance, cfg):
return run_xflux_pipeline(
prompt=prompt,
neg_prompt=neg_prompt,
image=image,
negative_image=negative_image,
lora_name=lora_name,
lora_weight=lora_weight,
lora_repo_id=lora_path,
control_type="depth" if use_depth else "hed" if use_hed else "canny",
repo_id="XLabs-AI/flux-controlnet-collections",
name="flux-depth-controlnet.safetensors",
device="cuda",
use_controlnet=use_controlnet,
model_type="flux-dev",
width=1024,
height=1024,
timestep_to_start_cfg=cfg,
num_steps=steps,
num_images_per_prompt=4,
use_lora=use_lora,
true_gs=true_gs,
use_ip=use_ip,
guidance=guidance
)
custom_css = """
body {
background: rgb(24, 24, 27);
}
.gradio-container {
background: rgb(24, 24, 27);
}
.app-container {
background: rgb(24, 24, 27);
}
gradio-app {
background: rgb(24, 24, 27);
}
.sidebar {
background: rgb(31, 31, 35);
border-right: 1px solid rgb(41, 41, 41);
}
"""
with gr.Blocks(css=custom_css) as demo:
with gr.Row(elem_classes="app-container"):
with gr.Column():
input_image = gr.Image(label="Image", type="filepath")
negative_image = gr.Image(label="Negative_image", type="filepath")
submit_btn = gr.Button("Submit")
with gr.Column():
prompt = gr.Textbox(label="Prompt")
neg_prompt = gr.Textbox(label="Neg Prompt")
steps = gr.Slider(step=1, minimum=1, maximum=64, value=28, label="Num Steps")
use_lora = gr.Checkbox(label="Use LORA", value=True)
lora_path = gr.Textbox(label="LoraPath", value="XLabs-AI/flux-lora-collection")
lora_name = gr.Textbox(label="LoraName", value="realism_lora.safetensors")
lora_weight = gr.Slider(step=0.1, minimum=0, maximum=1, value=0.7, label="Lora Weight")
controlnet = gr.Checkbox(label="Use Controlnet(by default uses canny)", value=True)
use_ip = gr.Checkbox(label="Use IP")
use_depth = gr.Checkbox(label="Use depth")
use_hed = gr.Checkbox(label="Use hed")
true_gs = gr.Slider(step=0.1, minimum=0, maximum=10, value=3.5, label="TrueGs")
guidance = gr.Slider(minimum=1, maximum=10, value=4, label="Guidance")
cfg = gr.Slider(minimum=1, maximum=10, value=1, label="CFG")
with gr.Column():
output = gr.Gallery(label="Galery output", elem_classes="galery", selected_index=0)
submit_btn.click(process_image, inputs=[input_image, prompt, steps, use_lora, controlnet, use_depth, use_hed, use_ip, lora_name, lora_path, lora_weight, negative_image, neg_prompt, true_gs, guidance, cfg], outputs=output)
demo.launch(share=True)