face / app.py
wangliujing70's picture
Update app.py
8b9d9ed verified
Raw History Blame Contribute Delete
2.86 kB
# 表情生成器:上传人脸 + Prompt 控制表情(轻量版)
import gradio as gr
import torch
from diffusers import StableDiffusionImg2ImgPipeline
from PIL import Image
# 使用 SD1.5 图像转图像版本
model_id = "runwayml/stable-diffusion-v1-5"
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
model_id,
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
).to("cuda" if torch.cuda.is_available() else "cpu")
# 表情 prompt 定义
expression_prompts = {
"neutral": "a person with a neutral expression",
"happy": "a smiling person",
"laughing": "a person laughing joyfully",
"funny": "a person making a funny face",
"sad": "a person looking sad",
"angry": "a person looking angry",
"crying": "a person crying",
"loving": "a person showing love",
"embarrassed": "a person looking embarrassed",
"surprised": "a surprised person",
"shocked": "a person with a shocked expression",
"thinking": "a person thinking deeply",
"winking": "a person winking",
"cool": "a person with a cool expression wearing sunglasses",
"relaxed": "a relaxed person",
"delicious": "a person enjoying delicious food",
"kissy": "a person making a kissy face",
"confident": "a confident person",
"sleepy": "a sleepy person",
"silly": "a person acting silly",
"confused": "a confused person"
}
@torch.inference_mode()
def generate_expression(image, expression):
if image.mode != "RGB":
image = image.convert("RGB")
# Resize to 512x512 for compatibility
image = image.resize((512, 512), Image.Resampling.LANCZOS)
prompt = expression_prompts[expression]
negative_prompt = "blurry, low quality, distorted, bad anatomy"
result = pipe(
image=image,
prompt=prompt,
negative_prompt=negative_prompt,
strength=0.6, # 稍微低一些以保留原图结构
guidance_scale=7.5,
num_inference_steps=30
)
return result.images[0], f"生成了表情:{expression}"
expression_list = list(expression_prompts.keys())
with gr.Blocks() as demo:
gr.Markdown("# 🎭 AI 表情生成器(上传图片)\n上传你的脸,选择一个表情,AI 会保留人脸并改变表情。")
with gr.Row():
with gr.Column():
image_input = gr.Image(type="pil", label="上传人脸图像")
expression_input = gr.Radio(expression_list, label="选择一个表情")
submit_btn = gr.Button("生成表情图像")
with gr.Column():
image_output = gr.Image(label="生成结果")
log_output = gr.Textbox(label="状态")
submit_btn.click(
fn=generate_expression,
inputs=[image_input, expression_input],
outputs=[image_output, log_output]
)
if __name__ == "__main__":
demo.launch()