Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,13 +4,13 @@ import gradio as gr
|
|
| 4 |
from transformers import pipeline
|
| 5 |
from diffusers import StableDiffusionPipeline
|
| 6 |
|
| 7 |
-
# 如果需要使用 Hugging Face 访问令牌,取消下面
|
| 8 |
# from huggingface_hub import login
|
| 9 |
# login(token=os.getenv("HUGGINGFACE_TOKEN"))
|
| 10 |
|
| 11 |
-
# Step 1: Prompt-to-Prompt 模块,使用 Flan-T5 生成结构化提示词
|
| 12 |
llm = pipeline(
|
| 13 |
-
"text2text-generation",
|
| 14 |
model="google/flan-t5-large",
|
| 15 |
device=0 if torch.cuda.is_available() else -1
|
| 16 |
)
|
|
@@ -30,17 +30,32 @@ sd_xl = sd_xl.to("cuda" if torch.cuda.is_available() else "cpu")
|
|
| 30 |
|
| 31 |
# 可选:语音输入模块,使用 Whisper
|
| 32 |
asr = pipeline(
|
| 33 |
-
"automatic-speech-recognition",
|
| 34 |
model="openai/whisper-base",
|
| 35 |
device=0 if torch.cuda.is_available() else -1
|
| 36 |
)
|
| 37 |
|
|
|
|
| 38 |
def transcribe(audio_path):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
text = asr(audio_path)["text"]
|
| 40 |
return text
|
| 41 |
|
| 42 |
|
| 43 |
def generate(description, model_choice, guidance_scale, negative_prompt, style):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
# 构造给 LLM 的指令
|
| 45 |
instruction = (
|
| 46 |
f"请将以下简短描述扩展为 Stable Diffusion 友好的提示词,包含细节和风格:\n"
|
|
@@ -49,6 +64,7 @@ def generate(description, model_choice, guidance_scale, negative_prompt, style):
|
|
| 49 |
)
|
| 50 |
result = llm(instruction, max_length=128)[0]["generated_text"].strip()
|
| 51 |
prompt = result
|
|
|
|
| 52 |
# 根据模型选择生成图像
|
| 53 |
pipeline_model = sd_xl if model_choice == "SDXL" else sd_v15
|
| 54 |
image = pipeline_model(
|
|
@@ -58,33 +74,35 @@ def generate(description, model_choice, guidance_scale, negative_prompt, style):
|
|
| 58 |
).images[0]
|
| 59 |
return prompt, image
|
| 60 |
|
|
|
|
| 61 |
# Step 3: 构建 Gradio 界面
|
| 62 |
with gr.Blocks(title="Prompt-to-Image Generator") as demo:
|
| 63 |
gr.Markdown("## 基于 LLM 的提示词生成与 Stable Diffusion 图像生成")
|
|
|
|
| 64 |
with gr.Row():
|
| 65 |
with gr.Column():
|
| 66 |
desc_input = gr.Textbox(label="文本描述", placeholder="例如:空中的魔法树屋")
|
| 67 |
style_dropdown = gr.Dropdown(
|
| 68 |
-
choices=["幻想风格", "赛博朋克",
|
| 69 |
label="选择风格"
|
| 70 |
)
|
| 71 |
model_radio = gr.Radio(
|
| 72 |
-
choices=["SD v1.5", "SDXL"],
|
| 73 |
-
value="SD v1.5",
|
| 74 |
label="选择模型"
|
| 75 |
)
|
| 76 |
guidance_slider = gr.Slider(
|
| 77 |
-
minimum=0, maximum=20, step=0.5, value=7.5,
|
| 78 |
label="Guidance Scale"
|
| 79 |
)
|
| 80 |
neg_text = gr.Textbox(
|
| 81 |
-
label="反向提示词",
|
| 82 |
placeholder="排除内容(如:低分辨率、水印)"
|
| 83 |
)
|
| 84 |
use_voice = gr.Checkbox(label="启用语音输入(加分项)")
|
| 85 |
-
# 移除 'source' 参数以兼容 Gradio 版本
|
| 86 |
audio_input = gr.Audio(type="filepath", label="语音输入")
|
| 87 |
generate_btn = gr.Button("生成图像")
|
|
|
|
| 88 |
with gr.Column():
|
| 89 |
prompt_output = gr.Textbox(label="生成的提示词")
|
| 90 |
image_output = gr.Image(label="生成的图像")
|
|
@@ -98,13 +116,16 @@ with gr.Blocks(title="Prompt-to-Image Generator") as demo:
|
|
| 98 |
inputs=[audio_input, use_voice],
|
| 99 |
outputs=desc_input
|
| 100 |
)
|
|
|
|
| 101 |
# 点击按钮生成提示词并绘图
|
| 102 |
generate_btn.click(
|
| 103 |
-
fn=generate,
|
| 104 |
-
inputs=[desc_input, model_radio, guidance_slider, neg_text, style_dropdown],
|
| 105 |
outputs=[prompt_output, image_output]
|
| 106 |
)
|
| 107 |
|
| 108 |
# Step 4: 启动应用
|
| 109 |
if __name__ == "__main__":
|
| 110 |
-
demo.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
|
|
|
|
|
| 4 |
from transformers import pipeline
|
| 5 |
from diffusers import StableDiffusionPipeline
|
| 6 |
|
| 7 |
+
# 如果需要使用 Hugging Face 访问令牌,取消下面两行的注释并设置环境变量 HUGGINGFACE_TOKEN
|
| 8 |
# from huggingface_hub import login
|
| 9 |
# login(token=os.getenv("HUGGINGFACE_TOKEN"))
|
| 10 |
|
| 11 |
+
# Step 1: Prompt-to-Prompt 模块,使用 Flan - T5 生成结构化提示词
|
| 12 |
llm = pipeline(
|
| 13 |
+
task="text2text-generation",
|
| 14 |
model="google/flan-t5-large",
|
| 15 |
device=0 if torch.cuda.is_available() else -1
|
| 16 |
)
|
|
|
|
| 30 |
|
| 31 |
# 可选:语音输入模块,使用 Whisper
|
| 32 |
asr = pipeline(
|
| 33 |
+
task="automatic-speech-recognition",
|
| 34 |
model="openai/whisper-base",
|
| 35 |
device=0 if torch.cuda.is_available() else -1
|
| 36 |
)
|
| 37 |
|
| 38 |
+
|
| 39 |
def transcribe(audio_path):
|
| 40 |
+
"""
|
| 41 |
+
对音频文件进行转录
|
| 42 |
+
:param audio_path: 音频文件路径
|
| 43 |
+
:return: 转录后的文本
|
| 44 |
+
"""
|
| 45 |
text = asr(audio_path)["text"]
|
| 46 |
return text
|
| 47 |
|
| 48 |
|
| 49 |
def generate(description, model_choice, guidance_scale, negative_prompt, style):
|
| 50 |
+
"""
|
| 51 |
+
根据输入生成图像
|
| 52 |
+
:param description: 文本描述
|
| 53 |
+
:param model_choice: 选择的模型
|
| 54 |
+
:param guidance_scale: 引导比例
|
| 55 |
+
:param negative_prompt: 反向提示词
|
| 56 |
+
:param style: 选择的风格
|
| 57 |
+
:return: 生成的提示词和图像
|
| 58 |
+
"""
|
| 59 |
# 构造给 LLM 的指令
|
| 60 |
instruction = (
|
| 61 |
f"请将以下简短描述扩展为 Stable Diffusion 友好的提示词,包含细节和风格:\n"
|
|
|
|
| 64 |
)
|
| 65 |
result = llm(instruction, max_length=128)[0]["generated_text"].strip()
|
| 66 |
prompt = result
|
| 67 |
+
|
| 68 |
# 根据模型选择生成图像
|
| 69 |
pipeline_model = sd_xl if model_choice == "SDXL" else sd_v15
|
| 70 |
image = pipeline_model(
|
|
|
|
| 74 |
).images[0]
|
| 75 |
return prompt, image
|
| 76 |
|
| 77 |
+
|
| 78 |
# Step 3: 构建 Gradio 界面
|
| 79 |
with gr.Blocks(title="Prompt-to-Image Generator") as demo:
|
| 80 |
gr.Markdown("## 基于 LLM 的提示词生成与 Stable Diffusion 图像生成")
|
| 81 |
+
|
| 82 |
with gr.Row():
|
| 83 |
with gr.Column():
|
| 84 |
desc_input = gr.Textbox(label="文本描述", placeholder="例如:空中的魔法树屋")
|
| 85 |
style_dropdown = gr.Dropdown(
|
| 86 |
+
choices=["幻想风格", "赛博朋克", "写实主义"],
|
| 87 |
label="选择风格"
|
| 88 |
)
|
| 89 |
model_radio = gr.Radio(
|
| 90 |
+
choices=["SD v1.5", "SDXL"],
|
| 91 |
+
value="SD v1.5",
|
| 92 |
label="选择模型"
|
| 93 |
)
|
| 94 |
guidance_slider = gr.Slider(
|
| 95 |
+
minimum=0, maximum=20, step=0.5, value=7.5,
|
| 96 |
label="Guidance Scale"
|
| 97 |
)
|
| 98 |
neg_text = gr.Textbox(
|
| 99 |
+
label="反向提示词",
|
| 100 |
placeholder="排除内容(如:低分辨率、水印)"
|
| 101 |
)
|
| 102 |
use_voice = gr.Checkbox(label="启用语音输入(加分项)")
|
|
|
|
| 103 |
audio_input = gr.Audio(type="filepath", label="语音输入")
|
| 104 |
generate_btn = gr.Button("生成图像")
|
| 105 |
+
|
| 106 |
with gr.Column():
|
| 107 |
prompt_output = gr.Textbox(label="生成的提示词")
|
| 108 |
image_output = gr.Image(label="生成的图像")
|
|
|
|
| 116 |
inputs=[audio_input, use_voice],
|
| 117 |
outputs=desc_input
|
| 118 |
)
|
| 119 |
+
|
| 120 |
# 点击按钮生成提示词并绘图
|
| 121 |
generate_btn.click(
|
| 122 |
+
fn=generate,
|
| 123 |
+
inputs=[desc_input, model_radio, guidance_slider, neg_text, style_dropdown],
|
| 124 |
outputs=[prompt_output, image_output]
|
| 125 |
)
|
| 126 |
|
| 127 |
# Step 4: 启动应用
|
| 128 |
if __name__ == "__main__":
|
| 129 |
+
demo.launch(server_name="0.0.0.0", server_port=7860, share=True)
|
| 130 |
+
|
| 131 |
+
|