Update app.py

app.py

@@ -7,8 +7,13 @@ from PIL import Image
 # === 模型定义 ===
 llm = pipeline("text2text-generation", model="google/flan-t5-large")
 stt = pipeline("automatic-speech-recognition", model="openai/whisper-small")
-sd_pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
-sd_pipe.to("cuda" if torch.cuda.is_available() else "cpu")
+
+# 加载 Stable Diffusion v1.5 模型（去掉 float16）
+pipe = StableDiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    safety_checker=None,  # 可选：避免屏蔽图像
+)
+pipe = pipe.to("cpu")  # 或者 "cuda" 如果你在 GPU 上部署
 
 # === Prompt-to-Prompt ===
 def refine_prompt(user_input):
@@ -18,7 +23,7 @@ def refine_prompt(user_input):
 
 # === Prompt-to-Image ===
 def generate_image(prompt, model_choice, num_steps, guidance):
-    image = sd_pipe(prompt, num_inference_steps=num_steps, guidance_scale=guidance).images[0]
+    image = pipe(prompt, num_inference_steps=num_steps, guidance_scale=guidance).images[0]
     return prompt, image
 
 # === 总流程 ===
@@ -27,6 +32,7 @@ def process_all(user_input, model_choice, steps, guidance):
     prompt, image = generate_image(refined, model_choice, steps, guidance)
     return refined, image
 
+# === 语音识别 ===
 def transcribe_audio(audio):
     text = stt(audio)["text"]
     return text
@@ -37,10 +43,10 @@ with gr.Blocks() as demo:
 
     with gr.Row():
         with gr.Column():
-            audio_input = gr.Audio(source="microphone", type="filepath", label="🎤 或点击录音")
+            audio_input = gr.Audio(type="filepath", label="🎤 或点击录音")
             voice_btn = gr.Button("使用语音输入")
             textbox = gr.Textbox(label="描述一句画面", placeholder="比如：空中的魔法树屋")
-            model_choice = gr.Radio(["SD v1.4", "SDXL"], value="SDXL", label="选择模型")  # 现在不会影响生成逻辑
+            model_choice = gr.Radio(["SD v1.4", "SDXL"], value="SDXL", label="选择模型")
             steps_slider = gr.Slider(10, 50, value=30, step=5, label="推理步数")
             guidance_slider = gr.Slider(5, 15, value=7.5, step=0.5, label="引导系数")
             submit_btn = gr.Button("生成图像 🎨")
@@ -52,6 +58,6 @@ with gr.Blocks() as demo:
     voice_btn.click(fn=transcribe_audio, inputs=audio_input, outputs=textbox)
     submit_btn.click(fn=process_all, inputs=[textbox, model_choice, steps_slider, guidance_slider], outputs=[refined_out, image_out])
 
-# === 启动 ===
 demo.launch()
 
+