update app

app.py

@@ -2,6 +2,7 @@ from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
 from diffusers import UniPCMultistepScheduler
 import gradio as gr
 import torch
+from gradio.components import Markdown
 
 # Models
 controlnet_pose = ControlNetModel.from_pretrained(
@@ -12,7 +13,7 @@ controlnet_canny = ControlNetModel.from_pretrained(
 )
 
 pipe = StableDiffusionControlNetPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
+    "dreamlike-art/dreamlike-anime-1.0",
     controlnet=[controlnet_pose, controlnet_canny],
     safety_checker=None, torch_dtype=torch.float16
 ).to('cuda')
@@ -26,13 +27,30 @@ pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
 pipe.enable_xformers_memory_efficient_attention()
 
 # Generator seed,
-generator = torch.manual_seed(3)
+generator = torch.manual_seed(0)
 
+negative_prompt = ("worst quality, low quality, lowres, bad anatomy, bad hands, "
+                   "missing fingers, extra digit, fewer digits")
+
+markdown = """
+## Generate controlled outputs with Mult-ControlNet and Stable Diffusion using 🤗Diffusers
+
+This Space uses pose lines and canny edged image as the additional conditioning. Please refer to the "Examples" for what kind of images are appropriate.
+The Followings are tools available to create such images. In this example, tool 1 is being used.
+1. [Character bones that look like Openpose for blender Ver4.7 Depth+Canny](https://toyxyz.gumroad.com/l/ciojz)
+2. [open-pose-editor](https://github.com/ZhUyU1997/open-pose-editor)
+
+"""
+# **This space using these models**:
+# - ControlNet Model (canny): [lllyasviel/sd-controlnet-canny](https://hf.co/lllyasviel/sd-controlnet-canny)
+# - ControlNet Model (openpose): [lllyasviel/sd-controlnet-openpose](https://hf.co/lllyasviel/sd-controlnet-openpose)
+# - SD Base Model: [dreamlike-art/dreamlike-anime-1.0](https://hf.co/dreamlike-art/dreamlike-anime-1.0)
 
 def generate_images(pose_image, canny_image, prompt):
     output = pipe(
-        prompt,
-        [pose_image, canny_image],
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        image=[pose_image, canny_image],
         generator=generator,
         num_images_per_prompt=3,
         num_inference_steps=20,
@@ -53,16 +71,23 @@ gr.Interface(
         gr.Textbox(
             label="Enter your prompt",
             max_lines=1,
-            placeholder="masterpiece, a professional portrait of woman wearing white shirts",
+            placeholder="Enter your prompt",
         ),
     ],
     outputs=gr.Gallery().style(grid=[2], height="auto"),
-    title="Generate controlled outputs with Mult-ControlNet and Stable Diffusion using 🤗Diffusers",
-    description="This Space uses pose lines and canny edged image as the additional conditioning. Please refer to the \"Examples\" for what kind of images are appropriate.",
+    description=markdown,
     examples=[
-        ["p2_clip.png",
-         "c2_clip.png",
-         "masterpiece, a professional portrait of woman wearing white shirts"
+        ["p11_clip.png",
+         "c11_clip.png",
+         "masterpiece, a professional portrait of woman wearing white shirts, smile, heart hand sign"
+         ],
+        # ["p13_clip.png",
+        #  "c13_clip.png",
+        #  "masterpiece, a professional portrait of woman wearing white shirts, smile"
+        #  ],
+        ["p12_clip.png",
+         "c12_clip.png",
+         "masterpiece, a professional portrait of woman wearing white shirts, smile"
          ],
     ],
     allow_flagging=False,

requirements.txt

@@ -1,4 +1,4 @@
-git+https://github.com/takuma104/diffusers.git@multi-controlnet-ext
+git+https://github.com/huggingface/diffusers
 transformers
 git+https://github.com/huggingface/accelerate
 torch==1.13.1