Ready to deplay as api only

Dockerfile

@@ -1,161 +0,0 @@
-# Stage 1: Build Cuda toolkit
-<<<<<<< HEAD
-FROM ubuntu:22.04 as cuda-setup
-
-
-ARG DEBIAN_FRONTEND=noninteractive
-
-# Install necessary libraries including libxml2
-RUN apt-get update && \
-    apt-get install -y gcc libxml2 && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-COPY cuda_12.4.0_550.54.14_linux.run .
-
-# Install wget, download cuda-toolkit and run
-RUN chmod +x cuda_12.4.0_550.54.14_linux.run && \
-    ./cuda_12.4.0_550.54.14_linux.run --silent --toolkit --override
-
-# Second Stage: Copy necessary CUDA directories install flash-attn
-FROM ubuntu:22.04 as base-layer
-
-# Copy the CUDA toolkit from the first stage
-COPY --from=cuda-setup /usr/local/cuda-12.4 /usr/local/cuda-12.4
-
-# Set environment variables to enable CUDA commands
-ENV PATH=/usr/local/cuda-12.4/bin:${PATH}
-ENV LD_LIBRARY_PATH=/usr/local/cuda-12.4/lib64:${LD_LIBRARY_PATH}
-
-# Install Python, pip, and virtualenv
-RUN apt-get update && \ 
-    apt-get install -y python3 python3-pip python3-venv git && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-# Create a virtual environment and install dependencies
-RUN python3 -m venv /venv
-ENV PATH="/venv/bin:$PATH"
-FROM ubuntu:22.04 as cuda-setup
-
-
-ARG DEBIAN_FRONTEND=noninteractive
-
-# Install necessary libraries including libxml2
-RUN apt-get update && \
-    apt-get install -y gcc libxml2 && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-COPY cuda_12.4.0_550.54.14_linux.run .
-
-# Install wget, download cuda-toolkit and run
-RUN chmod +x cuda_12.4.0_550.54.14_linux.run && \
-    ./cuda_12.4.0_550.54.14_linux.run --silent --toolkit --override
-
-# Second Stage: Copy necessary CUDA directories install flash-attn
-FROM ubuntu:22.04 as base-layer
-
-# Copy the CUDA toolkit from the first stage
-COPY --from=cuda-setup /usr/local/cuda-12.4 /usr/local/cuda-12.4
-
-# Set environment variables to enable CUDA commands
-ENV PATH=/usr/local/cuda-12.4/bin:${PATH}
-ENV LD_LIBRARY_PATH=/usr/local/cuda-12.4/lib64:${LD_LIBRARY_PATH}
-
-# Install Python, pip, and virtualenv
-RUN apt-get update && \ 
-    apt-get install -y python3 python3-pip python3-venv git && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-# Create a virtual environment and install dependencies
-RUN python3 -m venv /venv
-ENV PATH="/venv/bin:$PATH"
-
-# Llama.cpp requires the ENV variable be set to signal the CUDA build and be built with the CMAKE variables from pip for python use
-ENV LLAMA_CUBLAS=1
-RUN pip install --no-cache-dir torch packaging wheel && \
-    pip install flash-attn && \
-RUN pip install --no-cache-dir torch packaging wheel && \
-    pip install flash-attn && \
-    pip install gradio && \
-    CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama_cpp_python==0.2.55 && \
-    CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama_cpp_python==0.2.55 && \
-=======
-FROM drakosfire/cuda-base:latest as base-layer
-
-# Llama.cpp requires the ENV variable be set to signal the CUDA build and be built with the CMAKE variables from pip for python use
-ENV LLAMA_CUBLAS=1
-RUN apt-get update && \
-    apt-get install -y python3 python3-pip python3-venv && \
-    pip install gradio && \
-    CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python && \
->>>>>>> 9a956dd (Polished and launch to Hugging Face)
-    pip install pillow && \ 
-    pip install diffusers && \
-    pip install accelerate && \
-    pip install transformers && \
-<<<<<<< HEAD
-    pip install peft && \
-    pip install pip install PyGithub
-
-
-FROM ubuntu:22.04 as final-layer
-
-COPY --from=base-layer /usr/local/cuda-12.4 /usr/local/cuda-12.4
-COPY --from=base-layer /venv /venv
-
-ENV PATH=/usr/local/cuda-12.4/bin:/venv/bin:${PATH}
-ENV LD_LIBRARY_PATH=/usr/local/cuda-12.4/lib64:${LD_LIBRARY_PATH}
-ENV LLAMA_CPP_LIB=/venv/lib/python3.10/site-packages/llama_cpp/libllama.so
-ENV VIRTUAL_ENV=/venv
-
-# Install Python and create a user
-RUN apt-get update && apt-get install -y python3 python3-venv && apt-get clean && rm -rf /var/lib/apt/lists/* && \
-    useradd -m -u 1000 user
-    
-
-# Install Python and create a user
-RUN apt-get update && apt-get install -y python3 python3-venv && apt-get clean && rm -rf /var/lib/apt/lists/* && \
-    useradd -m -u 1000 user
-    
-ENV PATH="$VIRTUAL_ENV/bin:$PATH"
-# Set working directory and user
-COPY . /home/user/app
-# Set working directory and user
-COPY . /home/user/app
-WORKDIR /home/user/app
-RUN chown -R user:user /home/user/app/ && \
-    mkdir -p /home/user/app/output && \
-    chown -R user:user /home/user/app/image_temp && \
-    chown -R user:user /home/user/app/output 
-=======
-    pip install peft
-
-FROM base-layer as final-layer
-
-RUN useradd -m -u 1000 user 
-   
-    # mkdir -p /home/user/.cache && \  
-    # chmod 777 /home/user/.cache && \  
-    # chown -R user:user /home/user/app/ 
-# Set environment variables for copied builds of cuda and flash-attn in /venv
-
-
-ENV PATH=/usr/local/cuda-12.4/bin:/venv/bin:${PATH}
-ENV LD_LIBRARY_PATH=/usr/local/cuda-12.4/lib64:${LD_LIBRARY_PATH}
-
-ENV VIRTUAL_ENV=/venv
-RUN python3 -m venv $VIRTUAL_ENV
-ENV PATH="$VIRTUAL_ENV/bin:$PATH"
-# Set working directory and user
-WORKDIR /home/user/app
->>>>>>> 9a956dd (Polished and launch to Hugging Face)
-
-USER user
-    
-# Set the entrypoint
-EXPOSE 8000
-
-ENTRYPOINT ["python", "main.py"]

README.md

@@ -1,38 +1,31 @@
-<<<<<<< HEAD
----
-<<<<<<< HEAD
-license: mit
-title: Collectible Card Generator
-short_description: Use LLM and SD to make custom collectible cards
-app_file: main.py
-sdk: docker
-app_port: 8000
----
-=======
-title: CollectibleCardGenerator
-emoji: 🐠
-colorFrom: green
-colorTo: yellow
-sdk: docker
-pinned: false
-license: mit
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
->>>>>>> f6d0508 (initial commit)
-
----
-title: CollectibleCardGenerator
-emoji: 🐠
-colorFrom: green
-colorTo: yellow
-sdk: docker
-pinned: false
-license: mit
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
-=======
-# CardGenerator
-Takes user input and generates a collectible card with custom or LLM generated text and image generation
->>>>>>> 9a956dd (Polished and launch to Hugging Face)
+# Drakosfire's Dungeons and Dragons Item Card Generator
+
+Welcome to the Drakosfire's Dungeons and Dragons Item Card Generator! This innovative tool harnesses the power of AI to generate unique item cards for your D&D adventures. 
+
+## Overview
+
+This generator leverages an API call to [Replicate](https://replicate.com/) using Llama 3 70b, combined with a custom fine-tuned version of the Stable Diffusion SDXL model. You can find more about the specific modelthis project was based on at [Civitai](https://civitai.com/models/129681/sdxl-faetastic).
+
+## Key Features
+
+- **Custom Fine-Tuning**: The backbone of this generator is a fine-tuned Stable Diffusion SDXL model, specifically optimized for generating high-quality, fantasy-themed images on a unique card border that is themed by your imagination.
+
+- **Consistent Card Design**: To ensure each card maintains a uniform appearance, a LoRA (Locally Optimized Representation Approximation) technique was used. This involved training the model with a hand-crafted dataset of card border images, enabling the system to generate new cards with consistent text and image spaces where the generated text and images can be elegantly integrated.
+
+## How It Works
+
+1. Your intitial text along with the prompt is sent to Llama 3 70b to generate a structured python dictionary.
+2. This new text will populate in interactive text fields. If it isn't perfect you can edit the text to fit your item.
+3. THe final text field is the Stable Diffusion prompt, these generate like one sentence stories describing the scene of your item. This field can also be edited.
+4. **Image and Text Generation**: Now generate 4 card template without text and pick your favorite.
+5. Finally, add text to your favorite template.
+3. **Result**: The final product is a beautifully crafted D&D item card, ready for use in your gaming sessions.
+
+## Example Cards
+
+Below are a few examples of the item cards generated using our tool:
+
+![Example Card 1](url-to-example-card1)
+![Example Card 2](url-to-example-card2)
+
+We hope you enjoy enhancing your Dungeons and Dragons experience with this unique tool. Happy adventuring!

img2img.py

@@ -5,69 +5,40 @@ import time
 import utilities as u
 import card_generator as card
 from PIL import Image
+import replicate
+from pathlib import Path
+
 
-pipe = None
 start_time = time.time()
-torch.backends.cuda.matmul.allow_tf32 = True
-model_path = "/media/drakosfire/Shared/models/stable-diffusion/card-generator-v1/card-generator-v1.safetensors"
-lora_path = "/media/drakosfire/Shared/models/stable-diffusion/card-generator-v1/blank-card-template-5.safetensors"
-detail_lora_path = "/media/drakosfire/Shared/models/stable-diffusion/card-generator-v1/add-detail-xl.safetensors"
-mimic_lora_path = :"/media/drakosfire/Shared/CardGenerator/models/stable-diffusion/Loras/EnvyMimicXL01.safetensors"
 temp_image_path = "./image_temp/"
-card_pre_prompt = " blank magic card,high resolution, detailed intricate high quality border, textbox, high quality detailed magnum opus drawing of a "
-negative_prompts = "text, words, numbers, letters"
-image_list = []
-
-class img_generator():
 
-    def load_img_gen(self,prompt, item, mimic = None):
-        print(f"Input Prompt = {prompt} + Item : {item}")
-        prompt = card_pre_prompt + item + ' ' + prompt
-        print(prompt)
-        
-        
-        pipe = StableDiffusionXLImg2ImgPipeline.from_single_file(model_path,
-                                                        custom_pipeline="low_stable_diffusion",                                                       
-                                                            torch_dtype=torch.float16, 
-                                                            variant="fp16").to("cuda")  
-        # Load LoRAs for controlling image
-        #pipe.load_lora_weights(lora_path, weight_name = "blank-card-template-5.safetensors",adapter_name = 'blank-card-template')    
-        pipe.load_lora_weights(detail_lora_path, weight_name = "add-detail-xl.safetensors", adapter_name = "add-detail-xl")
-        
-        # If mimic keyword has been detected, load the mimic LoRA and set adapter values
-        if mimic:
-            print("MIMIC!")
-            pipe.load_lora_weights(mimic_lora_path, weight_name = "EnvyMimicXL01.safetensors", adapter_name = "EnvyMimicXL")
-            pipe.set_adapters(['blank-card-template', "add-detail-xl", "EnvyMimicXL"], adapter_weights = [0.9,0.9,1.0])
-        else : 
-            pipe.set_adapters([ "add-detail-xl"], adapter_weights = [0.9])       
-        pipe.enable_vae_slicing()
-        return pipe, prompt
 
-    def preview_and_generate_image(self,x,pipe, prompt, user_input_template, item):    
-        img_start = time.time()   
-        image = pipe(prompt=prompt,
-                    strength = .9,
-                    guidance_scale = 5,
-                    image= user_input_template,
-                    negative_prompt = negative_prompts,
-                    num_inference_steps=40,
-                    height = 1024, width = 768).images[0]
-        
-        image = image.save(temp_image_path+str(x) + f"{item}.png")
-        output_image_path = temp_image_path+str(x) + f"{item}.png"
-        img_time = time.time() - img_start
-        img_its = 50/img_time
-        print(f"image gen time = {img_time} and {img_its} it/s")
-            
-        # Delete the image variable to keep VRAM open to load the LLM
-        del image
-        print(f"Memory after del {torch.cuda.memory_allocated()}")
-        print(image_list)
-        total_time = time.time() - start_time
-        print(total_time)
 
-        return output_image_path
+def preview_and_generate_image(num_images,sd_prompt, user_input_template, item):    
+    # Replicate is expecting a path
+    card_template = Path(user_input_template[0][0])
+    print(card_template)
+    img_start = time.time()   
+    output=replicate.run(
+            "drakosfire/card-generator-v1:4e46bb44e6444d104b58e34a6d2ab66833aa84ba0dfe84ae57ef63d8e15da467",
+            input={
+            "item":item,
+            "sd_prompt":sd_prompt,
+            "num_images":num_images,
+            "input_template":card_template
+
+        }
+    )
+    
+    
+    img_time = time.time() - img_start
+    img_its = 35/img_time
+    print(f"image gen time = {img_time} and {img_its} it/s")
+    
+    total_time = time.time() - start_time
+    print(total_time)
+
+    return output
 
 
 

main.py

@@ -1,6 +1,5 @@
+import replicate
 import img2img
-from img2img import img_generator
-
 import card_generator as card
 import utilities as u
 import ctypes
@@ -107,20 +106,20 @@ with gr.Blocks() as demo:
     def delete_temp_images():
         image_list = u.directory_contents('./image_temp')
         u.delete_files(image_list)
-        img2img.image_list.clear()
+        u.image_list.clear()
     
     # Called when pressing button to generate image, updates gallery by returning the list of image URLs
     def generate_image_update_gallery(num_img, sd_prompt,item_name, built_template):
         delete_temp_images()
         print(f"sd_prompt is a {type(sd_prompt)}")
         image_list = []
-        image_generator = img_generator()
-        img_gen, prompt = image_generator.load_img_gen(sd_prompt, item_name)
+        
+        
         for x in range(num_img):
-            preview = image_generator.preview_and_generate_image(x,img_gen, prompt, built_template, item_name)
+            preview = img2img.preview_and_generate_image(x,sd_prompt, built_template, item_name)
             image_list.append(preview)
             yield image_list
-            #generate_gallery.change(image_list)
+            
         del preview
         u.reclaim_mem()
 
@@ -189,7 +188,8 @@ with gr.Blocks() as demo:
                                         object_fit = "contain",
                                         height = "auto",
                                         elem_id = "Template Gallery",
-                                        interactive=True) 
+                                        interactive=True,
+                                        type="filepath") 
     
     seed_image_gallery.select(assign_img_path, outputs = selected_seed_image)
     built_template_gallery.upload(u.receive_upload, inputs=built_template_gallery, outputs= selected_seed_image)
@@ -258,7 +258,7 @@ with gr.Blocks() as demo:
         generate_final_item_card = gr.Button(value = "Add Text", elem_id = "Generate user card")
     
     
-    card_gen_button.click(fn = generate_image_update_gallery, inputs =[num_image_to_generate,item_sd_prompt_output,item_name_output,built_template], outputs= generate_gallery)
+    card_gen_button.click(fn = generate_image_update_gallery, inputs =[num_image_to_generate,item_sd_prompt_output,item_name_output,built_template_gallery], outputs= generate_gallery)
     generate_gallery.select(assign_img_path, outputs = selected_generated_image)
 
         # Button logice calls function when button object is pressed, passing inputs and passing output to components

models

@@ -1 +0,0 @@
-/media/drakosfire/Shared/models

user_input.py

@@ -96,7 +96,7 @@ def prompt_user_input():
         print(image_path)
         
         card.render_text_on_card(user_card_image, output_dict)
-        u.delete_files(img2img.image_list)
+        u.delete_files(u.image_list)
 
 
 

utilities.py

@@ -7,6 +7,8 @@ import requests
 import os
 from PIL import Image
 from pathlib import Path
+
+image_list = []
 # Utility Functions to be called from all modules
 
 # Function to return a list of keys of a nested dictionary using it's key value (item or creature)