added assets, models, initial script

app.py

@@ -1,7 +1,86 @@
+import torch
+from diffusers import AutoPipelineForImage2Image
+from PIL import Image, ImageDraw, ImageFont
+import requests
+from io import BytesIO
 import gradio as gr
 
-def greet(name):
-    return "Hello " + name + "!!"
+# log gpu availabilitu
+print(f"Is CUDA available: {torch.cuda.is_available()}")
+print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")
 
-iface = gr.Interface(fn=greet, inputs="text", outputs="text")
-iface.launch()
+
+def image_to_template(generated_image, logo, button_text, punchline, theme_color):
+    # Resize logo if needed
+    logo = logo.resize((100, 100))  # Example size, adjust as needed
+
+    # Create a blank canvas with extra space for logo, punchline, and button
+    canvas_width = max(generated_image.width, logo.width) * 2
+    canvas_height = generated_image.height + logo.height + 100
+    canvas = Image.new('RGB', (canvas_width, canvas_height), 'white')
+    
+    # Paste the logo and the generated image onto the canvas
+    canvas.paste(logo, (10, 10))  # Adjust position as needed
+    canvas.paste(generated_image, (0, logo.height + 20))
+
+    # Add punchline and button
+    draw = ImageDraw.Draw(canvas)
+    font = ImageFont.load_default()  # Or use a custom font
+    text_color = theme_color
+
+    # Draw punchline
+    draw.text((10, logo.height + generated_image.height + 30), punchline, fill=text_color, font=font)
+
+    # Draw button
+    button_position = (10, logo.height + generated_image.height + 60)  # Adjust as needed
+    draw.rectangle([button_position, (canvas_width - 10, canvas_height - 10)], outline=theme_color, fill=text_color)
+    draw.text(button_position, button_text, font=font)
+
+    return canvas
+
+def generate_template(initial_image, logo, prompt, button_text, punchline, image_color, theme_color):
+    pipeline = AutoPipelineForImage2Image.from_pretrained(
+        "./models/kandinsky-2-2-decoder", 
+        torch_dtype=torch.float16, 
+        variant="fp16", 
+        use_safetensors=True
+    )
+
+    # pipeline.unet = torch.compile(pipeline.unet, mode="reduce-overhead", fullgraph=True)
+    pipeline.enable_model_cpu_offload()
+
+    prompt = f"{prompt}, include the color {image_color}"
+    negative_prompt = "low quality, bad quality, blurry, unprofessional"
+
+    generated_image = pipeline(
+        prompt=prompt, 
+        negative_prompt=negative_prompt,
+        image=initial_image,  
+        height=256, 
+        width=256).images[0]
+
+    template_image = image_to_template(generated_image, logo, button_text, punchline, theme_color)
+
+    return template_image
+
+# Set up Gradio interface
+iface = gr.Interface(
+    fn=generate_template,
+    inputs=[gr.inputs.Image(type="pil", label="Initial Image"), 
+            gr.inputs.Image(type="pil", label="Logo"), 
+            gr.inputs.Textbox(label="Prompt"), 
+            gr.inputs.Textbox(label="Button Text"),
+            gr.inputs.Textbox(label="Punchline"),
+            gr.inputs.ColorPicker(label="Image Color"),
+            gr.inputs.ColorPicker(label="Theme Color")],
+    outputs=[gr.outputs.Image(type="pil")],
+    title="Ad Template Generation Using Diffusion Models Demo",
+    description="Generate ad template based on your inputs using a trained model.",
+    concurrency_limit=2,
+    # examples=[
+    #     []
+    # ]
+)
+
+# Run the interface
+iface.launch()

models/kandinsky-2-2-decoder/model_index.json

@@ -0,0 +1,41 @@
+{
+  "_class_name": "KandinskyV22Img2ImgCombinedPipeline",
+  "_diffusers_version": "0.24.0",
+  "_name_or_path": "kandinsky-community/kandinsky-2-2-decoder",
+  "movq": [
+    "diffusers",
+    "VQModel"
+  ],
+  "prior_image_encoder": [
+    "transformers",
+    "CLIPVisionModelWithProjection"
+  ],
+  "prior_image_processor": [
+    "transformers",
+    "CLIPImageProcessor"
+  ],
+  "prior_prior": [
+    "diffusers",
+    "PriorTransformer"
+  ],
+  "prior_scheduler": [
+    "diffusers",
+    "UnCLIPScheduler"
+  ],
+  "prior_text_encoder": [
+    "transformers",
+    "CLIPTextModelWithProjection"
+  ],
+  "prior_tokenizer": [
+    "transformers",
+    "CLIPTokenizer"
+  ],
+  "scheduler": [
+    "diffusers",
+    "DDPMScheduler"
+  ],
+  "unet": [
+    "diffusers",
+    "UNet2DConditionModel"
+  ]
+}

models/kandinsky-2-2-decoder/movq/config.json

@@ -0,0 +1,34 @@
+{
+  "_class_name": "VQModel",
+  "_diffusers_version": "0.24.0",
+  "_name_or_path": "C:\\Users\\edesa\\.cache\\huggingface\\hub\\models--kandinsky-community--kandinsky-2-2-decoder\\snapshots\\9ae140d347fed8ce6e8bb3005dcc1f48543bb8e3\\movq",
+  "act_fn": "silu",
+  "block_out_channels": [
+    128,
+    256,
+    256,
+    512
+  ],
+  "down_block_types": [
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "AttnDownEncoderBlock2D"
+  ],
+  "in_channels": 3,
+  "latent_channels": 4,
+  "layers_per_block": 2,
+  "norm_num_groups": 32,
+  "norm_type": "spatial",
+  "num_vq_embeddings": 16384,
+  "out_channels": 3,
+  "sample_size": 32,
+  "scaling_factor": 0.18215,
+  "up_block_types": [
+    "AttnUpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D"
+  ],
+  "vq_embed_dim": 4
+}

models/kandinsky-2-2-decoder/movq/diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efb05eb02828bd06b46d4f0fd3bc05fd8c6c0e7803067b20d58c91900e5a820a
+size 135714846

models/kandinsky-2-2-decoder/prior_image_encoder/config.json

@@ -0,0 +1,23 @@
+{
+  "_name_or_path": "C:\\Users\\edesa\\.cache\\huggingface\\hub\\models--kandinsky-community--kandinsky-2-2-prior\\snapshots\\9fc51ad5732afc5d031724219d22e6c42179c5a8\\image_encoder",
+  "architectures": [
+    "CLIPVisionModelWithProjection"
+  ],
+  "attention_dropout": 0.0,
+  "dropout": 0.0,
+  "hidden_act": "gelu",
+  "hidden_size": 1664,
+  "image_size": 224,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "layer_norm_eps": 1e-05,
+  "model_type": "clip_vision_model",
+  "num_attention_heads": 16,
+  "num_channels": 3,
+  "num_hidden_layers": 48,
+  "patch_size": 14,
+  "projection_dim": 1280,
+  "torch_dtype": "float16",
+  "transformers_version": "4.35.2"
+}

models/kandinsky-2-2-decoder/prior_image_encoder/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dabf125eaf7f339da6b900a7a14f8190f02913b7d46cab80e09cd0755c4ce39c
+size 3689910512

models/kandinsky-2-2-decoder/prior_image_processor/preprocessor_config.json

@@ -0,0 +1,29 @@
+{
+  "crop_size": {
+    "height": 224,
+    "width": 224
+  },
+  "do_center_crop": true,
+  "do_convert_rgb": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "feature_extractor_type": "CLIPFeatureExtractor",
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_processor_type": "CLIPImageProcessor",
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "shortest_edge": 224
+  },
+  "use_square_size": false
+}

models/kandinsky-2-2-decoder/prior_prior/config.json

@@ -0,0 +1,20 @@
+{
+  "_class_name": "PriorTransformer",
+  "_diffusers_version": "0.24.0",
+  "_name_or_path": "C:\\Users\\edesa\\.cache\\huggingface\\hub\\models--kandinsky-community--kandinsky-2-2-prior\\snapshots\\9fc51ad5732afc5d031724219d22e6c42179c5a8\\prior",
+  "added_emb_type": "prd",
+  "additional_embeddings": 4,
+  "attention_head_dim": 64,
+  "clip_embed_dim": null,
+  "dropout": 0.0,
+  "embedding_dim": 1280,
+  "embedding_proj_dim": null,
+  "embedding_proj_norm_type": null,
+  "encoder_hid_proj_type": "linear",
+  "norm_in_type": null,
+  "num_attention_heads": 32,
+  "num_embeddings": 77,
+  "num_layers": 20,
+  "time_embed_act_fn": "silu",
+  "time_embed_dim": null
+}

models/kandinsky-2-2-decoder/prior_prior/diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa935a029f07936321ea2922f21cc7904f3260c92785ec61ad6adaa6f2de369b
+size 2052488952

models/kandinsky-2-2-decoder/prior_scheduler/scheduler_config.json

@@ -0,0 +1,10 @@
+{
+  "_class_name": "UnCLIPScheduler",
+  "_diffusers_version": "0.24.0",
+  "beta_schedule": "squaredcos_cap_v2",
+  "clip_sample": true,
+  "clip_sample_range": 10.0,
+  "num_train_timesteps": 1000,
+  "prediction_type": "sample",
+  "variance_type": "fixed_small_log"
+}

models/kandinsky-2-2-decoder/prior_text_encoder/config.json

@@ -0,0 +1,25 @@
+{
+  "_name_or_path": "C:\\Users\\edesa\\.cache\\huggingface\\hub\\models--kandinsky-community--kandinsky-2-2-prior\\snapshots\\9fc51ad5732afc5d031724219d22e6c42179c5a8\\text_encoder",
+  "architectures": [
+    "CLIPTextModelWithProjection"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "dropout": 0.0,
+  "eos_token_id": 2,
+  "hidden_act": "gelu",
+  "hidden_size": 1280,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 5120,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 77,
+  "model_type": "clip_text_model",
+  "num_attention_heads": 20,
+  "num_hidden_layers": 32,
+  "pad_token_id": 1,
+  "projection_dim": 1280,
+  "torch_dtype": "float16",
+  "transformers_version": "4.35.2",
+  "vocab_size": 49408
+}

models/kandinsky-2-2-decoder/prior_text_encoder/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec310df2af79c318e24d20511b601a591ca8cd4f1fce1d8dff822a356bcdb1f4
+size 1389382176

models/kandinsky-2-2-decoder/prior_tokenizer/special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<|endoftext|>",
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

models/kandinsky-2-2-decoder/prior_tokenizer/tokenizer_config.json

@@ -0,0 +1,30 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "49406": {
+      "content": "<|startoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "49407": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<|startoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "do_lower_case": true,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "model_max_length": 77,
+  "pad_token": "<|endoftext|>",
+  "tokenizer_class": "CLIPTokenizer",
+  "unk_token": "<|endoftext|>"
+}

models/kandinsky-2-2-decoder/scheduler/scheduler_config.json

@@ -0,0 +1,19 @@
+{
+  "_class_name": "DDPMScheduler",
+  "_diffusers_version": "0.24.0",
+  "beta_end": 0.012,
+  "beta_schedule": "linear",
+  "beta_start": 0.00085,
+  "clip_sample": true,
+  "clip_sample_range": 1.0,
+  "dynamic_thresholding_ratio": 0.995,
+  "num_train_timesteps": 1000,
+  "prediction_type": "epsilon",
+  "sample_max_value": 1.0,
+  "set_alpha_to_one": false,
+  "steps_offset": 1,
+  "thresholding": false,
+  "timestep_spacing": "leading",
+  "trained_betas": null,
+  "variance_type": "fixed_small"
+}

models/kandinsky-2-2-decoder/unet/config.json

@@ -0,0 +1,68 @@
+{
+  "_class_name": "UNet2DConditionModel",
+  "_diffusers_version": "0.24.0",
+  "_name_or_path": "C:\\Users\\edesa\\.cache\\huggingface\\hub\\models--kandinsky-community--kandinsky-2-2-decoder\\snapshots\\9ae140d347fed8ce6e8bb3005dcc1f48543bb8e3\\unet",
+  "act_fn": "silu",
+  "addition_embed_type": "image",
+  "addition_embed_type_num_heads": 64,
+  "addition_time_embed_dim": null,
+  "attention_head_dim": 64,
+  "attention_type": "default",
+  "block_out_channels": [
+    384,
+    768,
+    1152,
+    1536
+  ],
+  "center_input_sample": false,
+  "class_embed_type": null,
+  "class_embeddings_concat": false,
+  "conv_in_kernel": 3,
+  "conv_out_kernel": 3,
+  "cross_attention_dim": 768,
+  "cross_attention_norm": null,
+  "down_block_types": [
+    "ResnetDownsampleBlock2D",
+    "SimpleCrossAttnDownBlock2D",
+    "SimpleCrossAttnDownBlock2D",
+    "SimpleCrossAttnDownBlock2D"
+  ],
+  "downsample_padding": 1,
+  "dropout": 0.0,
+  "dual_cross_attention": false,
+  "encoder_hid_dim": 1280,
+  "encoder_hid_dim_type": "image_proj",
+  "flip_sin_to_cos": true,
+  "freq_shift": 0,
+  "in_channels": 4,
+  "layers_per_block": 3,
+  "mid_block_only_cross_attention": null,
+  "mid_block_scale_factor": 1,
+  "mid_block_type": "UNetMidBlock2DSimpleCrossAttn",
+  "norm_eps": 1e-05,
+  "norm_num_groups": 32,
+  "num_attention_heads": null,
+  "num_class_embeds": null,
+  "only_cross_attention": false,
+  "out_channels": 8,
+  "projection_class_embeddings_input_dim": null,
+  "resnet_out_scale_factor": 1.0,
+  "resnet_skip_time_act": false,
+  "resnet_time_scale_shift": "scale_shift",
+  "reverse_transformer_layers_per_block": null,
+  "sample_size": 64,
+  "time_cond_proj_dim": null,
+  "time_embedding_act_fn": null,
+  "time_embedding_dim": null,
+  "time_embedding_type": "positional",
+  "timestep_post_act": null,
+  "transformer_layers_per_block": 1,
+  "up_block_types": [
+    "SimpleCrossAttnUpBlock2D",
+    "SimpleCrossAttnUpBlock2D",
+    "SimpleCrossAttnUpBlock2D",
+    "ResnetUpsampleBlock2D"
+  ],
+  "upcast_attention": false,
+  "use_linear_projection": false
+}

models/kandinsky-2-2-decoder/unet/diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd317b5be472665cef98ed37ffe42016a5c4490c282e6f0c2b11205c3953a013
+size 2506194800

requirements.txt

@@ -0,0 +1,8 @@
+diffusers
+transformers
+--extra-index-url https://download.pytorch.org/whl/cu113
+torch
+gradio
+PIL
+requests
+safetensors