Upload fusion_t2i_CLIP_interrogator.ipynb
Browse files
fusion_t2i_CLIP_interrogator.ipynb
ADDED
@@ -0,0 +1,554 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"nbformat": 4,
|
3 |
+
"nbformat_minor": 0,
|
4 |
+
"metadata": {
|
5 |
+
"colab": {
|
6 |
+
"provenance": []
|
7 |
+
},
|
8 |
+
"kernelspec": {
|
9 |
+
"name": "python3",
|
10 |
+
"display_name": "Python 3"
|
11 |
+
},
|
12 |
+
"language_info": {
|
13 |
+
"name": "python"
|
14 |
+
}
|
15 |
+
},
|
16 |
+
"cells": [
|
17 |
+
{
|
18 |
+
"cell_type": "markdown",
|
19 |
+
"source": [
|
20 |
+
"Try this Free online SD 1.5 generator with the results: https://perchance.org/fusion-ai-image-generator\n",
|
21 |
+
"\n",
|
22 |
+
" This Notebook is a Stable-diffusion tool which allows you to find similiar prompts to an existing prompt. It uses the Nearest Neighbor decoder method listed here:https://arxiv.org/pdf/2303.03032"
|
23 |
+
],
|
24 |
+
"metadata": {
|
25 |
+
"id": "cRV2YWomjMBU"
|
26 |
+
}
|
27 |
+
},
|
28 |
+
{
|
29 |
+
"cell_type": "code",
|
30 |
+
"source": [
|
31 |
+
"import os\n",
|
32 |
+
"home_directory = '/content/'\n",
|
33 |
+
"using_Kaggle = os.environ.get('KAGGLE_URL_BASE','')\n",
|
34 |
+
"if using_Kaggle : home_directory = '/kaggle/working/'\n",
|
35 |
+
"%cd {home_directory}\n",
|
36 |
+
"\n",
|
37 |
+
"def fix_bad_symbols(txt):\n",
|
38 |
+
" result = txt\n",
|
39 |
+
" for symbol in ['^', '}', '{' , ')', '(', '[' , ']' , ':' , '=' ]:\n",
|
40 |
+
" result = result.replace(symbol,'\\\\' + symbol)\n",
|
41 |
+
" #------#\n",
|
42 |
+
" return result;\n",
|
43 |
+
"\n",
|
44 |
+
"def my_mkdirs(folder):\n",
|
45 |
+
" if os.path.exists(folder)==False:\n",
|
46 |
+
" os.makedirs(folder)\n",
|
47 |
+
"\n",
|
48 |
+
"#🔸🔹\n",
|
49 |
+
"# Load the data if not already loaded\n",
|
50 |
+
"try:\n",
|
51 |
+
" loaded\n",
|
52 |
+
"except:\n",
|
53 |
+
" from safetensors.torch import load_file , save_file\n",
|
54 |
+
" import json , torch , requests , math\n",
|
55 |
+
" import pandas as pd\n",
|
56 |
+
" from PIL import Image\n",
|
57 |
+
" #----#\n",
|
58 |
+
" %cd {home_directory}\n",
|
59 |
+
" !git clone https://huggingface.co/datasets/codeShare/fusion-t2i-generator-data\n",
|
60 |
+
" loaded = True\n",
|
61 |
+
" %cd {home_directory + 'fusion-t2i-generator-data/'}\n",
|
62 |
+
" !unzip vocab.zip\n",
|
63 |
+
" !unzip reference.zip\n",
|
64 |
+
"#------#\n",
|
65 |
+
"%cd {home_directory + 'fusion-t2i-generator-data/' + 'vocab'}\n",
|
66 |
+
"with open(f'prompts.json', 'r') as f:\n",
|
67 |
+
" data = json.load(f)\n",
|
68 |
+
" _df = pd.DataFrame({'count': data})['count']\n",
|
69 |
+
" prompts = {\n",
|
70 |
+
" key : value for key, value in _df.items()\n",
|
71 |
+
" }\n",
|
72 |
+
"#-------#\n",
|
73 |
+
"%cd {home_directory + 'fusion-t2i-generator-data/' + 'reference'}\n",
|
74 |
+
"with open(f'reference_prompts.json', 'r') as f:\n",
|
75 |
+
" data = json.load(f)\n",
|
76 |
+
" _df = pd.DataFrame({'count': data})['count']\n",
|
77 |
+
" target_prompts = {\n",
|
78 |
+
" key : value for key, value in _df.items()\n",
|
79 |
+
" }\n",
|
80 |
+
"#------#\n",
|
81 |
+
"with open(f'reference_urls.json', 'r') as f:\n",
|
82 |
+
" data = json.load(f)\n",
|
83 |
+
" _df = pd.DataFrame({'count': data})['count']\n",
|
84 |
+
" target_urls = {\n",
|
85 |
+
" key : value for key, value in _df.items()\n",
|
86 |
+
" }\n",
|
87 |
+
"from transformers import AutoTokenizer\n",
|
88 |
+
"tokenizer = AutoTokenizer.from_pretrained(\"openai/clip-vit-large-patch14\", clean_up_tokenization_spaces = False)\n",
|
89 |
+
"from transformers import CLIPProcessor, CLIPModel\n",
|
90 |
+
"processor = CLIPProcessor.from_pretrained(\"openai/clip-vit-large-patch14\" , clean_up_tokenization_spaces = True)\n",
|
91 |
+
"model = CLIPModel.from_pretrained(\"openai/clip-vit-large-patch14\")\n",
|
92 |
+
"logit_scale = model.logit_scale.exp() #logit_scale = 100.00000762939453\n",
|
93 |
+
"\n",
|
94 |
+
"f_add = torch.nn.quantized.FloatFunctional()\n",
|
95 |
+
"\n",
|
96 |
+
"index = 0\n",
|
97 |
+
"%cd {home_directory + 'fusion-t2i-generator-data/' + 'vocab'}\n",
|
98 |
+
"\n",
|
99 |
+
"vocab_encodings = torch.load('vocab_encodings.pt', weights_only=False)\n",
|
100 |
+
"for key in vocab_encodings:\n",
|
101 |
+
" index = index + 1;\n",
|
102 |
+
"#------#\n",
|
103 |
+
"NUM_VOCAB_ITEMS = index\n",
|
104 |
+
"\n",
|
105 |
+
"index = 0\n",
|
106 |
+
"%cd {home_directory + 'fusion-t2i-generator-data/' + 'reference'}\n",
|
107 |
+
"for key in torch.load('reference_text_and_image_encodings.pt', weights_only=False):\n",
|
108 |
+
" index = index + 1;\n",
|
109 |
+
"#------#\n",
|
110 |
+
"NUM_REFERENCE_ITEMS = index\n"
|
111 |
+
],
|
112 |
+
"metadata": {
|
113 |
+
"id": "TC5lMJrS1HCC"
|
114 |
+
},
|
115 |
+
"execution_count": null,
|
116 |
+
"outputs": []
|
117 |
+
},
|
118 |
+
{
|
119 |
+
"cell_type": "code",
|
120 |
+
"source": [
|
121 |
+
"# @title \t⚄ Use a pre-encoded prompt + image pair from the fusion gen (note: NSFW!)\n",
|
122 |
+
"# @markdown Choose a pre-encoded reference\n",
|
123 |
+
"index = 457 # @param {type:\"slider\", min:0, max:1666, step:1}\n",
|
124 |
+
"PROMPT_INDEX = index\n",
|
125 |
+
"prompt = target_prompts[f'{PROMPT_INDEX}']\n",
|
126 |
+
"url = target_urls[f'{PROMPT_INDEX}']\n",
|
127 |
+
"if url.find('perchance')>-1:\n",
|
128 |
+
" image = Image.open(requests.get(url, stream=True).raw)\n",
|
129 |
+
"#------#\n",
|
130 |
+
"# @markdown ⚖️ 🖼️ encoding <-----?-----> 📝 encoding </div> <br>\n",
|
131 |
+
"C = 0.5 # @param {type:\"slider\", min:0, max:1, step:0.01}\n",
|
132 |
+
"log_strength = 1 # @param {type:\"slider\", min:-5, max:5, step:0.01}\n",
|
133 |
+
"prompt_strength = math.pow(10 ,log_strength-1)\n",
|
134 |
+
"reference = torch.zeros(768)\n",
|
135 |
+
"%cd {home_directory + 'fusion-t2i-generator-data/' + 'reference'}\n",
|
136 |
+
"references = torch.load('reference_text_and_image_encodings.pt' , weights_only=False)\n",
|
137 |
+
"reference = torch.add(reference, C * references[index][0].dequantize())\n",
|
138 |
+
"reference = torch.add(reference, (1-C) * references[index][1].dequantize())\n",
|
139 |
+
"references = ''\n",
|
140 |
+
"# @markdown -----------\n",
|
141 |
+
"# @markdown 📝➕ Enhance similarity to prompt(s)\n",
|
142 |
+
"POS = '' # @param {type:'string' ,placeholder:'item1 , item2 , ...'}\n",
|
143 |
+
"log_strength = 1.06 # @param {type:\"slider\", min:-5, max:5, step:0.01}\n",
|
144 |
+
"pos_strength = math.pow(10 ,log_strength-1)\n",
|
145 |
+
"for _POS in POS.split(','):\n",
|
146 |
+
" inputs = tokenizer(text = _POS.strip(), truncation = True , padding=True, return_tensors=\"pt\")\n",
|
147 |
+
" text_features_POS = model.get_text_features(**inputs)\n",
|
148 |
+
" text_features_POS = text_features_POS/text_features_POS.norm(p=2, dim=-1, keepdim=True)\n",
|
149 |
+
" reference = torch.add(reference, pos_strength * text_features_POS)\n",
|
150 |
+
"# @markdown -----------\n",
|
151 |
+
"\n",
|
152 |
+
"# @markdown 🚫 Penalize similarity to prompt(s)\n",
|
153 |
+
"NEG = '' # @param {type:'string' , placeholder:'item1 , item2 , ...'}\n",
|
154 |
+
"log_strength = 1 # @param {type:\"slider\", min:-5, max:5, step:0.01}\n",
|
155 |
+
"neg_strength = math.pow(10 ,log_strength-1)\n",
|
156 |
+
"for _NEG in NEG.split(','):\n",
|
157 |
+
" inputs = tokenizer(text = _NEG.strip(), truncation = True , padding=True, return_tensors=\"pt\")\n",
|
158 |
+
" text_features_NEG = model.get_text_features(**inputs)\n",
|
159 |
+
" text_features_NEG = text_features_NEG/text_features_NEG.norm(p=2, dim=-1, keepdim=True)\n",
|
160 |
+
" reference = torch.add(reference, (-1) * neg_strength * text_features_NEG)\n",
|
161 |
+
"# @markdown -----------\n",
|
162 |
+
"# @markdown ⏩ Skip item(s) containing the word(s)\n",
|
163 |
+
"SKIP = 'futa ' # @param {type:'string' , placeholder:'item1 , item2 , ...'}\n",
|
164 |
+
"def isBlacklisted(txt):\n",
|
165 |
+
" if txt.strip().isnumeric(): return True\n",
|
166 |
+
" if blacklist.strip() == '': return False\n",
|
167 |
+
" for item in list(blacklist.split(',')):\n",
|
168 |
+
" if item.strip() == '' : continue\n",
|
169 |
+
" if txt.find(item.strip())> -1 : return True\n",
|
170 |
+
" #------#\n",
|
171 |
+
" return False\n",
|
172 |
+
"# @markdown -----------\n",
|
173 |
+
"# @markdown 🔍 How similar should the results be?\n",
|
174 |
+
"list_size = 1000 # @param {type:'number'}\n",
|
175 |
+
"start_at_index = 1 # @param {type:'number'}\n",
|
176 |
+
"# @markdown -----------\n",
|
177 |
+
"# @markdown Repeat output N times\n",
|
178 |
+
"N = 7 # @param {type:\"slider\", min:0, max:20, step:1}\n",
|
179 |
+
"# @markdown -----------\n",
|
180 |
+
"# @markdown ⚙️ Run the script?\n",
|
181 |
+
"run_script = True # @param {type:\"boolean\"}\n",
|
182 |
+
"enable = run_script\n",
|
183 |
+
"if (enable):\n",
|
184 |
+
" reference = reference/reference.norm(p=2, dim=-1, keepdim=True)\n",
|
185 |
+
" %cd {home_directory + 'fusion-t2i-generator-data/' + 'vocab'}\n",
|
186 |
+
" sims = torch.matmul(vocab_encodings.dequantize(),reference.t())\n",
|
187 |
+
" sorted , indices = torch.sort(sims,dim=0 , descending=True)\n",
|
188 |
+
"\n",
|
189 |
+
" average = torch.zeros(768)\n",
|
190 |
+
" for key in range(NUM_VOCAB_ITEMS):\n",
|
191 |
+
" if (key>=start_at_index and key < start_at_index + list_size):\n",
|
192 |
+
" average = torch.add(average, vocab_encodings[key].dequantize())\n",
|
193 |
+
" if (key>=start_at_index + list_size) : break\n",
|
194 |
+
" average = average * (1/max(1, list_size))\n",
|
195 |
+
" average = average/average.norm(p=2, dim=-1, keepdim=True)\n",
|
196 |
+
" print(average.norm(p=2, dim=-1, keepdim=True))\n",
|
197 |
+
" average = average.clone().detach();\n",
|
198 |
+
" variance = torch.zeros(1)\n",
|
199 |
+
" for key in range(NUM_VOCAB_ITEMS):\n",
|
200 |
+
" if (key>=start_at_index and key < start_at_index + list_size):\n",
|
201 |
+
" #dot product\n",
|
202 |
+
"\n",
|
203 |
+
" difference_to_average = 100 * (torch.ones(1) - torch.dot(average[0]\n",
|
204 |
+
" , vocab_encodings[key].dequantize()[0])/average.norm(p=2, dim=-1, keepdim=True))\n",
|
205 |
+
"\n",
|
206 |
+
" variance = torch.add(variance, difference_to_average * difference_to_average)\n",
|
207 |
+
" if (key>=start_at_index + list_size) : break\n",
|
208 |
+
" #--------#\n",
|
209 |
+
" variance = variance * (1/max(1, list_size))\n",
|
210 |
+
" variance= variance.clone().detach();\n",
|
211 |
+
" print(f'The variance for the selected range is {math.sqrt(variance.item())} units from average')\n",
|
212 |
+
"#---#\n",
|
213 |
+
" output = '{'\n",
|
214 |
+
" for _index in range(list_size):\n",
|
215 |
+
" output = output + prompts[f'{indices[min(_index+start_at_index,NUM_VOCAB_ITEMS-1)].item()}'] + '|'\n",
|
216 |
+
" #---------#\n",
|
217 |
+
" output = (output + '}').replace('|}' , '}</w>')\n",
|
218 |
+
" for iter in range(N):\n",
|
219 |
+
" print(output)\n",
|
220 |
+
"#-------#\n",
|
221 |
+
"image or print('No image found')"
|
222 |
+
],
|
223 |
+
"metadata": {
|
224 |
+
"id": "NqL_I3ZSrISq"
|
225 |
+
},
|
226 |
+
"execution_count": null,
|
227 |
+
"outputs": []
|
228 |
+
},
|
229 |
+
{
|
230 |
+
"cell_type": "code",
|
231 |
+
"source": [
|
232 |
+
"# Check the average value for this set\n",
|
233 |
+
"sims = torch.matmul(vocab_encodings.dequantize(),average.t())\n",
|
234 |
+
"sorted , indices = torch.sort(sims,dim=0 , descending=True)\n",
|
235 |
+
"for index in range(10):\n",
|
236 |
+
" print(prompts[f'{indices[index].item()}'])"
|
237 |
+
],
|
238 |
+
"metadata": {
|
239 |
+
"id": "XNHz0hfhHRUu"
|
240 |
+
},
|
241 |
+
"execution_count": 113,
|
242 |
+
"outputs": []
|
243 |
+
},
|
244 |
+
{
|
245 |
+
"cell_type": "code",
|
246 |
+
"source": [
|
247 |
+
"# @title ⚙️📝 Print the results (Advanced)\n",
|
248 |
+
"list_size = 1000 # @param {type:'number'}\n",
|
249 |
+
"start_at_index = 0 # @param {type:'number'}\n",
|
250 |
+
"print_Similarity = True # @param {type:\"boolean\"}\n",
|
251 |
+
"print_Prompts = True # @param {type:\"boolean\"}\n",
|
252 |
+
"print_Descriptions = True # @param {type:\"boolean\"}\n",
|
253 |
+
"compact_Output = True # @param {type:\"boolean\"}\n",
|
254 |
+
"newline_Separator = False # @param {type:\"boolean\"}\n",
|
255 |
+
"\n",
|
256 |
+
"import random\n",
|
257 |
+
"# @markdown -----------\n",
|
258 |
+
"# @markdown Mix with...\n",
|
259 |
+
"list_size2 = 1000 # @param {type:'number'}\n",
|
260 |
+
"start_at_index2 = 10000 # @param {type:'number'}\n",
|
261 |
+
"rate_percent = 0 # @param {type:\"slider\", min:0, max:100, step:1}\n",
|
262 |
+
"\n",
|
263 |
+
"# @markdown -----------\n",
|
264 |
+
"# @markdown Repeat output N times\n",
|
265 |
+
"N = 6 # @param {type:\"slider\", min:0, max:10, step:1}\n",
|
266 |
+
"\n",
|
267 |
+
"# title Show the 100 most similiar suffix and prefix text-encodings to the text encoding\n",
|
268 |
+
"RANGE = list_size\n",
|
269 |
+
"separator = '|'\n",
|
270 |
+
"if newline_Separator : separator = separator + '\\n'\n",
|
271 |
+
"\n",
|
272 |
+
"_prompts = ''\n",
|
273 |
+
"_sims = ''\n",
|
274 |
+
"for _index in range(start_at_index + RANGE):\n",
|
275 |
+
" if _index < start_at_index : continue\n",
|
276 |
+
" index = indices[_index].item()\n",
|
277 |
+
"\n",
|
278 |
+
" prompt = prompts[f'{index}']\n",
|
279 |
+
" if rate_percent >= random.randint(0,100) : prompt = prompts[f'{random.randint(start_at_index2 , start_at_index2 + list_size2)}']\n",
|
280 |
+
"\n",
|
281 |
+
" #Remove duplicates\n",
|
282 |
+
" if _prompts.find(prompt + separator)<=-1:\n",
|
283 |
+
" _sims = _sims + f'{round(100*sims[index].item(), 2)} %' + separator\n",
|
284 |
+
" #-------#\n",
|
285 |
+
" _prompts = _prompts.replace(prompt + separator,'')\n",
|
286 |
+
" _prompts = _prompts + prompt + separator\n",
|
287 |
+
" #------#\n",
|
288 |
+
"#------#\n",
|
289 |
+
"__prompts = fix_bad_symbols(__prompts)\n",
|
290 |
+
"__prompts = ('{' + _prompts + '}').replace(separator + '}', '}')\n",
|
291 |
+
"__sims = ('{' + _sims + '}').replace(separator + '}', '}')\n",
|
292 |
+
"#------#\n",
|
293 |
+
"\n",
|
294 |
+
"if(not print_Prompts): __prompts = ''\n",
|
295 |
+
"if(not print_Similarity): __sims = ''\n",
|
296 |
+
"\n",
|
297 |
+
"if(not compact_Output):\n",
|
298 |
+
" if(print_Descriptions):\n",
|
299 |
+
" print(f'The {start_at_index}-{start_at_index + RANGE} most similiar items to prompt : \\n\\n ')\n",
|
300 |
+
" for i in range(N) : print(__prompts)\n",
|
301 |
+
" print(f'The {start_at_index}-{start_at_index + RANGE} similarity % for items : \\n\\n' + __sims)\n",
|
302 |
+
" print('')\n",
|
303 |
+
" else:\n",
|
304 |
+
" for i in range(N) : print(__prompts)\n",
|
305 |
+
"else:\n",
|
306 |
+
" for i in range(N) : print(__prompts)\n",
|
307 |
+
"#-------#"
|
308 |
+
],
|
309 |
+
"metadata": {
|
310 |
+
"id": "EdBiAguJO9aX",
|
311 |
+
"cellView": "form"
|
312 |
+
},
|
313 |
+
"execution_count": null,
|
314 |
+
"outputs": []
|
315 |
+
},
|
316 |
+
{
|
317 |
+
"cell_type": "markdown",
|
318 |
+
"source": [
|
319 |
+
"The savefile can be used here : https://perchance.org/fusion-ai-image-generator"
|
320 |
+
],
|
321 |
+
"metadata": {
|
322 |
+
"id": "JldNmWy1iyvK"
|
323 |
+
}
|
324 |
+
},
|
325 |
+
{
|
326 |
+
"cell_type": "code",
|
327 |
+
"source": [
|
328 |
+
"# @title \t⚄ Create fusion-generator .json savefile from result\n",
|
329 |
+
"filename = 'blank.json'\n",
|
330 |
+
"path = '/content/text-to-image-prompts/fusion/'\n",
|
331 |
+
"\n",
|
332 |
+
"print(f'reading {filename}....')\n",
|
333 |
+
"_index = 0\n",
|
334 |
+
"%cd {path}\n",
|
335 |
+
"with open(f'{filename}', 'r') as f:\n",
|
336 |
+
" data = json.load(f)\n",
|
337 |
+
"#------#\n",
|
338 |
+
"_df = pd.DataFrame({'count': data})['count']\n",
|
339 |
+
"_savefile = {\n",
|
340 |
+
" key : value for key, value in _df.items()\n",
|
341 |
+
"}\n",
|
342 |
+
"#------#\n",
|
343 |
+
"from safetensors.torch import load_file\n",
|
344 |
+
"import json , os , torch\n",
|
345 |
+
"import pandas as pd\n",
|
346 |
+
"#----#\n",
|
347 |
+
"def my_mkdirs(folder):\n",
|
348 |
+
" if os.path.exists(folder)==False:\n",
|
349 |
+
" os.makedirs(folder)\n",
|
350 |
+
"#------#\n",
|
351 |
+
"savefile_prompt = ''\n",
|
352 |
+
"for i in range(N) : savefile_prompt = savefile_prompt + ' ' + __prompts\n",
|
353 |
+
"_savefile['main'] = savefile_prompt.replace('\\n', ' ').replace(' ', ' ').replace(' ', ' ')\n",
|
354 |
+
"#------#\n",
|
355 |
+
"save_filename = f'fusion_C05_X7_1000_{PROMPT_INDEX}.json'\n",
|
356 |
+
"output_folder = '/content/output/savefiles/'\n",
|
357 |
+
"my_mkdirs(output_folder)\n",
|
358 |
+
"#-----#\n",
|
359 |
+
"%cd {output_folder}\n",
|
360 |
+
"print(f'Saving segment {save_filename} to {output_folder}...')\n",
|
361 |
+
"with open(save_filename, 'w') as f:\n",
|
362 |
+
" json.dump(_savefile, f)\n"
|
363 |
+
],
|
364 |
+
"metadata": {
|
365 |
+
"id": "Q7vpNAXQilbf",
|
366 |
+
"cellView": "form"
|
367 |
+
},
|
368 |
+
"execution_count": null,
|
369 |
+
"outputs": []
|
370 |
+
},
|
371 |
+
{
|
372 |
+
"cell_type": "code",
|
373 |
+
"source": [
|
374 |
+
"# @title \t⚄ Create a savefile-set from the entire range of pre-encoded items\n",
|
375 |
+
"\n",
|
376 |
+
"# @markdown 📥 Load the data (only required one time)\n",
|
377 |
+
"load_the_data = True # @param {type:\"boolean\"}\n",
|
378 |
+
"\n",
|
379 |
+
"import math\n",
|
380 |
+
"from safetensors.torch import load_file\n",
|
381 |
+
"import json , os , torch\n",
|
382 |
+
"import pandas as pd\n",
|
383 |
+
"from PIL import Image\n",
|
384 |
+
"import requests\n",
|
385 |
+
"\n",
|
386 |
+
"def my_mkdirs(folder):\n",
|
387 |
+
" if os.path.exists(folder)==False:\n",
|
388 |
+
" os.makedirs(folder)\n",
|
389 |
+
"\n",
|
390 |
+
"# @markdown ⚖️ Set the value for C in the reference <br> <br> sim = C* text_enc + image_enc*(1-C) <br><br>\n",
|
391 |
+
"\n",
|
392 |
+
"C = 0.5 # @param {type:\"slider\", min:0, max:1, step:0.01}\n",
|
393 |
+
"\n",
|
394 |
+
"# @markdown 🚫 Penalize similarity to this prompt(optional)\n",
|
395 |
+
"if(load_the_data):\n",
|
396 |
+
" target_prompts , target_text_encodings , urls , target_image_encodings , NUM_ITEMS = getPromptsAndLinks('/content/text-to-image-prompts/fusion')\n",
|
397 |
+
" from transformers import AutoTokenizer\n",
|
398 |
+
" tokenizer = AutoTokenizer.from_pretrained(\"openai/clip-vit-large-patch14\", clean_up_tokenization_spaces = False)\n",
|
399 |
+
" from transformers import CLIPProcessor, CLIPModel\n",
|
400 |
+
" processor = CLIPProcessor.from_pretrained(\"openai/clip-vit-large-patch14\" , clean_up_tokenization_spaces = True)\n",
|
401 |
+
" model = CLIPModel.from_pretrained(\"openai/clip-vit-large-patch14\")\n",
|
402 |
+
" logit_scale = model.logit_scale.exp() #logit_scale = 100.00000762939453\n",
|
403 |
+
"#---------#\n",
|
404 |
+
"\n",
|
405 |
+
"filename = 'blank.json'\n",
|
406 |
+
"path = '/content/text-to-image-prompts/fusion/'\n",
|
407 |
+
"print(f'reading {filename}....')\n",
|
408 |
+
"_index = 0\n",
|
409 |
+
"%cd {path}\n",
|
410 |
+
"with open(f'{filename}', 'r') as f:\n",
|
411 |
+
" data = json.load(f)\n",
|
412 |
+
"#------#\n",
|
413 |
+
"_df = pd.DataFrame({'count': data})['count']\n",
|
414 |
+
"_blank = {\n",
|
415 |
+
" key : value for key, value in _df.items()\n",
|
416 |
+
"}\n",
|
417 |
+
"#------#\n",
|
418 |
+
"\n",
|
419 |
+
"root_savefile_name = 'fusion_C05_X7'\n",
|
420 |
+
"\n",
|
421 |
+
"%cd /content/\n",
|
422 |
+
"output_folder = '/content/output/savefiles/'\n",
|
423 |
+
"my_mkdirs(output_folder)\n",
|
424 |
+
"my_mkdirs('/content/output2/savefiles/')\n",
|
425 |
+
"my_mkdirs('/content/output3/savefiles/')\n",
|
426 |
+
"my_mkdirs('/content/output4/savefiles/')\n",
|
427 |
+
"my_mkdirs('/content/output5/savefiles/')\n",
|
428 |
+
"my_mkdirs('/content/output6/savefiles/')\n",
|
429 |
+
"my_mkdirs('/content/output7/savefiles/')\n",
|
430 |
+
"my_mkdirs('/content/output8/savefiles/')\n",
|
431 |
+
"my_mkdirs('/content/output9/savefiles/')\n",
|
432 |
+
"my_mkdirs('/content/output10/savefiles/')\n",
|
433 |
+
"my_mkdirs('/content/output11/savefiles/')\n",
|
434 |
+
"my_mkdirs('/content/output12/savefiles/')\n",
|
435 |
+
"my_mkdirs('/content/output13/savefiles/')\n",
|
436 |
+
"\n",
|
437 |
+
"\n",
|
438 |
+
"NEG = '' # @param {type:'string'}\n",
|
439 |
+
"strength = 1 # @param {type:\"slider\", min:-5, max:5, step:0.1}\n",
|
440 |
+
"\n",
|
441 |
+
"for index in range(1667):\n",
|
442 |
+
"\n",
|
443 |
+
" PROMPT_INDEX = index\n",
|
444 |
+
" prompt = target_prompts[f'{index}']\n",
|
445 |
+
" url = urls[f'{index}']\n",
|
446 |
+
" if url.find('perchance')>-1:\n",
|
447 |
+
" image = Image.open(requests.get(url, stream=True).raw)\n",
|
448 |
+
" else: continue #print(\"(No image for this ID)\")\n",
|
449 |
+
"\n",
|
450 |
+
" print(f\"no. {PROMPT_INDEX} : '{prompt}'\")\n",
|
451 |
+
" text_features_A = target_text_encodings[f'{index}']\n",
|
452 |
+
" image_features_A = target_image_encodings[f'{index}']\n",
|
453 |
+
" # text-similarity\n",
|
454 |
+
" sims = C * torch.matmul(text_tensor, text_features_A.t())\n",
|
455 |
+
"\n",
|
456 |
+
" neg_sims = 0*sims\n",
|
457 |
+
" if(NEG != ''):\n",
|
458 |
+
" # Get text features for user input\n",
|
459 |
+
" inputs = tokenizer(text = NEG, padding=True, return_tensors=\"pt\")\n",
|
460 |
+
" text_features_NEG = model.get_text_features(**inputs)\n",
|
461 |
+
" text_features_NEG = text_features_A/text_features_A.norm(p=2, dim=-1, keepdim=True)\n",
|
462 |
+
" # text-similarity\n",
|
463 |
+
" neg_sims = strength*torch.matmul(text_tensor, text_features_NEG.t())\n",
|
464 |
+
" #------#\n",
|
465 |
+
"\n",
|
466 |
+
" # plus image-similarity\n",
|
467 |
+
" sims = sims + (1-C) * torch.matmul(text_tensor, image_features_A.t()) * logit_scale\n",
|
468 |
+
"\n",
|
469 |
+
" # minus NEG-similarity\n",
|
470 |
+
" sims = sims - neg_sims\n",
|
471 |
+
"\n",
|
472 |
+
" # Sort the items\n",
|
473 |
+
" sorted , indices = torch.sort(sims,dim=0 , descending=True)\n",
|
474 |
+
"\n",
|
475 |
+
" # @markdown Repeat output N times\n",
|
476 |
+
" RANGE = 1000\n",
|
477 |
+
" NUM_CHUNKS = 10+\n",
|
478 |
+
" separator = '|'\n",
|
479 |
+
" _savefiles = {}\n",
|
480 |
+
" #-----#\n",
|
481 |
+
" for chunk in range(NUM_CHUNKS):\n",
|
482 |
+
" if chunk=<10:continue\n",
|
483 |
+
" start_at_index = chunk * RANGE\n",
|
484 |
+
" _prompts = ''\n",
|
485 |
+
" for _index in range(start_at_index + RANGE):\n",
|
486 |
+
" if _index < start_at_index : continue\n",
|
487 |
+
" index = indices[_index].item()\n",
|
488 |
+
" prompt = prompts[f'{index}']\n",
|
489 |
+
" _prompts = _prompts.replace(prompt + separator,'')\n",
|
490 |
+
" _prompts = _prompts + prompt + separator\n",
|
491 |
+
" #------#\n",
|
492 |
+
" _prompts = fix_bad_symbols(_prompts)\n",
|
493 |
+
" _prompts = ('{' + _prompts + '}').replace(separator + '}', '}')\n",
|
494 |
+
" _savefiles[f'{chunk}'] = _prompts\n",
|
495 |
+
" #---------#\n",
|
496 |
+
" save_filename = f'{root_savefile_name}_{start_at_index + RANGE}_{PROMPT_INDEX}.json'\n",
|
497 |
+
"\n",
|
498 |
+
"\n",
|
499 |
+
" if (chunk=<20 && chunk>10): %cd '/content/output2/savefiles/'\n",
|
500 |
+
" if (chunk<=30 && chunk>20): %cd '/content/output3/savefiles/'\n",
|
501 |
+
" if (chunk=<40 && chunk>30): %cd '/content/output4/savefiles/'\n",
|
502 |
+
" if (chunk<=50 && chunk>40): %cd '/content/output5/savefiles/'\n",
|
503 |
+
" if (chunk=<60 && chunk>50): %cd '/content/output6/savefiles/'\n",
|
504 |
+
" if (chunk<=70 && chunk>60): %cd '/content/output7/savefiles/'\n",
|
505 |
+
" if (chunk=<80 && chunk>70): %cd '/content/output8/savefiles/'\n",
|
506 |
+
" if (chunk<=90 && chunk>80): %cd '/content/output9/savefiles/'\n",
|
507 |
+
" if (chunk=<100 && chunk>90): %cd '/content/output10/savefiles/'\n",
|
508 |
+
" if (chunk<=110 && chunk>100): %cd '/content/output11/savefiles/'\n",
|
509 |
+
" if (chunk=<120 && chunk>110): %cd '/content/output12/savefiles/'\n",
|
510 |
+
" if (chunk<=130 && chunk>120): %cd '/content/output13/savefiles/'\n",
|
511 |
+
"\n",
|
512 |
+
"\n",
|
513 |
+
" #------#\n",
|
514 |
+
" print(f'Saving savefile {save_filename} to {output_folder}...')\n",
|
515 |
+
" with open(save_filename, 'w') as f:\n",
|
516 |
+
" json.dump(_savefiles, f)\n",
|
517 |
+
" #---------#\n",
|
518 |
+
" continue\n",
|
519 |
+
"#-----------#"
|
520 |
+
],
|
521 |
+
"metadata": {
|
522 |
+
"id": "x1uAVXZEoL0T",
|
523 |
+
"cellView": "form"
|
524 |
+
},
|
525 |
+
"execution_count": null,
|
526 |
+
"outputs": []
|
527 |
+
},
|
528 |
+
{
|
529 |
+
"cell_type": "code",
|
530 |
+
"source": [
|
531 |
+
"# Determine if this notebook is running on Colab or Kaggle\n",
|
532 |
+
"#Use https://www.kaggle.com/ if Google Colab GPU is busy\n",
|
533 |
+
"home_directory = '/content/'\n",
|
534 |
+
"using_Kaggle = os.environ.get('KAGGLE_URL_BASE','')\n",
|
535 |
+
"if using_Kaggle : home_directory = '/kaggle/working/'\n",
|
536 |
+
"%cd {home_directory}\n",
|
537 |
+
"#-------#\n",
|
538 |
+
"\n",
|
539 |
+
"# @title Download the text_encodings as .zip\n",
|
540 |
+
"import os\n",
|
541 |
+
"%cd {home_directory}\n",
|
542 |
+
"#os.remove(f'{home_directory}results.zip')\n",
|
543 |
+
"root_output_folder = home_directory + 'output/'\n",
|
544 |
+
"zip_dest = f'/content/results.zip' #drive/MyDrive\n",
|
545 |
+
"!zip -r {zip_dest} {root_output_folder}"
|
546 |
+
],
|
547 |
+
"metadata": {
|
548 |
+
"id": "zivBNrw9uSVD"
|
549 |
+
},
|
550 |
+
"execution_count": null,
|
551 |
+
"outputs": []
|
552 |
+
}
|
553 |
+
]
|
554 |
+
}
|