Upload fusion_t2i_CLIP_interrogator.ipynb
Browse files
Google Colab Notebooks/fusion_t2i_CLIP_interrogator.ipynb
CHANGED
@@ -29,7 +29,8 @@
|
|
29 |
"cell_type": "code",
|
30 |
"execution_count": null,
|
31 |
"metadata": {
|
32 |
-
"id": "UEYEdzjgOEOE"
|
|
|
33 |
},
|
34 |
"outputs": [],
|
35 |
"source": [
|
@@ -47,28 +48,22 @@
|
|
47 |
"\n",
|
48 |
"def fix_bad_symbols(txt):\n",
|
49 |
" result = txt\n",
|
50 |
-
" for symbol in ['}', '{' , ')', '(', '[' , ']' , ':' , '='
|
51 |
" result = result.replace(symbol,'\\\\' + symbol)\n",
|
52 |
-
" for symbol in ['^']:\n",
|
53 |
-
" result = result.replace(symbol,'')\n",
|
54 |
" #------#\n",
|
55 |
-
" result = result.replace('\\\\|','|').replace(' |',' |')\n",
|
56 |
" return result;\n",
|
57 |
"\n",
|
58 |
"\n",
|
59 |
"def getPrompts(_path, separator):\n",
|
60 |
-
"\n",
|
61 |
" path = _path + '/text'\n",
|
62 |
" path_enc = _path + '/text_encodings'\n",
|
63 |
" #-----#\n",
|
64 |
" index = 0\n",
|
65 |
-
" file_index = 0\n",
|
66 |
" prompts = {}\n",
|
67 |
" text_encodings = {}\n",
|
68 |
" _text_encodings = {}\n",
|
69 |
" #-----#\n",
|
70 |
" for filename in os.listdir(f'{path}'):\n",
|
71 |
-
"\n",
|
72 |
" print(f'reading {filename}....')\n",
|
73 |
" _index = 0\n",
|
74 |
" %cd {path}\n",
|
@@ -79,27 +74,13 @@
|
|
79 |
" _prompts = {\n",
|
80 |
" key : value for key, value in _df.items()\n",
|
81 |
" }\n",
|
|
|
|
|
|
|
82 |
" for key in _prompts:\n",
|
83 |
" _index = int(key)\n",
|
84 |
" value = _prompts[key]\n",
|
85 |
-
"\n",
|
86 |
-
" #Read the 'header' file in the JSON\n",
|
87 |
-
" if _index <= 0 :\n",
|
88 |
-
" _NUM_ITEMS = int(value)\n",
|
89 |
-
" prompts[f'{index}'] = _prompts[f'{_index}'] + separator\n",
|
90 |
-
" index = index + 1\n",
|
91 |
-
" continue\n",
|
92 |
-
" if _index <= 1 :\n",
|
93 |
-
" _file_name = f'{value}'\n",
|
94 |
-
" %cd {path_enc}\n",
|
95 |
-
" _text_encodings = load_file(f'{_file_name}.safetensors')\n",
|
96 |
-
" #Store text_encodings for the header items\n",
|
97 |
-
" text_encodings[f'{index-1}'] = _text_encodings[f'{_index-1}']\n",
|
98 |
-
" text_encodings[f'{index}'] = _text_encodings[f'{_index}']\n",
|
99 |
-
" #------#\n",
|
100 |
-
" prompts[f'{index}'] = _prompts[f'{_index}'] + separator\n",
|
101 |
-
" index = index + 1\n",
|
102 |
-
" continue\n",
|
103 |
" #------#\n",
|
104 |
" #Read the text_encodings + prompts\n",
|
105 |
" text_encodings[f'{index}'] = _text_encodings[f'{_index}']\n",
|
@@ -108,8 +89,6 @@
|
|
108 |
" continue\n",
|
109 |
" #-------#\n",
|
110 |
" #--------#\n",
|
111 |
-
" #_text_encodings.close() #close the text_encodings file\n",
|
112 |
-
" file_index = file_index + 1\n",
|
113 |
" #----------#\n",
|
114 |
" NUM_ITEMS = index -1\n",
|
115 |
" return prompts , text_encodings , NUM_ITEMS\n",
|
@@ -255,8 +234,7 @@
|
|
255 |
" #----------#\n",
|
256 |
" NUM_ITEMS = index -1\n",
|
257 |
" return prompts , text_encodings , urls , image_encodings , NUM_ITEMS\n",
|
258 |
-
"
|
259 |
-
"\n"
|
260 |
]
|
261 |
},
|
262 |
{
|
@@ -264,35 +242,57 @@
|
|
264 |
"source": [
|
265 |
"# @title 📚 Select items to sample from\n",
|
266 |
"\n",
|
267 |
-
"prompt_features =
|
268 |
-
"civitai_blue_set =
|
269 |
-
"
|
270 |
-
"
|
271 |
-
"
|
272 |
-
"prefix = False # @param {\"type\":\"boolean\",\"placeholder\":\"🔸\"}\n",
|
273 |
-
"emojis = False # @param {\"type\":\"boolean\",\"placeholder\":\"😃\"}\n",
|
274 |
"#------#\n",
|
275 |
-
"
|
276 |
-
"first_names =
|
277 |
-
"last_names =
|
278 |
-
"celebs =
|
279 |
"#-------#\n",
|
280 |
"danbooru_tags = True # @param {\"type\":\"boolean\",\"placeholder\":\"🎀\"}\n",
|
281 |
-
"lyrics =
|
282 |
-
"tripple_nouns =
|
283 |
"#-----#\n",
|
284 |
-
"female_fullnames =
|
285 |
"debug = False\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
286 |
"#------#\n",
|
287 |
"prompts = {}\n",
|
288 |
"text_encodings = {}\n",
|
289 |
"nA = 0\n",
|
290 |
"#--------#\n",
|
291 |
"\n",
|
292 |
-
"if
|
293 |
-
" url = '/content/text-to-image-prompts/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
294 |
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
295 |
"\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
296 |
"if tripple_nouns:\n",
|
297 |
" url = '/content/text-to-image-prompts/nouns'\n",
|
298 |
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
@@ -344,16 +344,6 @@
|
|
344 |
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
345 |
"#--------#\n",
|
346 |
"\n",
|
347 |
-
"if civitai_yellow_set:\n",
|
348 |
-
" url = '/content/text-to-image-prompts/civitai-prompts/yellow'\n",
|
349 |
-
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
350 |
-
"#--------#\n",
|
351 |
-
"\n",
|
352 |
-
"if artby_prompts:\n",
|
353 |
-
" url = '/content/text-to-image-prompts/artby'\n",
|
354 |
-
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
355 |
-
"#--------#\n",
|
356 |
-
"\n",
|
357 |
"if suffix :\n",
|
358 |
" tmp = '/content/text-to-image-prompts/vocab/text_encodings/suffix/'\n",
|
359 |
" for item in ['common','average','rare','weird','exotic'] :\n",
|
@@ -381,10 +371,13 @@
|
|
381 |
"text_tensor = torch.zeros(NUM_VOCAB_ITEMS,768)\n",
|
382 |
"for index in range(NUM_VOCAB_ITEMS):\n",
|
383 |
" text_tensor[index] = text_encodings[f'{index}']\n",
|
384 |
-
"#---------#\n"
|
|
|
|
|
385 |
],
|
386 |
"metadata": {
|
387 |
-
"id": "CF53WIAKObg3"
|
|
|
388 |
},
|
389 |
"execution_count": null,
|
390 |
"outputs": []
|
@@ -394,7 +387,7 @@
|
|
394 |
"source": [
|
395 |
"# @title \t⚄ Use a pre-encoded prompt + image pair from the fusion gen (note: NSFW!)\n",
|
396 |
"# @markdown 🖼️ Choose a pre-encoded reference\n",
|
397 |
-
"index =
|
398 |
"PROMPT_INDEX = index\n",
|
399 |
"\n",
|
400 |
"import math\n",
|
@@ -411,7 +404,7 @@
|
|
411 |
"neg_strength = math.pow(10 ,log_strength-1)\n",
|
412 |
"\n",
|
413 |
"# @markdown ⏩ Skip item(s) containing the word\n",
|
414 |
-
"SKIP = '
|
415 |
"\n",
|
416 |
"# @markdown ⚖️ How important is the prompt vs. the image? <br> <br>\n",
|
417 |
"# @markdown <div style = \"font-size: 10px; \"> sim_ref = (10^(log_strength-1)) * ( C* text_encoding + image_encoding*(1-C) ) </div> <br>\n",
|
@@ -609,7 +602,8 @@
|
|
609 |
"#------#"
|
610 |
],
|
611 |
"metadata": {
|
612 |
-
"id": "XW3914T8O2uf"
|
|
|
613 |
},
|
614 |
"execution_count": null,
|
615 |
"outputs": []
|
@@ -680,7 +674,8 @@
|
|
680 |
"#-------#"
|
681 |
],
|
682 |
"metadata": {
|
683 |
-
"id": "EdBiAguJO9aX"
|
|
|
684 |
},
|
685 |
"execution_count": null,
|
686 |
"outputs": []
|
@@ -734,7 +729,8 @@
|
|
734 |
" json.dump(_savefile, f)\n"
|
735 |
],
|
736 |
"metadata": {
|
737 |
-
"id": "Q7vpNAXQilbf"
|
|
|
738 |
},
|
739 |
"execution_count": null,
|
740 |
"outputs": []
|
@@ -744,19 +740,26 @@
|
|
744 |
"source": [
|
745 |
"# @title \t⚄ Create a savefile-set from the entire range of pre-encoded items\n",
|
746 |
"\n",
|
747 |
-
"#image_index = 0 # @param {type:'number'}\n",
|
748 |
"# @markdown 📥 Load the data (only required one time)\n",
|
749 |
"load_the_data = True # @param {type:\"boolean\"}\n",
|
750 |
"\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
751 |
"# @markdown ⚖️ Set the value for C in the reference <br> <br> sim = C* text_enc + image_enc*(1-C) <br><br>\n",
|
752 |
"\n",
|
753 |
"C = 0.5 # @param {type:\"slider\", min:0, max:1, step:0.01}\n",
|
754 |
"\n",
|
755 |
"# @markdown 🚫 Penalize similarity to this prompt(optional)\n",
|
756 |
-
"\n",
|
757 |
"if(load_the_data):\n",
|
758 |
-
" from PIL import Image\n",
|
759 |
-
" import requests\n",
|
760 |
" target_prompts , target_text_encodings , urls , target_image_encodings , NUM_ITEMS = getPromptsAndLinks('/content/text-to-image-prompts/fusion')\n",
|
761 |
" from transformers import AutoTokenizer\n",
|
762 |
" tokenizer = AutoTokenizer.from_pretrained(\"openai/clip-vit-large-patch14\", clean_up_tokenization_spaces = False)\n",
|
@@ -780,16 +783,31 @@
|
|
780 |
"}\n",
|
781 |
"#------#\n",
|
782 |
"\n",
|
783 |
-
"root_savefile_name = '
|
|
|
|
|
784 |
"output_folder = '/content/output/savefiles/'\n",
|
785 |
"my_mkdirs(output_folder)\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
786 |
"NEG = '' # @param {type:'string'}\n",
|
787 |
"strength = 1 # @param {type:\"slider\", min:-5, max:5, step:0.1}\n",
|
788 |
"\n",
|
789 |
"for index in range(1667):\n",
|
790 |
"\n",
|
791 |
" PROMPT_INDEX = index\n",
|
792 |
-
"\n",
|
793 |
" prompt = target_prompts[f'{index}']\n",
|
794 |
" url = urls[f'{index}']\n",
|
795 |
" if url.find('perchance')>-1:\n",
|
@@ -797,110 +815,79 @@
|
|
797 |
" else: continue #print(\"(No image for this ID)\")\n",
|
798 |
"\n",
|
799 |
" print(f\"no. {PROMPT_INDEX} : '{prompt}'\")\n",
|
|
|
|
|
|
|
|
|
800 |
"\n",
|
801 |
-
"\n",
|
802 |
-
" if(
|
803 |
-
"
|
804 |
-
"
|
805 |
-
"\n",
|
|
|
806 |
" # text-similarity\n",
|
807 |
-
"
|
808 |
-
"
|
809 |
-
" neg_sims = 0*sims\n",
|
810 |
-
" if(NEG != ''):\n",
|
811 |
-
"\n",
|
812 |
-
" # Get text features for user input\n",
|
813 |
-
" inputs = tokenizer(text = NEG, padding=True, return_tensors=\"pt\")\n",
|
814 |
-
" text_features_NEG = model.get_text_features(**inputs)\n",
|
815 |
-
" text_features_NEG = text_features_A/text_features_A.norm(p=2, dim=-1, keepdim=True)\n",
|
816 |
-
"\n",
|
817 |
-
" # text-similarity\n",
|
818 |
-
" neg_sims = strength*torch.matmul(text_tensor, text_features_NEG.t())\n",
|
819 |
-
" #------#\n",
|
820 |
-
"\n",
|
821 |
-
" # plus image-similarity\n",
|
822 |
-
" sims = sims + (1-C) * torch.matmul(text_tensor, image_features_A.t()) * logit_scale\n",
|
823 |
-
"\n",
|
824 |
-
" # minus NEG-similarity\n",
|
825 |
-
" sims = sims - neg_sims\n",
|
826 |
-
"\n",
|
827 |
-
" # Sort the items\n",
|
828 |
-
" sorted , indices = torch.sort(sims,dim=0 , descending=True)\n",
|
829 |
-
"\n",
|
830 |
-
" # @title ⚙️📝 Print the results (Advanced)\n",
|
831 |
-
" list_size = 1000 # param {type:'number'}\n",
|
832 |
-
" start_at_index = 0 # param {type:'number'}\n",
|
833 |
-
" print_Similarity = True # param {type:\"boolean\"}\n",
|
834 |
-
" print_Prompts = True # param {type:\"boolean\"}\n",
|
835 |
-
" print_Prefix = True # param {type:\"boolean\"}\n",
|
836 |
-
" print_Descriptions = True # param {type:\"boolean\"}\n",
|
837 |
-
" compact_Output = True # param {type:\"boolean\"}\n",
|
838 |
-
"\n",
|
839 |
-
" # @markdown -----------\n",
|
840 |
-
" # @markdown ⚙️📝 Printing options\n",
|
841 |
-
" newline_Separator = False # @param {type:\"boolean\"}\n",
|
842 |
"\n",
|
843 |
-
"
|
844 |
-
"
|
845 |
-
" start_at_index2 = 10000 # param {type:'number'}\n",
|
846 |
-
" rate_percent = 0 # param {type:\"slider\", min:0, max:100, step:1}\n",
|
847 |
"\n",
|
848 |
-
"
|
849 |
-
"
|
850 |
"\n",
|
851 |
-
"
|
852 |
-
"
|
853 |
-
" separator = '|'\n",
|
854 |
-
" if newline_Separator : separator = separator + '\\n'\n",
|
855 |
"\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
856 |
" _prompts = ''\n",
|
857 |
-
" _sims = ''\n",
|
858 |
" for _index in range(start_at_index + RANGE):\n",
|
859 |
" if _index < start_at_index : continue\n",
|
860 |
" index = indices[_index].item()\n",
|
861 |
-
"\n",
|
862 |
" prompt = prompts[f'{index}']\n",
|
863 |
-
" if rate_percent >= random.randint(0,100) : prompt = prompts[f'{random.randint(start_at_index2 , start_at_index2 + list_size2)}']\n",
|
864 |
-
"\n",
|
865 |
-
" #Remove duplicates\n",
|
866 |
-
" if _prompts.find(prompt + separator)<=-1:\n",
|
867 |
-
" _sims = _sims + f'{round(100*sims[index].item(), 2)} %' + separator\n",
|
868 |
-
" #-------#\n",
|
869 |
" _prompts = _prompts.replace(prompt + separator,'')\n",
|
870 |
" _prompts = _prompts + prompt + separator\n",
|
871 |
-
" #------#\n",
|
872 |
" #------#\n",
|
873 |
" _prompts = fix_bad_symbols(_prompts)\n",
|
874 |
-
"
|
875 |
-
"
|
876 |
-
"
|
877 |
-
"
|
878 |
-
"
|
879 |
-
"
|
880 |
-
"
|
881 |
-
"
|
882 |
-
"
|
883 |
-
"
|
884 |
-
"
|
885 |
-
"
|
886 |
-
"
|
887 |
-
"
|
888 |
-
"
|
889 |
-
"
|
|
|
|
|
|
|
|
|
890 |
" #------#\n",
|
891 |
-
" save_filename = f'{root_savefile_name}{PROMPT_INDEX}.json'\n",
|
892 |
-
" #-----#\n",
|
893 |
-
" %cd {output_folder}\n",
|
894 |
" print(f'Saving savefile {save_filename} to {output_folder}...')\n",
|
895 |
" with open(save_filename, 'w') as f:\n",
|
896 |
-
" json.dump(
|
897 |
" #---------#\n",
|
898 |
" continue\n",
|
899 |
-
"
|
900 |
-
"\n"
|
901 |
],
|
902 |
"metadata": {
|
903 |
-
"id": "x1uAVXZEoL0T"
|
|
|
904 |
},
|
905 |
"execution_count": null,
|
906 |
"outputs": []
|
@@ -925,53 +912,8 @@
|
|
925 |
"!zip -r {zip_dest} {root_output_folder}"
|
926 |
],
|
927 |
"metadata": {
|
928 |
-
"id": "zivBNrw9uSVD"
|
929 |
-
|
930 |
-
"execution_count": null,
|
931 |
-
"outputs": []
|
932 |
-
},
|
933 |
-
{
|
934 |
-
"cell_type": "code",
|
935 |
-
"source": [
|
936 |
-
"output_folder = '/content/output/fusion-gen-savefiles/'\n",
|
937 |
-
"index = 0\n",
|
938 |
-
"path = '/content/text-to-image-prompts/fusion-gen-savefiles'\n",
|
939 |
-
"\n",
|
940 |
-
"def my_mkdirs(folder):\n",
|
941 |
-
" if os.path.exists(folder)==False:\n",
|
942 |
-
" os.makedirs(folder)\n",
|
943 |
-
"\n",
|
944 |
-
"my_mkdirs(output_folder)\n",
|
945 |
-
"for filename in os.listdir(f'{path}'):\n",
|
946 |
-
" if filename.find('fusion_C05_X7_1000_')<=-1: continue\n",
|
947 |
-
" print(f'reading {filename}...')\n",
|
948 |
-
" %cd {path}\n",
|
949 |
-
" with open(f'{filename}', 'r') as f:\n",
|
950 |
-
" data = json.load(f)\n",
|
951 |
-
" _df = pd.DataFrame({'count': data})['count']\n",
|
952 |
-
" _savefile = {\n",
|
953 |
-
" key : value for key, value in _df.items()\n",
|
954 |
-
" }\n",
|
955 |
-
"\n",
|
956 |
-
" _savefile2 = {}\n",
|
957 |
-
"\n",
|
958 |
-
" for key in _savefile:\n",
|
959 |
-
" _savefile2[key] = _savefile[key]\n",
|
960 |
-
" if(key == \"_main\") :\n",
|
961 |
-
" _savefile2[key] = \"Prompt input only ✏️\"\n",
|
962 |
-
" print(\"changed\")\n",
|
963 |
-
" #----------#\n",
|
964 |
-
"\n",
|
965 |
-
" save_filename = f'fusion_C05_X7_1000_{index}.json'\n",
|
966 |
-
" index = index + 1\n",
|
967 |
-
"\n",
|
968 |
-
" %cd {output_folder}\n",
|
969 |
-
" print(f'Saving savefile {save_filename} to {output_folder}...')\n",
|
970 |
-
" with open(save_filename, 'w') as f:\n",
|
971 |
-
" json.dump(_savefile2, f)"
|
972 |
-
],
|
973 |
-
"metadata": {
|
974 |
-
"id": "A3ASDnO3IzSL"
|
975 |
},
|
976 |
"execution_count": null,
|
977 |
"outputs": []
|
|
|
29 |
"cell_type": "code",
|
30 |
"execution_count": null,
|
31 |
"metadata": {
|
32 |
+
"id": "UEYEdzjgOEOE",
|
33 |
+
"cellView": "form"
|
34 |
},
|
35 |
"outputs": [],
|
36 |
"source": [
|
|
|
48 |
"\n",
|
49 |
"def fix_bad_symbols(txt):\n",
|
50 |
" result = txt\n",
|
51 |
+
" for symbol in ['^', '}', '{' , ')', '(', '[' , ']' , ':' , '=' ]:\n",
|
52 |
" result = result.replace(symbol,'\\\\' + symbol)\n",
|
|
|
|
|
53 |
" #------#\n",
|
|
|
54 |
" return result;\n",
|
55 |
"\n",
|
56 |
"\n",
|
57 |
"def getPrompts(_path, separator):\n",
|
|
|
58 |
" path = _path + '/text'\n",
|
59 |
" path_enc = _path + '/text_encodings'\n",
|
60 |
" #-----#\n",
|
61 |
" index = 0\n",
|
|
|
62 |
" prompts = {}\n",
|
63 |
" text_encodings = {}\n",
|
64 |
" _text_encodings = {}\n",
|
65 |
" #-----#\n",
|
66 |
" for filename in os.listdir(f'{path}'):\n",
|
|
|
67 |
" print(f'reading {filename}....')\n",
|
68 |
" _index = 0\n",
|
69 |
" %cd {path}\n",
|
|
|
74 |
" _prompts = {\n",
|
75 |
" key : value for key, value in _df.items()\n",
|
76 |
" }\n",
|
77 |
+
" _file_name = _prompts[f'{1}']\n",
|
78 |
+
" %cd {path_enc}\n",
|
79 |
+
" _text_encodings = load_file(f'{_file_name}.safetensors')\n",
|
80 |
" for key in _prompts:\n",
|
81 |
" _index = int(key)\n",
|
82 |
" value = _prompts[key]\n",
|
83 |
+
" if _index<2:continue\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
84 |
" #------#\n",
|
85 |
" #Read the text_encodings + prompts\n",
|
86 |
" text_encodings[f'{index}'] = _text_encodings[f'{_index}']\n",
|
|
|
89 |
" continue\n",
|
90 |
" #-------#\n",
|
91 |
" #--------#\n",
|
|
|
|
|
92 |
" #----------#\n",
|
93 |
" NUM_ITEMS = index -1\n",
|
94 |
" return prompts , text_encodings , NUM_ITEMS\n",
|
|
|
234 |
" #----------#\n",
|
235 |
" NUM_ITEMS = index -1\n",
|
236 |
" return prompts , text_encodings , urls , image_encodings , NUM_ITEMS\n",
|
237 |
+
"#--------#"
|
|
|
238 |
]
|
239 |
},
|
240 |
{
|
|
|
242 |
"source": [
|
243 |
"# @title 📚 Select items to sample from\n",
|
244 |
"\n",
|
245 |
+
"prompt_features = True # @param {\"type\":\"boolean\",\"placeholder\":\"🦜\"}\n",
|
246 |
+
"civitai_blue_set = True # @param {\"type\":\"boolean\",\"placeholder\":\"📘\"}\n",
|
247 |
+
"suffix = True # @param {\"type\":\"boolean\",\"placeholder\":\"🔹\"}\n",
|
248 |
+
"prefix = True # @param {\"type\":\"boolean\",\"placeholder\":\"🔸\"}\n",
|
249 |
+
"emojis = True # @param {\"type\":\"boolean\",\"placeholder\":\"😃\"}\n",
|
|
|
|
|
250 |
"#------#\n",
|
251 |
+
"\n",
|
252 |
+
"first_names = True # @param {\"type\":\"boolean\",\"placeholder\":\"🔹\"}\n",
|
253 |
+
"last_names = True # @param {\"type\":\"boolean\",\"placeholder\":\"🔸\"}\n",
|
254 |
+
"celebs = True # @param {\"type\":\"boolean\",\"placeholder\":\"🆔👨\"}\n",
|
255 |
"#-------#\n",
|
256 |
"danbooru_tags = True # @param {\"type\":\"boolean\",\"placeholder\":\"🎀\"}\n",
|
257 |
+
"lyrics = True # @param {\"type\":\"boolean\",\"placeholder\":\"🎼\"}\n",
|
258 |
+
"tripple_nouns = True # @param {\"type\":\"boolean\",\"placeholder\":\"🎼\"}\n",
|
259 |
"#-----#\n",
|
260 |
+
"female_fullnames = True # @param {\"type\":\"boolean\",\"placeholder\":\"😃\"}\n",
|
261 |
"debug = False\n",
|
262 |
+
"\n",
|
263 |
+
"civitai_red_set = True # @param {\"type\":\"boolean\",\"placeholder\":\"📕\"}\n",
|
264 |
+
"e621 = True # @param {\"type\":\"boolean\",\"placeholder\":\"😃\"}\n",
|
265 |
+
"prefix_suffix_pairs = True # @param {\"type\":\"boolean\",\"placeholder\":\"😃\"}\n",
|
266 |
+
"suffix_tripple = True # @param {\"type\":\"boolean\",\"placeholder\":\"😃\"}\n",
|
267 |
+
"suffix_quad = True # @param {\"type\":\"boolean\",\"placeholder\":\"😃\"}\n",
|
268 |
"#------#\n",
|
269 |
"prompts = {}\n",
|
270 |
"text_encodings = {}\n",
|
271 |
"nA = 0\n",
|
272 |
"#--------#\n",
|
273 |
"\n",
|
274 |
+
"if civitai_red_set:\n",
|
275 |
+
" url = '/content/text-to-image-prompts/civitai-prompts/red'\n",
|
276 |
+
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
277 |
+
"\n",
|
278 |
+
"if e621:\n",
|
279 |
+
" url = '/content/text-to-image-prompts/e621'\n",
|
280 |
+
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
281 |
+
"\n",
|
282 |
+
"if prefix_suffix_pairs:\n",
|
283 |
+
" url = '/content/text-to-image-prompts/prefix_suffix_pairs'\n",
|
284 |
+
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
285 |
+
"\n",
|
286 |
+
"if suffix_tripple:\n",
|
287 |
+
" url = '/content/text-to-image-prompts/suffix_tripple'\n",
|
288 |
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
289 |
"\n",
|
290 |
+
"if suffix_quad:\n",
|
291 |
+
" url = '/content/text-to-image-prompts/suffix_quad'\n",
|
292 |
+
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
293 |
+
"\n",
|
294 |
+
"\n",
|
295 |
+
"\n",
|
296 |
"if tripple_nouns:\n",
|
297 |
" url = '/content/text-to-image-prompts/nouns'\n",
|
298 |
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
|
|
344 |
" prompts , text_encodings, nA = append_from_url(prompts , text_encodings, nA , url , '')\n",
|
345 |
"#--------#\n",
|
346 |
"\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
347 |
"if suffix :\n",
|
348 |
" tmp = '/content/text-to-image-prompts/vocab/text_encodings/suffix/'\n",
|
349 |
" for item in ['common','average','rare','weird','exotic'] :\n",
|
|
|
371 |
"text_tensor = torch.zeros(NUM_VOCAB_ITEMS,768)\n",
|
372 |
"for index in range(NUM_VOCAB_ITEMS):\n",
|
373 |
" text_tensor[index] = text_encodings[f'{index}']\n",
|
374 |
+
"#---------#\n",
|
375 |
+
"print(f\"Done loading vocabulary for the interrogator\")\n",
|
376 |
+
"print(f\"Vocab size is : {NUM_VOCAB_ITEMS} items\")"
|
377 |
],
|
378 |
"metadata": {
|
379 |
+
"id": "CF53WIAKObg3",
|
380 |
+
"cellView": "form"
|
381 |
},
|
382 |
"execution_count": null,
|
383 |
"outputs": []
|
|
|
387 |
"source": [
|
388 |
"# @title \t⚄ Use a pre-encoded prompt + image pair from the fusion gen (note: NSFW!)\n",
|
389 |
"# @markdown 🖼️ Choose a pre-encoded reference\n",
|
390 |
+
"index = 682 # @param {type:\"slider\", min:0, max:1666, step:1}\n",
|
391 |
"PROMPT_INDEX = index\n",
|
392 |
"\n",
|
393 |
"import math\n",
|
|
|
404 |
"neg_strength = math.pow(10 ,log_strength-1)\n",
|
405 |
"\n",
|
406 |
"# @markdown ⏩ Skip item(s) containing the word\n",
|
407 |
+
"SKIP = 'futa ' # @param {type:'string' , placeholder:'item1 , item2 , ...'}\n",
|
408 |
"\n",
|
409 |
"# @markdown ⚖️ How important is the prompt vs. the image? <br> <br>\n",
|
410 |
"# @markdown <div style = \"font-size: 10px; \"> sim_ref = (10^(log_strength-1)) * ( C* text_encoding + image_encoding*(1-C) ) </div> <br>\n",
|
|
|
602 |
"#------#"
|
603 |
],
|
604 |
"metadata": {
|
605 |
+
"id": "XW3914T8O2uf",
|
606 |
+
"cellView": "form"
|
607 |
},
|
608 |
"execution_count": null,
|
609 |
"outputs": []
|
|
|
674 |
"#-------#"
|
675 |
],
|
676 |
"metadata": {
|
677 |
+
"id": "EdBiAguJO9aX",
|
678 |
+
"cellView": "form"
|
679 |
},
|
680 |
"execution_count": null,
|
681 |
"outputs": []
|
|
|
729 |
" json.dump(_savefile, f)\n"
|
730 |
],
|
731 |
"metadata": {
|
732 |
+
"id": "Q7vpNAXQilbf",
|
733 |
+
"cellView": "form"
|
734 |
},
|
735 |
"execution_count": null,
|
736 |
"outputs": []
|
|
|
740 |
"source": [
|
741 |
"# @title \t⚄ Create a savefile-set from the entire range of pre-encoded items\n",
|
742 |
"\n",
|
|
|
743 |
"# @markdown 📥 Load the data (only required one time)\n",
|
744 |
"load_the_data = True # @param {type:\"boolean\"}\n",
|
745 |
"\n",
|
746 |
+
"import math\n",
|
747 |
+
"from safetensors.torch import load_file\n",
|
748 |
+
"import json , os , torch\n",
|
749 |
+
"import pandas as pd\n",
|
750 |
+
"from PIL import Image\n",
|
751 |
+
"import requests\n",
|
752 |
+
"\n",
|
753 |
+
"def my_mkdirs(folder):\n",
|
754 |
+
" if os.path.exists(folder)==False:\n",
|
755 |
+
" os.makedirs(folder)\n",
|
756 |
+
"\n",
|
757 |
"# @markdown ⚖️ Set the value for C in the reference <br> <br> sim = C* text_enc + image_enc*(1-C) <br><br>\n",
|
758 |
"\n",
|
759 |
"C = 0.5 # @param {type:\"slider\", min:0, max:1, step:0.01}\n",
|
760 |
"\n",
|
761 |
"# @markdown 🚫 Penalize similarity to this prompt(optional)\n",
|
|
|
762 |
"if(load_the_data):\n",
|
|
|
|
|
763 |
" target_prompts , target_text_encodings , urls , target_image_encodings , NUM_ITEMS = getPromptsAndLinks('/content/text-to-image-prompts/fusion')\n",
|
764 |
" from transformers import AutoTokenizer\n",
|
765 |
" tokenizer = AutoTokenizer.from_pretrained(\"openai/clip-vit-large-patch14\", clean_up_tokenization_spaces = False)\n",
|
|
|
783 |
"}\n",
|
784 |
"#------#\n",
|
785 |
"\n",
|
786 |
+
"root_savefile_name = 'fusion_C05_X7'\n",
|
787 |
+
"\n",
|
788 |
+
"%cd /content/\n",
|
789 |
"output_folder = '/content/output/savefiles/'\n",
|
790 |
"my_mkdirs(output_folder)\n",
|
791 |
+
"my_mkdirs('/content/output2/savefiles/')\n",
|
792 |
+
"my_mkdirs('/content/output3/savefiles/')\n",
|
793 |
+
"my_mkdirs('/content/output4/savefiles/')\n",
|
794 |
+
"my_mkdirs('/content/output5/savefiles/')\n",
|
795 |
+
"my_mkdirs('/content/output6/savefiles/')\n",
|
796 |
+
"my_mkdirs('/content/output7/savefiles/')\n",
|
797 |
+
"my_mkdirs('/content/output8/savefiles/')\n",
|
798 |
+
"my_mkdirs('/content/output9/savefiles/')\n",
|
799 |
+
"my_mkdirs('/content/output10/savefiles/')\n",
|
800 |
+
"my_mkdirs('/content/output11/savefiles/')\n",
|
801 |
+
"my_mkdirs('/content/output12/savefiles/')\n",
|
802 |
+
"my_mkdirs('/content/output13/savefiles/')\n",
|
803 |
+
"\n",
|
804 |
+
"\n",
|
805 |
"NEG = '' # @param {type:'string'}\n",
|
806 |
"strength = 1 # @param {type:\"slider\", min:-5, max:5, step:0.1}\n",
|
807 |
"\n",
|
808 |
"for index in range(1667):\n",
|
809 |
"\n",
|
810 |
" PROMPT_INDEX = index\n",
|
|
|
811 |
" prompt = target_prompts[f'{index}']\n",
|
812 |
" url = urls[f'{index}']\n",
|
813 |
" if url.find('perchance')>-1:\n",
|
|
|
815 |
" else: continue #print(\"(No image for this ID)\")\n",
|
816 |
"\n",
|
817 |
" print(f\"no. {PROMPT_INDEX} : '{prompt}'\")\n",
|
818 |
+
" text_features_A = target_text_encodings[f'{index}']\n",
|
819 |
+
" image_features_A = target_image_encodings[f'{index}']\n",
|
820 |
+
" # text-similarity\n",
|
821 |
+
" sims = C * torch.matmul(text_tensor, text_features_A.t())\n",
|
822 |
"\n",
|
823 |
+
" neg_sims = 0*sims\n",
|
824 |
+
" if(NEG != ''):\n",
|
825 |
+
" # Get text features for user input\n",
|
826 |
+
" inputs = tokenizer(text = NEG, padding=True, return_tensors=\"pt\")\n",
|
827 |
+
" text_features_NEG = model.get_text_features(**inputs)\n",
|
828 |
+
" text_features_NEG = text_features_A/text_features_A.norm(p=2, dim=-1, keepdim=True)\n",
|
829 |
" # text-similarity\n",
|
830 |
+
" neg_sims = strength*torch.matmul(text_tensor, text_features_NEG.t())\n",
|
831 |
+
" #------#\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
832 |
"\n",
|
833 |
+
" # plus image-similarity\n",
|
834 |
+
" sims = sims + (1-C) * torch.matmul(text_tensor, image_features_A.t()) * logit_scale\n",
|
|
|
|
|
835 |
"\n",
|
836 |
+
" # minus NEG-similarity\n",
|
837 |
+
" sims = sims - neg_sims\n",
|
838 |
"\n",
|
839 |
+
" # Sort the items\n",
|
840 |
+
" sorted , indices = torch.sort(sims,dim=0 , descending=True)\n",
|
|
|
|
|
841 |
"\n",
|
842 |
+
" # @markdown Repeat output N times\n",
|
843 |
+
" RANGE = 1000\n",
|
844 |
+
" NUM_CHUNKS = 10+\n",
|
845 |
+
" separator = '|'\n",
|
846 |
+
" _savefiles = {}\n",
|
847 |
+
" #-----#\n",
|
848 |
+
" for chunk in range(NUM_CHUNKS):\n",
|
849 |
+
" if chunk=<10:continue\n",
|
850 |
+
" start_at_index = chunk * RANGE\n",
|
851 |
" _prompts = ''\n",
|
|
|
852 |
" for _index in range(start_at_index + RANGE):\n",
|
853 |
" if _index < start_at_index : continue\n",
|
854 |
" index = indices[_index].item()\n",
|
|
|
855 |
" prompt = prompts[f'{index}']\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
856 |
" _prompts = _prompts.replace(prompt + separator,'')\n",
|
857 |
" _prompts = _prompts + prompt + separator\n",
|
|
|
858 |
" #------#\n",
|
859 |
" _prompts = fix_bad_symbols(_prompts)\n",
|
860 |
+
" _prompts = ('{' + _prompts + '}').replace(separator + '}', '}')\n",
|
861 |
+
" _savefiles[f'{chunk}'] = _prompts\n",
|
862 |
+
" #---------#\n",
|
863 |
+
" save_filename = f'{root_savefile_name}_{start_at_index + RANGE}_{PROMPT_INDEX}.json'\n",
|
864 |
+
"\n",
|
865 |
+
"\n",
|
866 |
+
" if (chunk=<20 && chunk>10): %cd '/content/output2/savefiles/'\n",
|
867 |
+
" if (chunk<=30 && chunk>20): %cd '/content/output3/savefiles/'\n",
|
868 |
+
" if (chunk=<40 && chunk>30): %cd '/content/output4/savefiles/'\n",
|
869 |
+
" if (chunk<=50 && chunk>40): %cd '/content/output5/savefiles/'\n",
|
870 |
+
" if (chunk=<60 && chunk>50): %cd '/content/output6/savefiles/'\n",
|
871 |
+
" if (chunk<=70 && chunk>60): %cd '/content/output7/savefiles/'\n",
|
872 |
+
" if (chunk=<80 && chunk>70): %cd '/content/output8/savefiles/'\n",
|
873 |
+
" if (chunk<=90 && chunk>80): %cd '/content/output9/savefiles/'\n",
|
874 |
+
" if (chunk=<100 && chunk>90): %cd '/content/output10/savefiles/'\n",
|
875 |
+
" if (chunk<=110 && chunk>100): %cd '/content/output11/savefiles/'\n",
|
876 |
+
" if (chunk=<120 && chunk>110): %cd '/content/output12/savefiles/'\n",
|
877 |
+
" if (chunk<=130 && chunk>120): %cd '/content/output13/savefiles/'\n",
|
878 |
+
"\n",
|
879 |
+
"\n",
|
880 |
" #------#\n",
|
|
|
|
|
|
|
881 |
" print(f'Saving savefile {save_filename} to {output_folder}...')\n",
|
882 |
" with open(save_filename, 'w') as f:\n",
|
883 |
+
" json.dump(_savefiles, f)\n",
|
884 |
" #---------#\n",
|
885 |
" continue\n",
|
886 |
+
"#-----------#"
|
|
|
887 |
],
|
888 |
"metadata": {
|
889 |
+
"id": "x1uAVXZEoL0T",
|
890 |
+
"cellView": "form"
|
891 |
},
|
892 |
"execution_count": null,
|
893 |
"outputs": []
|
|
|
912 |
"!zip -r {zip_dest} {root_output_folder}"
|
913 |
],
|
914 |
"metadata": {
|
915 |
+
"id": "zivBNrw9uSVD",
|
916 |
+
"cellView": "form"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
917 |
},
|
918 |
"execution_count": null,
|
919 |
"outputs": []
|