Upload fusion_t2i_CLIP_interrogator.ipynb
Browse files
Google Colab Jupyter Notebooks/fusion_t2i_CLIP_interrogator.ipynb
CHANGED
@@ -175,6 +175,92 @@
|
|
175 |
"id": "Xf9zoq-Za3wi"
|
176 |
}
|
177 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
178 |
{
|
179 |
"cell_type": "code",
|
180 |
"source": [
|
@@ -289,92 +375,6 @@
|
|
289 |
"execution_count": null,
|
290 |
"outputs": []
|
291 |
},
|
292 |
-
{
|
293 |
-
"cell_type": "markdown",
|
294 |
-
"source": [
|
295 |
-
"# 🖼️ Image encoders (optional)"
|
296 |
-
],
|
297 |
-
"metadata": {
|
298 |
-
"id": "f9_AcquM7AYZ"
|
299 |
-
}
|
300 |
-
},
|
301 |
-
{
|
302 |
-
"cell_type": "code",
|
303 |
-
"source": [
|
304 |
-
"# @title ⚄ 📷💭 Use pre-encoded image+prompt pair\n",
|
305 |
-
"loaded_ref = False\n",
|
306 |
-
"try:\n",
|
307 |
-
" ref\n",
|
308 |
-
" loaded_ref = True\n",
|
309 |
-
"except:ref = torch.zeros(dim).to(dtype = dot_dtype)\n",
|
310 |
-
"if loaded_ref : prev_ref = ref.clone().detach()\n",
|
311 |
-
"\n",
|
312 |
-
"try:prompt\n",
|
313 |
-
"except: prompt = ''\n",
|
314 |
-
"\n",
|
315 |
-
"# @markdown 🖼️+📝 Choose a pre-encoded reference (note: some results are NSFW!)\n",
|
316 |
-
"index = 596 # @param {type:\"slider\", min:0, max:1666, step:1}\n",
|
317 |
-
"PROMPT_INDEX = index\n",
|
318 |
-
"prompt = target_prompts[f'{PROMPT_INDEX}']\n",
|
319 |
-
"url = target_urls[f'{PROMPT_INDEX}']\n",
|
320 |
-
"if url.find('perchance')>-1:\n",
|
321 |
-
" image = Image.open(requests.get(url, stream=True).raw)\n",
|
322 |
-
"#------#\n",
|
323 |
-
"%cd {home_directory + 'fusion-t2i-generator-data/' + 'reference'}\n",
|
324 |
-
"references = torch.load('reference_text_and_image_encodings.pt' , weights_only=False)\n",
|
325 |
-
"# @markdown ⚖️ 🖼️ encoding <-----?-----> 📝 encoding </div> <br>\n",
|
326 |
-
"C = 0.3 # @param {type:\"slider\", min:0, max:1, step:0.01}\n",
|
327 |
-
"log_strength = 1 # @param {type:\"slider\", min:-5, max:5, step:0.01}\n",
|
328 |
-
"method = 'Add to existing ref' # @param [\"Refresh\" , \"Add to existing ref\" , \"Subtract from existing ref\" , \"Do nothing\"]\n",
|
329 |
-
"image_size = 0.57 # @param {type:\"slider\", min:0, max:1, step:0.01}\n",
|
330 |
-
"show_encoding = True # @param {type:\"boolean\"}\n",
|
331 |
-
"\n",
|
332 |
-
"if(not method == 'Do nothing'):\n",
|
333 |
-
" if method == 'Refresh': ref = torch.zeros(dim).to(dtype = dot_dtype)\n",
|
334 |
-
" if method == 'Subtract from existing ref':\n",
|
335 |
-
" ref = torch.sub(ref, math.pow(10 ,log_strength-1) * C * references[index][0].dequantize().to(dtype = torch.float32))\n",
|
336 |
-
" ref = torch.sub(ref, math.pow(10 ,log_strength-1) * (1-C) * references[index][1].dequantize().to(dtype = torch.float32))\n",
|
337 |
-
" else:\n",
|
338 |
-
" ref = torch.add(ref, math.pow(10 ,log_strength-1) * C * references[index][0].dequantize().to(dtype = torch.float32))\n",
|
339 |
-
" ref = torch.add(ref, math.pow(10 ,log_strength-1) * (1-C) * references[index][1].dequantize().to(dtype = torch.float32))\n",
|
340 |
-
" #---------#\n",
|
341 |
-
" references = '' # Clear up memory\n",
|
342 |
-
" ref = ref/ref.norm(p=2, dim=-1, keepdim=True)\n",
|
343 |
-
" ref = ref.clone().detach()\n",
|
344 |
-
" #------#\n",
|
345 |
-
" # create figure\n",
|
346 |
-
" fig = plt.figure(figsize=(10*image_size, 10*image_size))\n",
|
347 |
-
" fig.patch.set_facecolor((56/255,56/255,56/255))\n",
|
348 |
-
" rows = 1\n",
|
349 |
-
" columns = 1\n",
|
350 |
-
" if show_encoding: columns = columns+1\n",
|
351 |
-
" if show_encoding and loaded_ref : columns = columns+1\n",
|
352 |
-
" fig.add_subplot(rows, columns, 1)\n",
|
353 |
-
" plt.imshow(image)\n",
|
354 |
-
" plt.axis('off')\n",
|
355 |
-
" plt.title(f\"Reference image at index={index}\" , color='white' , fontsize=round(20*image_size))\n",
|
356 |
-
" #-----#\n",
|
357 |
-
" if show_encoding and loaded_ref:\n",
|
358 |
-
" fig.add_subplot(rows, columns, columns-1)\n",
|
359 |
-
" plt.imshow( visualize(prev_ref))\n",
|
360 |
-
" plt.axis('off')\n",
|
361 |
-
" plt.title(\"Encoding (before)\" , color='white' , fontsize=round(20*image_size))\n",
|
362 |
-
" print(f'Prompt for this image : \\n\\n \"{prompt} \" \\n\\n')\n",
|
363 |
-
"\n",
|
364 |
-
" if show_encoding:\n",
|
365 |
-
" fig.add_subplot(rows, columns, columns)\n",
|
366 |
-
" plt.imshow( visualize(ref))\n",
|
367 |
-
" plt.axis('off')\n",
|
368 |
-
" plt.title(\"Encoding (now)\" , color='white' , fontsize=round(20*image_size))\n",
|
369 |
-
" #------#\n"
|
370 |
-
],
|
371 |
-
"metadata": {
|
372 |
-
"id": "BwrEs5zVB0Sb",
|
373 |
-
"cellView": "form"
|
374 |
-
},
|
375 |
-
"execution_count": null,
|
376 |
-
"outputs": []
|
377 |
-
},
|
378 |
{
|
379 |
"cell_type": "markdown",
|
380 |
"source": [
|
@@ -540,12 +540,10 @@
|
|
540 |
{
|
541 |
"cell_type": "markdown",
|
542 |
"source": [
|
543 |
-
"# CLIP
|
544 |
-
"\n",
|
545 |
-
"**Save the reference prior to running the Interrogator**"
|
546 |
],
|
547 |
"metadata": {
|
548 |
-
"id": "
|
549 |
}
|
550 |
},
|
551 |
{
|
|
|
175 |
"id": "Xf9zoq-Za3wi"
|
176 |
}
|
177 |
},
|
178 |
+
{
|
179 |
+
"cell_type": "code",
|
180 |
+
"source": [
|
181 |
+
"# @title ⚄ 📷💭 Use pre-encoded image+prompt pair\n",
|
182 |
+
"loaded_ref = False\n",
|
183 |
+
"try:\n",
|
184 |
+
" ref\n",
|
185 |
+
" loaded_ref = True\n",
|
186 |
+
"except:ref = torch.zeros(dim).to(dtype = dot_dtype)\n",
|
187 |
+
"if loaded_ref : prev_ref = ref.clone().detach()\n",
|
188 |
+
"\n",
|
189 |
+
"try:prompt\n",
|
190 |
+
"except: prompt = ''\n",
|
191 |
+
"\n",
|
192 |
+
"# @markdown 🖼️+📝 Choose a pre-encoded reference (note: some results are NSFW!)\n",
|
193 |
+
"index = 596 # @param {type:\"slider\", min:0, max:1666, step:1}\n",
|
194 |
+
"PROMPT_INDEX = index\n",
|
195 |
+
"prompt = target_prompts[f'{PROMPT_INDEX}']\n",
|
196 |
+
"url = target_urls[f'{PROMPT_INDEX}']\n",
|
197 |
+
"if url.find('perchance')>-1:\n",
|
198 |
+
" image = Image.open(requests.get(url, stream=True).raw)\n",
|
199 |
+
"#------#\n",
|
200 |
+
"%cd {home_directory + 'fusion-t2i-generator-data/' + 'reference'}\n",
|
201 |
+
"references = torch.load('reference_text_and_image_encodings.pt' , weights_only=False)\n",
|
202 |
+
"# @markdown ⚖️ 🖼️ encoding <-----?-----> 📝 encoding </div> <br>\n",
|
203 |
+
"C = 0.3 # @param {type:\"slider\", min:0, max:1, step:0.01}\n",
|
204 |
+
"log_strength = 1 # @param {type:\"slider\", min:-5, max:5, step:0.01}\n",
|
205 |
+
"method = 'Add to existing ref' # @param [\"Refresh\" , \"Add to existing ref\" , \"Subtract from existing ref\" , \"Do nothing\"]\n",
|
206 |
+
"image_size = 0.57 # @param {type:\"slider\", min:0, max:1, step:0.01}\n",
|
207 |
+
"show_encoding = True # @param {type:\"boolean\"}\n",
|
208 |
+
"\n",
|
209 |
+
"if(not method == 'Do nothing'):\n",
|
210 |
+
" if method == 'Refresh': ref = torch.zeros(dim).to(dtype = dot_dtype)\n",
|
211 |
+
" if method == 'Subtract from existing ref':\n",
|
212 |
+
" ref = torch.sub(ref, math.pow(10 ,log_strength-1) * C * references[index][0].dequantize().to(dtype = torch.float32))\n",
|
213 |
+
" ref = torch.sub(ref, math.pow(10 ,log_strength-1) * (1-C) * references[index][1].dequantize().to(dtype = torch.float32))\n",
|
214 |
+
" else:\n",
|
215 |
+
" ref = torch.add(ref, math.pow(10 ,log_strength-1) * C * references[index][0].dequantize().to(dtype = torch.float32))\n",
|
216 |
+
" ref = torch.add(ref, math.pow(10 ,log_strength-1) * (1-C) * references[index][1].dequantize().to(dtype = torch.float32))\n",
|
217 |
+
" #---------#\n",
|
218 |
+
" references = '' # Clear up memory\n",
|
219 |
+
" ref = ref/ref.norm(p=2, dim=-1, keepdim=True)\n",
|
220 |
+
" ref = ref.clone().detach()\n",
|
221 |
+
" #------#\n",
|
222 |
+
" # create figure\n",
|
223 |
+
" fig = plt.figure(figsize=(10*image_size, 10*image_size))\n",
|
224 |
+
" fig.patch.set_facecolor((56/255,56/255,56/255))\n",
|
225 |
+
" rows = 1\n",
|
226 |
+
" columns = 1\n",
|
227 |
+
" if show_encoding: columns = columns+1\n",
|
228 |
+
" if show_encoding and loaded_ref : columns = columns+1\n",
|
229 |
+
" fig.add_subplot(rows, columns, 1)\n",
|
230 |
+
" plt.imshow(image)\n",
|
231 |
+
" plt.axis('off')\n",
|
232 |
+
" plt.title(f\"Reference image at index={index}\" , color='white' , fontsize=round(20*image_size))\n",
|
233 |
+
" #-----#\n",
|
234 |
+
" if show_encoding and loaded_ref:\n",
|
235 |
+
" fig.add_subplot(rows, columns, columns-1)\n",
|
236 |
+
" plt.imshow( visualize(prev_ref))\n",
|
237 |
+
" plt.axis('off')\n",
|
238 |
+
" plt.title(\"Encoding (before)\" , color='white' , fontsize=round(20*image_size))\n",
|
239 |
+
" print(f'Prompt for this image : \\n\\n \"{prompt} \" \\n\\n')\n",
|
240 |
+
"\n",
|
241 |
+
" if show_encoding:\n",
|
242 |
+
" fig.add_subplot(rows, columns, columns)\n",
|
243 |
+
" plt.imshow( visualize(ref))\n",
|
244 |
+
" plt.axis('off')\n",
|
245 |
+
" plt.title(\"Encoding (now)\" , color='white' , fontsize=round(20*image_size))\n",
|
246 |
+
" #------#\n"
|
247 |
+
],
|
248 |
+
"metadata": {
|
249 |
+
"id": "BwrEs5zVB0Sb",
|
250 |
+
"cellView": "form"
|
251 |
+
},
|
252 |
+
"execution_count": null,
|
253 |
+
"outputs": []
|
254 |
+
},
|
255 |
+
{
|
256 |
+
"cell_type": "markdown",
|
257 |
+
"source": [
|
258 |
+
"# Other methods"
|
259 |
+
],
|
260 |
+
"metadata": {
|
261 |
+
"id": "f9_AcquM7AYZ"
|
262 |
+
}
|
263 |
+
},
|
264 |
{
|
265 |
"cell_type": "code",
|
266 |
"source": [
|
|
|
375 |
"execution_count": null,
|
376 |
"outputs": []
|
377 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
378 |
{
|
379 |
"cell_type": "markdown",
|
380 |
"source": [
|
|
|
540 |
{
|
541 |
"cell_type": "markdown",
|
542 |
"source": [
|
543 |
+
"# Search prompts using CLIP"
|
|
|
|
|
544 |
],
|
545 |
"metadata": {
|
546 |
+
"id": "UqrYOkhlEQdM"
|
547 |
}
|
548 |
},
|
549 |
{
|