#@title Prepare the Concepts Library to be used import requests import os import gradio as gr import wget import torch from torch import autocast from diffusers import StableDiffusionPipeline from huggingface_hub import HfApi from transformers import CLIPTextModel, CLIPTokenizer import html community_icon_html = """""" loading_icon_html = """""" share_js = """async () => { async function uploadFile(file){ const UPLOAD_URL = 'https://huggingface.co/uploads'; const response = await fetch(UPLOAD_URL, { method: 'POST', headers: { 'Content-Type': file.type, 'X-Requested-With': 'XMLHttpRequest', }, body: file, /// <- File inherits from Blob }); const url = await response.text(); return url; } const gradioEl = document.querySelector('body > gradio-app'); const imgEls = gradioEl.querySelectorAll('#generated-gallery img'); const promptTxt = gradioEl.querySelector('#prompt_input input').value; const shareBtnEl = gradioEl.querySelector('#share-btn'); const shareIconEl = gradioEl.querySelector('#share-btn-share-icon'); const loadingIconEl = gradioEl.querySelector('#share-btn-loading-icon'); if(!imgEls.length){ return; }; shareBtnEl.style.pointerEvents = 'none'; shareIconEl.style.display = 'none'; loadingIconEl.style.removeProperty('display'); const files = await Promise.all( [...imgEls].map(async (imgEl) => { const res = await fetch(imgEl.src); const blob = await res.blob(); const imgId = Date.now() % 200; const fileName = `diffuse-the-rest-${{imgId}}.png`; return new File([blob], fileName, { type: 'image/png' }); }) ); const REGEX_CONCEPT = /<(((?!<.+>).)+)>/gm; const matches = [...promptTxt.matchAll(REGEX_CONCEPT)]; const concepts = matches.map(m => m[1]); const conceptLibraryMd = concepts.map(c => `${c}`).join(`, `); const urls = await Promise.all(files.map((f) => uploadFile(f))); const htmlImgs = urls.map(url => ``); const htmlImgsMd = htmlImgs.join(`\n`); const descriptionMd = `#### Prompt: ${promptTxt.replace(//g, '\\\>')} #### Concepts Used: ${conceptLibraryMd} #### Generations:
${htmlImgsMd}
`; const params = new URLSearchParams({ title: promptTxt, description: descriptionMd, }); const paramsStr = params.toString(); window.open(`https://huggingface.co/spaces/sd-concepts-library/stable-diffusion-conceptualizer/discussions/new?${paramsStr}`, '_blank'); shareBtnEl.style.removeProperty('pointer-events'); shareIconEl.style.removeProperty('display'); loadingIconEl.style.display = 'none'; }""" api = HfApi() models_list = api.list_models(author="sd-concepts-library", sort="likes", direction=-1) models = [] # NEW LOGIN ATTEMPT {{{ api_key = os.environ['api_key'] my_token = api_key pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", revision="fp16", torch_dtype=torch.float16, use_auth_token=my_token).to("cuda") # }}} # pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", use_auth_token=True, revision="fp16", torch_dtype=torch.float16).to("cuda") def load_learned_embed_in_clip(learned_embeds_path, text_encoder, tokenizer, token=None): loaded_learned_embeds = torch.load(learned_embeds_path, map_location="cpu") _old_token = token # separate token and the embeds trained_token = list(loaded_learned_embeds.keys())[0] embeds = loaded_learned_embeds[trained_token] # cast to dtype of text_encoder dtype = text_encoder.get_input_embeddings().weight.dtype # add the token in tokenizer token = token if token is not None else trained_token num_added_tokens = tokenizer.add_tokens(token) i = 1 print("start while loop **************") while(num_added_tokens == 0): token = f"{token[:-1]}-{i}>" num_added_tokens = tokenizer.add_tokens(token) print("i --> ", i) print("token --> ", token) print("num_added_tokens --> ", num_added_tokens) i+=1 print("end while loop **************") # resize the token embeddings text_encoder.resize_token_embeddings(len(tokenizer)) # get the id for the token and assign the embeds token_id = tokenizer.convert_tokens_to_ids(token) print("&&&&&&&&&&&&&&&&") print("learned_embeds_path --> ", learned_embeds_path) print("text_encoder --> ", text_encoder) print("tokenizer --> ", tokenizer) print("_old_token --> ", _old_token) print("token --> ", token) print("trained_token --> ", trained_token) print("dtype --> ", dtype) print("num_added_tokens --> ", num_added_tokens) print("text_encoder --> ", text_encoder) print("token_id --> ", token_id) print("embeds --> ", embeds) print("&&&&&&&&&&&&&&&&") text_encoder.get_input_embeddings().weight.data[token_id] = embeds # <------ POINT OF FAILURE return token ahx_model_list = [model for model in models_list if "ahx" in model.modelId] # UNDER CONSTRUCTION --------------------------------------------------------------- from time import sleep print("--------------------------------------------------") print("--------------------------------------------------") print("Setting up the public library........") print("--------------------------------------------------") for model in ahx_model_list: model_content = {} model_id = model.modelId model_content["id"] = model_id embeds_url = f"https://huggingface.co/{model_id}/resolve/main/learned_embeds.bin" os.makedirs(model_id,exist_ok = True) if not os.path.exists(f"{model_id}/learned_embeds.bin"): try: wget.download(embeds_url, out=model_id) except: # print("FAILURE: <-------------------------------------------------------------------") # print("model -->", model) # print("model_id -->", model_id) # print("CONTINUING - MODEL NOT LOADING") continue token_identifier = f"https://huggingface.co/{model_id}/raw/main/token_identifier.txt" response = requests.get(token_identifier) token_name = response.text concept_type = f"https://huggingface.co/{model_id}/raw/main/type_of_concept.txt" response = requests.get(concept_type) concept_name = response.text model_content["concept_type"] = concept_name images = [] for i in range(4): url = f"https://huggingface.co/{model_id}/resolve/main/concept_images/{i}.jpeg" image_download = requests.get(url) url_code = image_download.status_code if(url_code == 200): file = open(f"{model_id}/{i}.jpeg", "wb") ## Creates the file for image file.write(image_download.content) ## Saves file content file.close() images.append(f"{model_id}/{i}.jpeg") model_content["images"] = images #if token cannot be loaded, skip it try: learned_token = load_learned_embed_in_clip(f"{model_id}/learned_embeds.bin", pipe.text_encoder, pipe.tokenizer, token_name) print("success / model loaded:") print("model -->", model) print("model_id -->", model_id) except: print("FAILURE: <- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -") print("model -->", model) print("model_id -->", model_id) print("CONTINUING - MODEL NOT LOADING") continue model_content["token"] = learned_token models.append(model_content) models.append(model_content) print("--------------------------------------------------") sleep(5) print("--------------------------------------------------") print("--------------------------------------------------") print("--------------------------------------------------") sleep(60) #@title Run the app to navigate around [the Library](https://huggingface.co/sd-concepts-library) #@markdown Click the `Running on public URL:` result to run the Gradio app SELECT_LABEL = "Select concept" def assembleHTML(model): html_gallery = '' html_gallery = html_gallery+'''
''' cap = 0 for model in models: html_gallery = html_gallery+f'''

{html.escape(model["token"])}

''' cap += 1 if(cap == 99): break html_gallery = html_gallery+'''
''' return html_gallery def title_block(title, id): return gr.Markdown(f"### [`{title}`](https://huggingface.co/{id})") def image_block(image_list, concept_type): return gr.Gallery( label=concept_type, value=image_list, elem_id="gallery" ).style(grid=[2], height="auto") def checkbox_block(): checkbox = gr.Checkbox(label=SELECT_LABEL).style(container=False) return checkbox def infer(text): #with autocast("cuda"): images_list = pipe( [text], num_inference_steps=50, guidance_scale=7.5 ) #output_images = [] #for i, image in enumerate(images_list.images): # output_images.append(image) return images_list.images, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True) # idetnical to `infer` function without gradio state updates for share btn def infer_examples(text): #with autocast("cuda"): images_list = pipe( [text], num_inference_steps=50, guidance_scale=7.5 ) #output_images = [] #for i, image in enumerate(images_list["sample"]): # output_images.append(image) return images_list.images css = ''' .gradio-container {font-family: 'IBM Plex Sans', sans-serif} #top_title{margin-bottom: .5em} #top_title h2{margin-bottom: 0; text-align: center} /*#main_row{flex-wrap: wrap; gap: 1em; max-height: 550px; overflow-y: scroll; flex-direction: row}*/ #component-3{height: 760px; overflow: auto} #component-9{position: sticky;top: 0;align-self: flex-start;} @media (min-width: 768px){#main_row > div{flex: 1 1 32%; margin-left: 0 !important}} .gr-prose code::before, .gr-prose code::after {content: "" !important} ::-webkit-scrollbar {width: 10px} ::-webkit-scrollbar-track {background: #f1f1f1} ::-webkit-scrollbar-thumb {background: #888} ::-webkit-scrollbar-thumb:hover {background: #555} .gr-button {white-space: nowrap} .gr-button:focus { border-color: rgb(147 197 253 / var(--tw-border-opacity)); outline: none; box-shadow: var(--tw-ring-offset-shadow), var(--tw-ring-shadow), var(--tw-shadow, 0 0 #0000); --tw-border-opacity: 1; --tw-ring-offset-shadow: var(--tw-ring-inset) 0 0 0 var(--tw-ring-offset-width) var(--tw-ring-offset-color); --tw-ring-shadow: var(--tw-ring-inset) 0 0 0 calc(3px var(--tw-ring-offset-width)) var(--tw-ring-color); --tw-ring-color: rgb(191 219 254 / var(--tw-ring-opacity)); --tw-ring-opacity: .5; } #prompt_input{flex: 1 3 auto; width: auto !important;} #prompt_area{margin-bottom: .75em} #prompt_area > div:first-child{flex: 1 3 auto} .animate-spin { animation: spin 1s linear infinite; } @keyframes spin { from { transform: rotate(0deg); } to { transform: rotate(360deg); } } #share-btn-container { display: flex; padding-left: 0.5rem !important; padding-right: 0.5rem !important; background-color: #000000; justify-content: center; align-items: center; border-radius: 9999px !important; width: 13rem; } #share-btn { all: initial; color: #ffffff;font-weight: 600; cursor:pointer; font-family: 'IBM Plex Sans', sans-serif; margin-left: 0.5rem !important; padding-top: 0.25rem !important; padding-bottom: 0.25rem !important; } #share-btn * { all: unset; } ''' # examples = ["a in style", "a style mecha robot", "a piano being played by ", "Candid photo of , high resolution photo, trending on artstation, interior design"] examples = [] with gr.Blocks(css=css) as demo: state = gr.Variable({ 'selected': -1 }) state = {} def update_state(i): global checkbox_states if(checkbox_states[i]): checkbox_states[i] = False state[i] = False else: state[i] = True checkbox_states[i] = True gr.HTML('''

🧑‍🚀 Astronaut Horse Concept Loader


Run your own text prompts into fine-tuned artist concepts, see the example below. Currently only loading two artist concepts while testing. Soon will automatically be able to add all concepts from astronaut horse artist collaborations. There's some buggy stuff here that'll be cleared up next week but I wanted to at least get this usable for the weekend!


http://www.astronaut.horse

Prompt Examples Using Artist Token:

  • "a photograph of pink crystals in the style of <artist>"
  • "a painting of a horse in the style of <ivan-stripes>"

Currently-Usable Concept Tokens:

  • <artist>
  • <ivan-stripes>
''') with gr.Row(): # with gr.Column(): # gr.Markdown(f"") # with gr.Row(): # image_blocks = [] # #for i, model in enumerate(models): # with gr.Box().style(border=None): # gr.HTML(assembleHTML(models)) # #title_block(model["token"], model["id"]) # #image_blocks.append(image_block(model["images"], model["concept_type"])) with gr.Column(): with gr.Box(): with gr.Row(elem_id="prompt_area").style(mobile_collapse=False, equal_height=True): text = gr.Textbox( label="Enter your prompt", placeholder="Enter your prompt", show_label=False, max_lines=1, elem_id="prompt_input" ).style( border=(True, False, True, True), rounded=(True, False, False, True), container=False, ) btn = gr.Button("generate image",elem_id="run_btn").style( margin=False, rounded=(False, True, True, False), ) with gr.Row().style(): infer_outputs = gr.Gallery(show_label=False, elem_id="generated-gallery").style(grid=[2], height="512px") with gr.Row(): gr.HTML("

") with gr.Row(): gr.Examples(examples=examples, fn=infer_examples, inputs=[text], outputs=infer_outputs, cache_examples=True) with gr.Group(elem_id="share-btn-container"): community_icon = gr.HTML(community_icon_html, visible=False) loading_icon = gr.HTML(loading_icon_html, visible=False) checkbox_states = {} inputs = [text] btn.click( infer, inputs=inputs, outputs=[infer_outputs, community_icon, loading_icon] ) # after loading_icon on line 392.5 # share_button = gr.Button("", elem_id="share-btn", visible=False) # and update outputs=[...] on line 398 to match this # outputs=[infer_outputs, community_icon, loading_icon, share_button] # then this has to be added after line 399 # share_button.click( # None, # [], # [], # _js=share_js, # ) demo.queue(max_size=20).launch()