import spaces import gradio as gr import numpy as np # DiffuseCraft from dc import (infer, _infer, pass_result, get_diffusers_model_list, get_samplers, get_vaes, enable_model_recom_prompt, enable_diffusers_model_detail, get_t2i_model_info, get_all_lora_tupled_list, update_loras, apply_lora_prompt, download_my_lora, search_civitai_lora, select_civitai_lora, search_civitai_lora_json, preset_quality, preset_styles, process_style_prompt) # Translator from llmdolphin import (dolphin_respond_auto, dolphin_parse_simple, get_llm_formats, get_dolphin_model_format, get_dolphin_models, get_dolphin_model_info, select_dolphin_model, select_dolphin_format, get_dolphin_sysprompt) # Tagger from tagger.v2 import v2_upsampling_prompt, V2_ALL_MODELS from tagger.utils import (gradio_copy_text, gradio_copy_prompt, COPY_ACTION_JS, V2_ASPECT_RATIO_OPTIONS, V2_RATING_OPTIONS, V2_LENGTH_OPTIONS, V2_IDENTITY_OPTIONS) from tagger.tagger import (predict_tags_wd, convert_danbooru_to_e621_prompt, remove_specific_prompt, insert_recom_prompt, compose_prompt_to_copy, translate_prompt, select_random_character) from tagger.fl2sd3longcap import predict_tags_fl2_sd3 def description_ui(): gr.Markdown( """ ## Danbooru Tags Transformer V2 Demo with WD Tagger & SD3 Long Captioner (Image =>) Prompt => Upsampled longer prompt - Mod of p1atdev's [Danbooru Tags Transformer V2 Demo](https://huggingface.co/spaces/p1atdev/danbooru-tags-transformer-v2) and [WD Tagger with 🤗 transformers](https://huggingface.co/spaces/p1atdev/wd-tagger-transformers). - Models: p1atdev's [wd-swinv2-tagger-v3-hf](https://huggingface.co/p1atdev/wd-swinv2-tagger-v3-hf), [dart-v2-moe-sft](https://huggingface.co/p1atdev/dart-v2-moe-sft), [dart-v2-sft](https://huggingface.co/p1atdev/dart-v2-sft)\ , gokaygokay's [Florence-2-SD3-Captioner](https://huggingface.co/gokaygokay/Florence-2-SD3-Captioner) """ ) MAX_SEED = np.iinfo(np.int32).max MAX_IMAGE_SIZE = 1216 css = """ #container { margin: 0 auto; !important; } #col-container { margin: 0 auto; !important; } #result { max-width: 520px; max-height: 520px; margin: 0px auto; !important; } .lora { min-width: 480px; !important; } #model-info { text-align: center; !important; } """ with gr.Blocks(fill_width=True, elem_id="container", css=css, delete_cache=(60, 3600)) as demo: with gr.Tab("Image Generator"): with gr.Column(elem_id="col-container"): with gr.Row(): prompt = gr.Text(label="Prompt", show_label=False, lines=1, max_lines=8, placeholder="Enter your prompt", container=False) with gr.Row(): run_button = gr.Button("Run") run_translate_button = gr.Button("Translate") result = gr.Image(label="Result", elem_id="result", format="png", show_label=False, interactive=False, show_download_button=True, show_share_button=False, container=True) with gr.Accordion("Advanced Settings", open=False): with gr.Row(): negative_prompt = gr.Text(label="Negative prompt", lines=1, max_lines=6, placeholder="Enter a negative prompt", value="(low quality, worst quality:1.2), very displeasing, watermark, signature, ugly") with gr.Row(): seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0) randomize_seed = gr.Checkbox(label="Randomize seed", value=True) with gr.Row(): width = gr.Slider(label="Width", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024) # 832 height = gr.Slider(label="Height", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024) # 1216 guidance_scale = gr.Slider(label="Guidance scale", minimum=0.0, maximum=30.0, step=0.1, value=7) num_inference_steps = gr.Slider(label="Number of inference steps", minimum=1, maximum=100, step=1, value=28) with gr.Row(): with gr.Column(scale=4): model_name = gr.Dropdown(label="Model", info="You can enter a huggingface model repo_id to want to use.", choices=get_diffusers_model_list(), value=get_diffusers_model_list()[0], allow_custom_value=True, interactive=True, min_width=320) model_info = gr.Markdown(elem_id="model-info") with gr.Column(scale=1): model_detail = gr.Checkbox(label="Show detail of model in list", value=False) with gr.Row(): sampler = gr.Dropdown(label="Sampler", choices=get_samplers(), value="Euler a") vae_model = gr.Dropdown(label="VAE Model", choices=get_vaes(), value=get_vaes()[0]) with gr.Accordion("LoRA", open=True, visible=True): with gr.Row(): with gr.Column(): with gr.Row(): lora1 = gr.Dropdown(label="LoRA 1", choices=get_all_lora_tupled_list(), value="", allow_custom_value=True, elem_classes="lora", min_width=320) lora1_wt = gr.Slider(minimum=-2, maximum=2, step=0.01, value=1.00, label="LoRA 1: weight") with gr.Row(): lora1_info = gr.Textbox(label="", info="Example of prompt:", value="", show_copy_button=True, interactive=False, visible=False) lora1_copy = gr.Button(value="Copy example to prompt", visible=False) lora1_md = gr.Markdown(value="", visible=False) with gr.Column(): with gr.Row(): lora2 = gr.Dropdown(label="LoRA 2", choices=get_all_lora_tupled_list(), value="", allow_custom_value=True, elem_classes="lora", min_width=320) lora2_wt = gr.Slider(minimum=-2, maximum=2, step=0.01, value=1.00, label="LoRA 2: weight") with gr.Row(): lora2_info = gr.Textbox(label="", info="Example of prompt:", value="", show_copy_button=True, interactive=False, visible=False) lora2_copy = gr.Button(value="Copy example to prompt", visible=False) lora2_md = gr.Markdown(value="", visible=False) with gr.Column(): with gr.Row(): lora3 = gr.Dropdown(label="LoRA 3", choices=get_all_lora_tupled_list(), value="", allow_custom_value=True, elem_classes="lora", min_width=320) lora3_wt = gr.Slider(minimum=-2, maximum=2, step=0.01, value=1.00, label="LoRA 3: weight") with gr.Row(): lora3_info = gr.Textbox(label="", info="Example of prompt:", value="", show_copy_button=True, interactive=False, visible=False) lora3_copy = gr.Button(value="Copy example to prompt", visible=False) lora3_md = gr.Markdown(value="", visible=False) with gr.Column(): with gr.Row(): lora4 = gr.Dropdown(label="LoRA 4", choices=get_all_lora_tupled_list(), value="", allow_custom_value=True, elem_classes="lora", min_width=320) lora4_wt = gr.Slider(minimum=-2, maximum=2, step=0.01, value=1.00, label="LoRA 4: weight") with gr.Row(): lora4_info = gr.Textbox(label="", info="Example of prompt:", value="", show_copy_button=True, interactive=False, visible=False) lora4_copy = gr.Button(value="Copy example to prompt", visible=False) lora4_md = gr.Markdown(value="", visible=False) with gr.Column(): with gr.Row(): lora5 = gr.Dropdown(label="LoRA 5", choices=get_all_lora_tupled_list(), value="", allow_custom_value=True, elem_classes="lora", min_width=320) lora5_wt = gr.Slider(minimum=-2, maximum=2, step=0.01, value=1.00, label="LoRA 5: weight") with gr.Row(): lora5_info = gr.Textbox(label="", info="Example of prompt:", value="", show_copy_button=True, interactive=False, visible=False) lora5_copy = gr.Button(value="Copy example to prompt", visible=False) lora5_md = gr.Markdown(value="", visible=False) with gr.Accordion("From URL", open=True, visible=True): with gr.Row(): lora_search_civitai_query = gr.Textbox(label="Query", placeholder="oomuro sakurako...", lines=1) lora_search_civitai_basemodel = gr.CheckboxGroup(label="Search LoRA for", choices=["Pony", "SD 1.5", "SDXL 1.0"], value=["Pony", "SDXL 1.0"]) lora_search_civitai_submit = gr.Button("Search on Civitai") with gr.Row(): lora_search_civitai_result = gr.Dropdown(label="Search Results", choices=[("", "")], value="", allow_custom_value=True, visible=False) lora_search_civitai_json = gr.JSON(value={}, visible=False) lora_search_civitai_desc = gr.Markdown(value="", visible=False) lora_download_url = gr.Textbox(label="URL", placeholder="http://...my_lora_url.safetensors", lines=1) lora_download = gr.Button("Get and set LoRA and apply to prompt") with gr.Row(): quality_selector = gr.Radio(label="Quality Tag Presets", interactive=True, choices=list(preset_quality.keys()), value="None", scale=3) style_selector = gr.Radio(label="Style Presets", interactive=True, choices=list(preset_styles.keys()), value="None", scale=3) recom_prompt = gr.Checkbox(label="Recommended prompt", value=True, scale=1) with gr.Accordion("Translation Settings", open=False): chatbot = gr.Chatbot(likeable=False, render_markdown=False, visible=False) # component for auto-translation chat_model = gr.Dropdown(choices=get_dolphin_models(), value=get_dolphin_models()[0][1], allow_custom_value=True, label="Model") chat_model_info = gr.Markdown(value=get_dolphin_model_info(get_dolphin_models()[0][1]), label="Model info") chat_format = gr.Dropdown(choices=get_llm_formats(), value=get_dolphin_model_format(get_dolphin_models()[0][1]), label="Message format") with gr.Row(): chat_tokens = gr.Slider(minimum=1, maximum=4096, value=512, step=1, label="Max tokens") chat_temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature") chat_topp = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p") chat_topk = gr.Slider(minimum=0, maximum=100, value=40, step=1, label="Top-k") chat_rp = gr.Slider(minimum=0.0, maximum=2.0, value=1.1, step=0.1, label="Repetition penalty") chat_sysmsg = gr.Textbox(value=get_dolphin_sysprompt(), label="System message") examples = gr.Examples( examples = [ ["souryuu asuka langley, 1girl, neon genesis evangelion, plugsuit, pilot suit, red bodysuit, sitting, crossing legs, black eye patch, cat hat, throne, symmetrical, looking down, from bottom, looking at viewer, outdoors"], ["sailor moon, magical girl transformation, sparkles and ribbons, soft pastel colors, crescent moon motif, starry night sky background, shoujo manga style"], ["kafuu chino, 1girl, solo"], ["1girl"], ["beautiful sunset"], ], inputs=[prompt], ) gr.on( #lambda x: None, inputs=None, outputs=result).then( triggers=[run_button.click, prompt.submit], fn=infer, inputs=[prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, model_name, lora1, lora1_wt, lora2, lora2_wt, lora3, lora3_wt, lora4, lora4_wt, lora5, lora5_wt, sampler, vae_model], outputs=[result], queue=True, show_progress="full", show_api=True, ) gr.on( #lambda x: None, inputs=None, outputs=result).then( triggers=[run_translate_button.click], fn=_infer, # dummy fn for api inputs=[prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, model_name, lora1, lora1_wt, lora2, lora2_wt, lora3, lora3_wt, lora4, lora4_wt, lora5, lora5_wt, sampler, vae_model], outputs=[result], queue=False, show_api=True, api_name="infer_translate", ).success( fn=dolphin_respond_auto, inputs=[prompt, chatbot], outputs=[chatbot], queue=True, show_progress="full", show_api=False, ).success( fn=dolphin_parse_simple, inputs=[prompt, chatbot], outputs=[prompt], queue=False, show_api=False, ).success( fn=infer, inputs=[prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, model_name, lora1, lora1_wt, lora2, lora2_wt, lora3, lora3_wt, lora4, lora4_wt, lora5, lora5_wt, sampler, vae_model], outputs=[result], queue=True, show_progress="full", show_api=False, ).success(lambda: None, None, chatbot, queue=False, show_api=False)\ .success(pass_result, [result], [result], queue=False, show_api=False) # dummy fn for api gr.on( triggers=[lora1.change, lora1_wt.change, lora2.change, lora2_wt.change, lora3.change, lora3_wt.change, lora4.change, lora4_wt.change, lora5.change, lora5_wt.change], fn=update_loras, inputs=[prompt, lora1, lora1_wt, lora2, lora2_wt, lora3, lora3_wt, lora4, lora4_wt, lora5, lora5_wt], outputs=[prompt, lora1, lora1_wt, lora1_info, lora1_copy, lora1_md, lora2, lora2_wt, lora2_info, lora2_copy, lora2_md, lora3, lora3_wt, lora3_info, lora3_copy, lora3_md, lora4, lora4_wt, lora4_info, lora4_copy, lora4_md, lora5, lora5_wt, lora5_info, lora5_copy, lora5_md], queue=False, trigger_mode="once", show_api=False, ) lora1_copy.click(apply_lora_prompt, [prompt, lora1_info], [prompt], queue=False, show_api=False) lora2_copy.click(apply_lora_prompt, [prompt, lora2_info], [prompt], queue=False, show_api=False) lora3_copy.click(apply_lora_prompt, [prompt, lora3_info], [prompt], queue=False, show_api=False) lora4_copy.click(apply_lora_prompt, [prompt, lora4_info], [prompt], queue=False, show_api=False) lora5_copy.click(apply_lora_prompt, [prompt, lora5_info], [prompt], queue=False, show_api=False) gr.on( triggers=[lora_search_civitai_submit.click, lora_search_civitai_query.submit], fn=search_civitai_lora, inputs=[lora_search_civitai_query, lora_search_civitai_basemodel], outputs=[lora_search_civitai_result, lora_search_civitai_desc, lora_search_civitai_submit, lora_search_civitai_query], scroll_to_output=True, queue=True, show_api=False, ) lora_search_civitai_json.change(search_civitai_lora_json, [lora_search_civitai_query, lora_search_civitai_basemodel], [lora_search_civitai_json], queue=True, show_api=True) # fn for api lora_search_civitai_result.change(select_civitai_lora, [lora_search_civitai_result], [lora_download_url, lora_search_civitai_desc], scroll_to_output=True, queue=False, show_api=False) gr.on( triggers=[lora_download.click, lora_download_url.submit], fn=download_my_lora, inputs=[lora_download_url,lora1, lora2, lora3, lora4, lora5], outputs=[lora1, lora2, lora3, lora4, lora5], scroll_to_output=True, queue=True, show_api=False, ) recom_prompt.change(enable_model_recom_prompt, [recom_prompt], [recom_prompt], queue=False, show_api=False) gr.on( triggers=[quality_selector.change, style_selector.change], fn=process_style_prompt, inputs=[prompt, negative_prompt, style_selector, quality_selector], outputs=[prompt, negative_prompt], queue=False, trigger_mode="once", ) model_detail.change(enable_diffusers_model_detail, [model_detail, model_name], [model_detail, model_name], queue=False, show_api=False) model_name.change(get_t2i_model_info, [model_name], [model_info], queue=False, show_api=False) chat_model.change(select_dolphin_model, [chat_model], [chat_model, chat_format, chat_model_info], queue=True, show_progress="full", show_api=False)\ .success(lambda: None, None, chatbot, queue=False, show_api=False) chat_format.change(select_dolphin_format, [chat_format], [chat_format], queue=False, show_api=False)\ .success(lambda: None, None, chatbot, queue=False, show_api=False) # Tagger with gr.Tab("Tags Transformer with Tagger"): with gr.Column(): with gr.Group(): input_image = gr.Image(label="Input image", type="pil", sources=["upload", "clipboard"], height=256) with gr.Accordion(label="Advanced options", open=False): general_threshold = gr.Slider(label="Threshold", minimum=0.0, maximum=1.0, value=0.3, step=0.01, interactive=True) character_threshold = gr.Slider(label="Character threshold", minimum=0.0, maximum=1.0, value=0.8, step=0.01, interactive=True) input_tag_type = gr.Radio(label="Convert tags to", info="danbooru for Animagine, e621 for Pony.", choices=["danbooru", "e621"], value="danbooru") recom_prompt = gr.Radio(label="Insert reccomended prompt", choices=["None", "Animagine", "Pony"], value="None", interactive=True) image_algorithms = gr.CheckboxGroup(["Use WD Tagger", "Use Florence-2-SD3-Long-Captioner"], label="Algorithms", value=["Use WD Tagger"]) keep_tags = gr.Radio(label="Remove tags leaving only the following", choices=["body", "dress", "all"], value="all") generate_from_image_btn = gr.Button(value="GENERATE TAGS FROM IMAGE", size="lg", variant="primary") with gr.Group(): with gr.Row(): input_character = gr.Textbox(label="Character tags", placeholder="hatsune miku") input_copyright = gr.Textbox(label="Copyright tags", placeholder="vocaloid") random_character = gr.Button(value="Random character 🎲", size="sm") input_general = gr.TextArea(label="General tags", lines=4, placeholder="1girl, ...", value="") input_tags_to_copy = gr.Textbox(value="", visible=False) with gr.Row(): copy_input_btn = gr.Button(value="Copy to clipboard", size="sm", interactive=False) copy_prompt_btn_input = gr.Button(value="Copy to primary prompt", size="sm", interactive=False) translate_input_prompt_button = gr.Button(value="Translate prompt to English", size="sm", variant="secondary") tag_type = gr.Radio(label="Output tag conversion", info="danbooru for Animagine, e621 for Pony.", choices=["danbooru", "e621"], value="e621", visible=False) input_rating = gr.Radio(label="Rating", choices=list(V2_RATING_OPTIONS), value="explicit") with gr.Accordion(label="Advanced options", open=False): input_aspect_ratio = gr.Radio(label="Aspect ratio", info="The aspect ratio of the image.", choices=list(V2_ASPECT_RATIO_OPTIONS), value="square") input_length = gr.Radio(label="Length", info="The total length of the tags.", choices=list(V2_LENGTH_OPTIONS), value="very_long") input_identity = gr.Radio(label="Keep identity", info="How strictly to keep the identity of the character or subject. If you specify the detail of subject in the prompt, you should choose `strict`. Otherwise, choose `none` or `lax`. `none` is very creative but sometimes ignores the input prompt.", choices=list(V2_IDENTITY_OPTIONS), value="lax") input_ban_tags = gr.Textbox(label="Ban tags", info="Tags to ban from the output.", placeholder="alternate costumen, ...", value="censored") model_name = gr.Dropdown(label="Model", choices=list(V2_ALL_MODELS.keys()), value=list(V2_ALL_MODELS.keys())[0]) dummy_np = gr.Textbox(label="Negative prompt", value="", visible=False) recom_animagine = gr.Textbox(label="Animagine reccomended prompt", value="Animagine", visible=False) recom_pony = gr.Textbox(label="Pony reccomended prompt", value="Pony", visible=False) generate_btn = gr.Button(value="GENERATE TAGS", size="lg", variant="primary") with gr.Row(): with gr.Group(): output_text = gr.TextArea(label="Output tags", interactive=False, show_copy_button=True) with gr.Row(): copy_btn = gr.Button(value="Copy to clipboard", size="sm", interactive=False) copy_prompt_btn = gr.Button(value="Copy to primary prompt", size="sm", interactive=False) with gr.Group(): output_text_pony = gr.TextArea(label="Output tags (Pony e621 style)", interactive=False, show_copy_button=True) with gr.Row(): copy_btn_pony = gr.Button(value="Copy to clipboard", size="sm", interactive=False) copy_prompt_btn_pony = gr.Button(value="Copy to primary prompt", size="sm", interactive=False) random_character.click(select_random_character, [input_copyright, input_character], [input_copyright, input_character], queue=False, show_api=False) translate_input_prompt_button.click(translate_prompt, [input_general], [input_general], queue=False, show_api=False) translate_input_prompt_button.click(translate_prompt, [input_character], [input_character], queue=False, show_api=False) translate_input_prompt_button.click(translate_prompt, [input_copyright], [input_copyright], queue=False, show_api=False) generate_from_image_btn.click( lambda: ("", "", ""), None, [input_copyright, input_character, input_general], queue=False, show_api=False, ).success( predict_tags_wd, [input_image, input_general, image_algorithms, general_threshold, character_threshold], [input_copyright, input_character, input_general, copy_input_btn], show_api=False, ).success( predict_tags_fl2_sd3, [input_image, input_general, image_algorithms], [input_general], show_api=False, ).success( remove_specific_prompt, [input_general, keep_tags], [input_general], queue=False, show_api=False, ).success( convert_danbooru_to_e621_prompt, [input_general, input_tag_type], [input_general], queue=False, show_api=False, ).success( insert_recom_prompt, [input_general, dummy_np, recom_prompt], [input_general, dummy_np], queue=False, show_api=False, ).success(lambda: gr.update(interactive=True), None, [copy_prompt_btn_input], queue=False, show_api=False) copy_input_btn.click(compose_prompt_to_copy, [input_character, input_copyright, input_general], [input_tags_to_copy], show_api=False)\ .success(gradio_copy_text, [input_tags_to_copy], js=COPY_ACTION_JS, show_api=False) copy_prompt_btn_input.click(compose_prompt_to_copy, inputs=[input_character, input_copyright, input_general], outputs=[input_tags_to_copy], show_api=False)\ .success(gradio_copy_prompt, inputs=[input_tags_to_copy], outputs=[prompt], show_api=False) generate_btn.click( v2_upsampling_prompt, [model_name, input_copyright, input_character, input_general, input_rating, input_aspect_ratio, input_length, input_identity, input_ban_tags], [output_text], show_api=False, ).success( convert_danbooru_to_e621_prompt, [output_text, tag_type], [output_text_pony], queue=False, show_api=False, ).success( insert_recom_prompt, [output_text, dummy_np, recom_animagine], [output_text, dummy_np], queue=False, show_api=False, ).success( insert_recom_prompt, [output_text_pony, dummy_np, recom_pony], [output_text_pony, dummy_np], queue=False, show_api=False, ).success(lambda: (gr.update(interactive=True), gr.update(interactive=True), gr.update(interactive=True), gr.update(interactive=True)), None, [copy_btn, copy_btn_pony, copy_prompt_btn, copy_prompt_btn_pony], queue=False, show_api=False) copy_btn.click(gradio_copy_text, [output_text], js=COPY_ACTION_JS, show_api=False) copy_btn_pony.click(gradio_copy_text, [output_text_pony], js=COPY_ACTION_JS, show_api=False) copy_prompt_btn.click(gradio_copy_prompt, inputs=[output_text], outputs=[prompt], show_api=False) copy_prompt_btn_pony.click(gradio_copy_prompt, inputs=[output_text_pony], outputs=[prompt], show_api=False) demo.queue() demo.launch()