diff --git a/scripts/infinite-zoom.py b/scripts/infinite-zoom.py index 12c74c4..0b65d36 100644 --- a/scripts/infinite-zoom.py +++ b/scripts/infinite-zoom.py @@ -1,9 +1,9 @@ import sys import os import time +import json +from jsonschema import validate -basedir = os.getcwd() -sys.path.extend(basedir + "/extensions/infinite-zoom-automatic1111-webui/") import numpy as np import gradio as gr from PIL import Image @@ -12,7 +12,7 @@ import json from iz_helpers import shrink_and_paste_on_blank, write_video from webui import wrap_gradio_gpu_call -from modules import script_callbacks +from modules import script_callbacks, scripts import modules.shared as shared from modules.processing import ( process_images, @@ -20,31 +20,23 @@ from modules.processing import ( StableDiffusionProcessingImg2Img, ) +import scripts.postprocessing_upscale + from modules.ui import create_output_panel, plaintext_to_html +import modules.sd_models +import modules.sd_samplers -available_samplers = [ - "DDIM", - "Euler a", - "Euler", - "LMS", - "Heun", - "DPM2", - "DPM2 a", - "DPM++ 2S a", - "DPM++ 2M", - "DPM++ SDE", - "DPM fast", - "DPM adaptive", - "LMS Karras", - "DPM2 Karras", - "DPM2 a Karras", - "DPM++ 2S a Karras", - "DPM++ 2M Karras", - "DPM++ SDE Karras", -] -default_prompt = "A psychedelic jungle with trees that have glowing, fractal-like patterns, Simon stalenhag poster 1920s style, street level view, hyper futuristic, 8k resolution, hyper realistic" -default_negative_prompt = "frames, borderline, text, character, duplicate, error, out of frame, watermark, low quality, ugly, deformed, blur" +from modules import scripts +usefulDirs = scripts.basedir().split(os.sep)[-2:] # contains install and our extension foldername +jsonprompt_schemafile = usefulDirs[0]+"/"+usefulDirs[1]+"/scripts/promptschema.json" +available_samplers = [s.name for s in modules.sd_samplers.samplers] + +default_prompt = '{"prompts":{"data":[[0,"Cat"],["1","Dog"],["2","Happy Pets"]],"headers":["outpaint steps","prompt"]},"negPrompt":"ugly"}' +empty_prompt = '{"prompts":{"data":[],"headers":["outpaint steps","prompt"]},"negPrompt":""}' + +#must be python dict +invalid_prompt ={"prompts":{"data":[[0,"Your prompt-json is invalid, please check Settings"]],"headers":["outpaint steps","prompt"]},"negPrompt":"Invalid prompt-json"} def closest_upper_divisible_by_eight(num): if num % 8 == 0: @@ -52,6 +44,13 @@ def closest_upper_divisible_by_eight(num): else: return math.ceil(num / 8) * 8 +def do_upscaleImg(curImg,upscale_do, upscaler_name,upscale_by): + if (not upscale_do): return curImg + pp= scripts.postprocessing_upscale.scripts_postprocessing.PostprocessedImage(curImg) + ups = scripts.postprocessing_upscale.ScriptPostprocessingUpscale() + ups.process(pp, upscale_mode=2, upscale_by=upscale_by, upscale_to_width=None, upscale_to_height=None, upscale_crop=False, upscaler_1_name=upscaler_name, upscaler_2_name=None, upscaler_2_visibility=0.0) + return pp.image + def renderTxt2Img(prompt, negative_prompt, sampler, steps, cfg_scale, width, height): processed = None @@ -72,7 +71,6 @@ def renderTxt2Img(prompt, negative_prompt, sampler, steps, cfg_scale, width, hei processed = process_images(p) return processed - def renderImg2Img( prompt, negative_prompt, @@ -90,6 +88,7 @@ def renderImg2Img( inpainting_padding, ): processed = None + p = StableDiffusionProcessingImg2Img( sd_model=shared.sd_model, outpath_samples=shared.opts.outdir_img2img_samples, @@ -133,6 +132,7 @@ def create_zoom( guidance_scale, num_inference_steps, custom_init_image, + custom_exit_image, video_frame_rate, video_zoom_mode, video_start_frame_dupe_amount, @@ -147,6 +147,9 @@ def create_zoom( outputsizeH, batchcount, sampler, + upscale_do, + upscaler_name, + upscale_by, progress=None, ): for i in range(batchcount): @@ -158,6 +161,7 @@ def create_zoom( guidance_scale, num_inference_steps, custom_init_image, + custom_exit_image, video_frame_rate, video_zoom_mode, video_start_frame_dupe_amount, @@ -171,7 +175,11 @@ def create_zoom( outputsizeW, outputsizeH, sampler, + upscale_do, + upscaler_name, + upscale_by, progress, + ) return result @@ -183,6 +191,7 @@ def create_zoom_single( guidance_scale, num_inference_steps, custom_init_image, + custom_exit_image, video_frame_rate, video_zoom_mode, video_start_frame_dupe_amount, @@ -196,6 +205,9 @@ def create_zoom_single( outputsizeW, outputsizeH, sampler, + upscale_do, + upscaler_name, + upscale_by, progress=None, ): # try: @@ -229,6 +241,15 @@ def create_zoom_single( (width, height), resample=Image.LANCZOS ) else: + modelname = shared.opts.data.get("infzoom_txt2img_model") + if (modelname): + # switch to txt2img model + checkinfo = modules.sd_models.checkpoint_alisases[modelname] + if (not checkinfo): + raise NameError("Checklist not found in registry") + if progress: progress(0, desc="Loading Model for txt2img: " + checkinfo.name) + modules.sd_models.load_model(checkinfo) + processed = renderTxt2Img( prompts[min(k for k in prompts.keys() if k >= 0)], negative_prompt, @@ -240,25 +261,36 @@ def create_zoom_single( ) current_image = processed.images[0] + mask_width = math.trunc(width / 4) # was initially 512px => 128px mask_height = math.trunc(height / 4) # was initially 512px => 128px num_interpol_frames = round(video_frame_rate * zoom_speed) all_frames = [] - all_frames.append(current_image) + + + if upscale_do and progress: + progress(0, desc="upscaling inital image") + + all_frames.append(do_upscaleImg(current_image,upscale_do, upscaler_name,upscale_by) if upscale_do else current_image) + + inmodelname = shared.opts.data.get("infzoom_inpainting_model") + if (inmodelname): + # switch to inpaint model now + checkinfo = modules.sd_models.checkpoint_alisases[inmodelname] + if (not checkinfo): + raise NameError("Checklist not found in registry") + if progress: progress(0, desc="Loading Model for inpainting/img2img: " + checkinfo.name) + modules.sd_models.load_model(checkinfo) + for i in range(num_outpainting_steps): print_out = "Outpaint step: " + str(i + 1) + " / " + str(num_outpainting_steps) print(print_out) - # if progress is not None: - # progress( - # ((i + 1) / num_outpainting_steps), - # desc=print_out, - # ) + if progress: progress( ((i + 1) / num_outpainting_steps), desc=print_out) + prev_image_fix = current_image - prev_image = shrink_and_paste_on_blank(current_image, mask_width, mask_height) - current_image = prev_image # create mask (black image with white mask_width width edges) @@ -267,6 +299,7 @@ def create_zoom_single( # inpainting step current_image = current_image.convert("RGB") + processed = renderImg2Img( prompts[max(k for k in prompts.keys() if k <= i)], negative_prompt, @@ -341,8 +374,21 @@ def create_zoom_single( interpol_image.paste(prev_image_fix_crop, mask=prev_image_fix_crop) - all_frames.append(interpol_image) - all_frames.append(current_image) + if upscale_do and progress: + progress( + ((i + 1) / num_outpainting_steps), + desc="upscaling interpol" + ) + + all_frames.append(do_upscaleImg(interpol_image, upscale_do, upscaler_name,upscale_by) if upscale_do else interpol_image) + + if upscale_do and progress: + progress( + ((i + 1) / num_outpainting_steps), + desc="upscaling current" + ) + + all_frames.append(do_upscaleImg(current_image,upscale_do, upscaler_name,upscale_by) if upscale_do else current_image) video_file_name = "infinite_zoom_" + str(int(time.time())) + ".mp4" output_path = shared.opts.data.get( @@ -372,17 +418,26 @@ def create_zoom_single( ) -def exportPrompts(p, np): - print("prompts:" + str(p) + "\n" + str(np)) - - +def validatePromptJson_throws(data): + with open(jsonprompt_schemafile, "r") as s: schema = json.load(s) + validate(instance=data, schema=schema) + def putPrompts(files): - file_paths = [file.name for file in files] - with open(files.name, "r") as f: - file_contents = f.read() - data = json.loads(file_contents) - print(data) - return [gr.DataFrame.update(data["prompts"]), gr.Textbox.update(data["negPrompt"])] + + try: + with open(files.name, 'r') as f: + file_contents = f.read() + data = json.loads(file_contents) + validatePromptJson_throws(data) + return [gr.DataFrame.update(data["prompts"]), gr.Textbox.update(data["negPrompt"])] + + except Exception: + gr.Error("loading your prompt failed. It seems to be invalid. Your prompt table is preserved.") + print("[InfiniteZoom:] Loading your prompt failed. It seems to be invalid. Your prompt table is preserved.") + return [gr.DataFrame.update(), gr.Textbox.update()] + +def clearPrompts(): + return [gr.DataFrame.update(value=[[0,"Infinite Zoom. Start over"]]), gr.Textbox.update("")] def on_ui_tabs(): @@ -396,10 +451,12 @@ def on_ui_tabs(): """ ) - generate_btn = gr.Button(value="Generate video", variant="primary") - interrupt = gr.Button(value="Interrupt", elem_id="interrupt_training") + with gr.Row(): + generate_btn = gr.Button(value="Generate video", variant="primary") + interrupt = gr.Button(value="Interrupt", elem_id="interrupt_training") with gr.Row(): with gr.Column(scale=1, variant="panel"): + with gr.Tab("Main"): main_outpaint_steps = gr.Slider( minimum=2, @@ -409,18 +466,29 @@ def on_ui_tabs(): label="Total Outpaint Steps", info="The more it is, the longer your videos will be", ) + + # safe reading json prompt + pr = shared.opts.data.get("infzoom_defPrompt",default_prompt) + if (not pr): pr = empty_prompt + + try: + jpr = json.loads(pr) + validatePromptJson_throws(jpr) + except Exception: + jpr = invalid_prompt + main_prompts = gr.Dataframe( type="array", headers=["outpaint step", "prompt"], datatype=["number", "str"], row_count=1, col_count=(2, "fixed"), - value=[[0, default_prompt]], + value=jpr["prompts"], wrap=True, ) main_negative_prompt = gr.Textbox( - value=default_negative_prompt, label="Negative Prompt" + value=jpr["negPrompt"], label="Negative Prompt" ) # these button will be moved using JS unde the dataframe view as small ones @@ -447,6 +515,10 @@ def on_ui_tabs(): outputs=[main_prompts, main_negative_prompt], inputs=[importPrompts_button], ) + + clearPrompts_button= gr.Button(value="Clear prompts",variant="secondary",elem_classes="sm infzoom_tab_butt", elem_id="infzoom_clP_butt") + clearPrompts_button.click(fn=clearPrompts,inputs=[],outputs=[main_prompts,main_negative_prompt]) + main_sampler = gr.Dropdown( label="Sampler", choices=available_samplers, @@ -483,7 +555,10 @@ def on_ui_tabs(): value=50, label="Sampling Steps for each outpaint", ) - init_image = gr.Image(type="pil", label="custom initial image") + with gr.Row(): + init_image = gr.Image(type="pil", label="custom initial image") + exit_image = gr.Image(type="pil", label="custom exit image", visible=False) #TODO: implement exit-image rendering + batchcount_slider = gr.Slider( minimum=1, maximum=25, @@ -545,6 +620,19 @@ def on_ui_tabs(): label="masked padding", minimum=0, maximum=256, value=0 ) + with gr.Tab("Post proccess"): + upscale_do = gr.Checkbox(False, label="Enable Upscale") + upscaler_name = gr.Dropdown(label='Upscaler', elem_id="infZ_upscaler", choices=[x.name for x in shared.sd_upscalers], value=shared.sd_upscalers[0].name) + + upscale_by = gr.Slider( + label="Upscale by factor", minimum=1, maximum=8, value=1 + ) + with gr.Accordion("Help",open=False): + gr.Markdown("""# Performance critical +Depending on amount of frames and which upscaler you choose it might took a long time to render. +Our best experience and trade-off is the R-ERSGAn4x upscaler. +""") + with gr.Column(scale=1, variant="compact"): output_video = gr.Video(label="Output").style(width=512, height=512) ( @@ -553,7 +641,7 @@ def on_ui_tabs(): html_info, html_log, ) = create_output_panel( - "infinit-zoom", shared.opts.outdir_img2img_samples + "infinite-zoom", shared.opts.outdir_img2img_samples ) generate_btn.click( fn=wrap_gradio_gpu_call(create_zoom, extra_outputs=[None, '', '']), @@ -564,6 +652,7 @@ def on_ui_tabs(): main_guidance_scale, sampling_step, init_image, + exit_image, video_frame_rate, video_zoom_mode, video_start_frame_dupe_amount, @@ -578,6 +667,10 @@ def on_ui_tabs(): main_height, batchcount_slider, main_sampler, + upscale_do, + upscaler_name, + upscale_by + ], outputs=[output_video, out_image, generation_info, html_info, html_log], ) @@ -640,10 +733,42 @@ def on_ui_settings(): "Writing videos has dependency to an existing FFPROBE executable on your machine. D/L here (https://github.com/BtbN/FFmpeg-Builds/releases) your OS variant and point to your installation path", gr.Textbox, {"interactive": True}, - section=section, - ), + section=section + ) ) + shared.opts.add_option( + "infzoom_txt2img_model", + shared.OptionInfo( + shared.list_checkpoint_tiles[0], + "Name of your desired model to render keyframes (txt2img)", + gr.Dropdown, + lambda: {"choices": shared.list_checkpoint_tiles()}, + section=section + ) + ) + + shared.opts.add_option( + "infzoom_inpainting_model", + shared.OptionInfo( + "sd-v1-5-inpainting.ckpt", + "Name of your desired inpaint model (img2img-inpaint). Default is vanilla sd-v1-5-inpainting.ckpt ", + gr.Dropdown, + lambda: {"choices": shared.list_checkpoint_tiles()}, + section=section + ) + ) + shared.opts.add_option( + "infzoom_defPrompt", + shared.OptionInfo( + default_prompt, + "Default prompt-setup to start with'", + gr.Code, + {"interactive": True, "language":"json"}, + section=section + ) + ) + script_callbacks.on_ui_tabs(on_ui_tabs) script_callbacks.on_ui_settings(on_ui_settings) diff --git a/scripts/promptschema.json b/scripts/promptschema.json new file mode 100644 index 0000000..fcafb0f --- /dev/null +++ b/scripts/promptschema.json @@ -0,0 +1,55 @@ +{ + "$schema": "http://json-schema.org/draft-07/schema#", + "type": "object", + "properties": { + "prompts": { + "type": "object", + "properties": { + "data": { + "type": "array", + "items": { + "type": "array", + "items": [ + { + "oneOf": [ + { + "type": "integer", + "minimum": 0 + }, + { + "type": "string" + } + ] + }, + { + "type": "string" + } + ], + "minItems": 2, + "maxItems": 2, + "uniqueItems": true + }, + "minItems": 1 + }, + "headers": { + "type": "array", + "items": { + "type": "string" + }, + "minItems": 2 + } + }, + "required": [ + "data", + "headers" + ] + }, + "negPrompt": { + "type": "string" + } + }, + "required": [ + "prompts", + "negPrompt" + ] +} \ No newline at end of file