import gradio as gr import spaces from clip_slider_pipeline import CLIPSliderFlux from diffusers import FluxPipeline, AutoencoderTiny import torch import numpy as np import cv2 from PIL import Image from diffusers.utils import load_image from diffusers.pipelines.flux.pipeline_flux_controlnet import FluxControlNetPipeline from diffusers.models.controlnet_flux import FluxControlNetModel from diffusers.utils import export_to_gif def process_controlnet_img(image): controlnet_img = np.array(image) controlnet_img = cv2.Canny(controlnet_img, 100, 200) controlnet_img = HWC3(controlnet_img) controlnet_img = Image.fromarray(controlnet_img) # load pipelines taef1 = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=torch.bfloat16).to("cuda") pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", vae=taef1, torch_dtype=torch.bfloat16) pipe.transformer.to(memory_format=torch.channels_last) pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune", fullgraph=True) #pipe.enable_model_cpu_offload() clip_slider = CLIPSliderFlux(pipe, device=torch.device("cuda")) base_model = 'black-forest-labs/FLUX.1-schnell' controlnet_model = 'InstantX/FLUX.1-dev-Controlnet-Canny-alpha' # controlnet = FluxControlNetModel.from_pretrained(controlnet_model, torch_dtype=torch.bfloat16) # pipe_controlnet = FluxControlNetPipeline.from_pretrained(base_model, controlnet=controlnet, torch_dtype=torch.bfloat16) # t5_slider_controlnet = T5SliderFlux(sd_pipe=pipe_controlnet,device=torch.device("cuda")) @spaces.GPU(duration=200) def generate(concept_1, concept_2, scale, prompt, seed, recalc_directions, iterations, steps, interm_steps, guidance_scale, x_concept_1, x_concept_2, avg_diff_x, img2img_type = None, img = None, controlnet_scale= None, ip_adapter_scale=None, ): slider_x = [concept_1, concept_2] # check if avg diff for directions need to be re-calculated print("slider_x", slider_x) print("x_concept_1", x_concept_1, "x_concept_2", x_concept_2) #torch.manual_seed(seed) if not sorted(slider_x) == sorted([x_concept_1, x_concept_2]) or recalc_directions: #avg_diff = clip_slider.find_latent_direction(slider_x[0], slider_x[1], num_iterations=iterations).to(torch.float16) avg_diff = clip_slider.find_latent_direction(slider_x[0], slider_x[1], num_iterations=iterations) x_concept_1, x_concept_2 = slider_x[0], slider_x[1] images = [] high_scale = scale low_scale = -1 * scale for i in range(interm_steps): cur_scale = low_scale + (high_scale - low_scale) * i / (steps - 1) image = clip_slider.generate(prompt, #guidance_scale=guidance_scale, scale=cur_scale, seed=seed, num_inference_steps=steps, avg_diff=avg_diff) images.append(image) canvas = Image.new('RGB', (256*interm_steps, 256)) for i, im in enumerate(images): canvas.paste(im.resize((256,256)), (256 * i, 0)) comma_concepts_x = f"{slider_x[1]}, {slider_x[0]}" avg_diff_x = avg_diff.cpu() return gr.update(label=comma_concepts_x, interactive=True, value=scale), x_concept_1, x_concept_2, avg_diff_x, export_to_gif(images, "clip.gif", fps=5), canvas @spaces.GPU def update_scales(x,prompt,seed, steps, interm_steps, guidance_scale, avg_diff_x, img2img_type = None, img = None, controlnet_scale= None, ip_adapter_scale=None,): print("Hola", x) avg_diff = avg_diff_x.cuda() # for spectrum generation images = [] high_scale = x low_scale = -1 * x if img2img_type=="controlnet canny" and img is not None: control_img = process_controlnet_img(img) image = t5_slider_controlnet.generate(prompt, guidance_scale=guidance_scale, image=control_img, controlnet_conditioning_scale =controlnet_scale, scale=x, seed=seed, num_inference_steps=steps, avg_diff=avg_diff) elif img2img_type=="ip adapter" and img is not None: image = clip_slider.generate(prompt, guidance_scale=guidance_scale, ip_adapter_image=img, scale=x,seed=seed, num_inference_steps=steps, avg_diff=avg_diff) else: for i in range(interm_steps): cur_scale = low_scale + (high_scale - low_scale) * i / (steps - 1) image = clip_slider.generate(prompt, #guidance_scale=guidance_scale, scale=cur_scale, seed=seed, num_inference_steps=steps, avg_diff=avg_diff) images.append(image) canvas = Image.new('RGB', (256*interm_steps, 256)) for i, im in enumerate(images): canvas.paste(im.resize((256,256)), (256 * i, 0)) return export_to_gif(images, "clip.gif", fps=5), canvas def reset_recalc_directions(): return True css = ''' #group { position: relative; width: 600px; /* Increased width */ height: 600px; /* Increased height */ margin-bottom: 20px; background-color: white; } #x { position: absolute; bottom: 20px; /* Moved further down */ left: 30px; /* Adjusted left margin */ width: 540px; /* Increased width to match the new container size */ } #y { position: absolute; bottom: 200px; /* Increased bottom margin to ensure proper spacing from #x */ left: 20px; /* Adjusted left margin */ width: 540px; /* Increased width to match the new container size */ transform: rotate(-90deg); transform-origin: left bottom; } #image_out { position: absolute; width: 80%; /* Adjust width as needed */ right: 10px; top: 10px; /* Increased top margin to clear space occupied by #x */ } ''' intro = """

Latent Navigation

Exploring CLIP text space with FLUX.1 schnell 🪐

code | Duplicate Space

""" with gr.Blocks() as demo: gr.HTML(intro) x_concept_1 = gr.State("") x_concept_2 = gr.State("") # y_concept_1 = gr.State("") # y_concept_2 = gr.State("") avg_diff_x = gr.State() #avg_diff_y = gr.State() recalc_directions = gr.State(False) #with gr.Tab("text2image"): with gr.Row(): with gr.Column(): with gr.Row(): concept_1 = gr.Textbox(label="A concept to compare") concept_2 = gr.Textbox(label="Concept to compare") #slider_x = gr.Dropdown(label="Slider concept range", allow_custom_value=True, multiselect=True, max_choices=2) #slider_y = gr.Dropdown(label="Slider Y concept range", allow_custom_value=True, multiselect=True, max_choices=2) prompt = gr.Textbox(label="Prompt") x = gr.Slider(minimum=0, value=1.25, step=0.1, maximum=2.5, info="the strength to scale in each direction") submit = gr.Button("find directions") with gr.Column(): with gr.Group(elem_id="group"): #y = gr.Slider(minimum=-10, value=0, maximum=10, elem_id="y", interactive=False) output_image = gr.Image(elem_id="image_out") image_seq = gr.Image() # with gr.Row(): # generate_butt = gr.Button("generate") with gr.Accordion(label="advanced options", open=False): iterations = gr.Slider(label = "num iterations", minimum=0, value=200, maximum=400) steps = gr.Slider(label = "num inference steps", minimum=1, value=4, maximum=10) interm_steps = gr.Slider(label = "num of intermediate images", minimum=1, value=5, maximum=65) guidance_scale = gr.Slider( label="Guidance scale", minimum=0.1, maximum=10.0, step=0.1, value=5, ) seed = gr.Slider(minimum=0, maximum=np.iinfo(np.int32).max, label="Seed", interactive=True, randomize=True) # with gr.Tab(label="image2image"): # with gr.Row(): # with gr.Column(): # image = gr.ImageEditor(type="pil", image_mode="L", crop_size=(512, 512)) # slider_x_a = gr.Dropdown(label="Slider X concept range", allow_custom_value=True, multiselect=True, max_choices=2) # slider_y_a = gr.Dropdown(label="Slider X concept range", allow_custom_value=True, multiselect=True, max_choices=2) # img2img_type = gr.Radio(["controlnet canny", "ip adapter"], label="", info="", visible=False, value="controlnet canny") # prompt_a = gr.Textbox(label="Prompt") # submit_a = gr.Button("Submit") # with gr.Column(): # with gr.Group(elem_id="group"): # x_a = gr.Slider(minimum=-10, value=0, maximum=10, elem_id="x", interactive=False) # y_a = gr.Slider(minimum=-10, value=0, maximum=10, elem_id="y", interactive=False) # output_image_a = gr.Image(elem_id="image_out") # with gr.Row(): # generate_butt_a = gr.Button("generate") # with gr.Accordion(label="advanced options", open=False): # iterations_a = gr.Slider(label = "num iterations", minimum=0, value=200, maximum=300) # steps_a = gr.Slider(label = "num inference steps", minimum=1, value=8, maximum=30) # guidance_scale_a = gr.Slider( # label="Guidance scale", # minimum=0.1, # maximum=10.0, # step=0.1, # value=5, # ) # controlnet_conditioning_scale = gr.Slider( # label="controlnet conditioning scale", # minimum=0.5, # maximum=5.0, # step=0.1, # value=0.7, # ) # ip_adapter_scale = gr.Slider( # label="ip adapter scale", # minimum=0.5, # maximum=5.0, # step=0.1, # value=0.8, # visible=False # ) # seed_a = gr.Slider(minimum=0, maximum=np.iinfo(np.int32).max, label="Seed", interactive=True, randomize=True) # submit.click(fn=generate, # inputs=[slider_x, slider_y, prompt, seed, iterations, steps, guidance_scale, x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x, avg_diff_y], # outputs=[x, y, x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x, avg_diff_y, output_image]) submit.click(fn=generate, inputs=[concept_1, concept_2, x, prompt, seed, recalc_directions, iterations, steps, interm_steps, guidance_scale, x_concept_1, x_concept_2, avg_diff_x], outputs=[x, x_concept_1, x_concept_2, avg_diff_x, output_image, image_seq]) iterations.change(fn=reset_recalc_directions, outputs=[recalc_directions]) seed.change(fn=reset_recalc_directions, outputs=[recalc_directions]) x.release(fn=update_scales, inputs=[x, prompt, seed, steps, interm_steps, guidance_scale, avg_diff_x], outputs=[output_image, image_seq], trigger_mode='always_last') # generate_butt_a.click(fn=update_scales, inputs=[x_a,y_a, prompt_a, seed_a, steps_a, guidance_scale_a, avg_diff_x, avg_diff_y, img2img_type, image, controlnet_conditioning_scale, ip_adapter_scale], outputs=[output_image_a]) # submit_a.click(fn=generate, # inputs=[slider_x_a, slider_y_a, prompt_a, seed_a, iterations_a, steps_a, guidance_scale_a, x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x, avg_diff_y, img2img_type, image, controlnet_conditioning_scale, ip_adapter_scale], # outputs=[x_a, y_a, x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x, avg_diff_y, output_image_a]) if __name__ == "__main__": demo.launch()