init

app.py

@@ -1,49 +1,98 @@
+import os
 import gradio as gr
 import numpy as np
 import random
-#import spaces #[uncomment to use ZeroGPU]
+import spaces
 from diffusers import DiffusionPipeline
 import torch
+import json
+import logging
+from diffusers import DiffusionPipeline, AutoencoderTiny, AutoencoderKL
+from live_preview_helpers import calculate_shift, retrieve_timesteps, flux_pipe_call_that_returns_an_iterable_of_images
 
+from huggingface_hub import hf_hub_download, HfFileSystem, ModelCard, snapshot_download
+import copy
+import random
+import time
+
+
+# init
+dtype = torch.bfloat16
 device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo" #Replace to the model you would like to use
+base_model = "black-forest-labs/FLUX.1-dev"
 
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
+taef1 = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype).to(device)
+good_vae = AutoencoderKL.from_pretrained(base_model, subfolder="vae", torch_dtype=dtype).to(device)
+pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=dtype, vae=taef1).to(device)
 
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
+MAX_SEED = 2**32-1
 
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
+pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(pipe)
 
-#@spaces.GPU #[uncomment to use ZeroGPU]
-def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, progress=gr.Progress(track_tqdm=True)):
+class calculateDuration:
+    def __init__(self, activity_name=""):
+        self.activity_name = activity_name
 
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-        
-    generator = torch.Generator().manual_seed(seed)
+    def __enter__(self):
+        self.start_time = time.time()
+        return self
+    
+    def __exit__(self, exc_type, exc_value, traceback):
+        self.end_time = time.time()
+        self.elapsed_time = self.end_time - self.start_time
+        if self.activity_name:
+            print(f"Elapsed time for {self.activity_name}: {self.elapsed_time:.6f} seconds")
+        else:
+            print(f"Elapsed time: {self.elapsed_time:.6f} seconds")
+
+
+@spaces.GPU(duration=70)
+def generate_image(prompt, steps, seed, cfg_scale, width, height, lora_scale, progress):
+    pipe.to("cuda")
+    generator = torch.Generator(device="cuda").manual_seed(seed)
+    with calculateDuration("Generating image"):
+        # Generate image
+        for img in pipe.flux_pipe_call_that_returns_an_iterable_of_images(
+            prompt=prompt,
+            num_inference_steps=steps,
+            guidance_scale=cfg_scale,
+            width=width,
+            height=height,
+            generator=generator,
+            joint_attention_kwargs={"scale": lora_scale},
+            output_type="pil",
+            good_vae=good_vae,
+        ):
+            yield img
+
+
+def run_lora(prompt, cfg_scale, steps, lora_repo, lora_name, randomize_seed, seed, width, height, lora_scale, progress=gr.Progress(track_tqdm=True)):
     
-    image = pipe(
-        prompt = prompt, 
-        negative_prompt = negative_prompt,
-        guidance_scale = guidance_scale, 
-        num_inference_steps = num_inference_steps, 
-        width = width, 
-        height = height,
-        generator = generator
-    ).images[0] 
+    with calculateDuration("Unloading LoRA"):
+        pipe.unload_lora_weights()
+        
+    # Load LoRA weights
+    with calculateDuration(f"Loading LoRA weights for {lora_repo} {lora_name}"):
+       pipe.load_lora_weights(lora_repo, weight_name=lora_name)
+
+    # Set random seed for reproducibility
+    with calculateDuration("Randomizing seed"):
+        if randomize_seed:
+            seed = random.randint(0, MAX_SEED)
+
+    image_generator = generate_image(prompt, steps, seed, cfg_scale, width, height, lora_scale, progress)
     
-    return image, seed
+    # Consume the generator to get the final image
+    final_image = None
+    step_counter = 0
+    for image in image_generator:
+        step_counter+=1
+        final_image = image
+        progress_bar = f'<div class="progress-container"><div class="progress-bar" style="--current: {step_counter}; --total: {steps};"></div></div>'
+        yield image, seed, gr.update(value=progress_bar, visible=True)
+        
+    yield final_image, seed, gr.update(value=progress_bar, visible=False)
 
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
 
 css="""
 #col-container {
@@ -56,87 +105,62 @@ with gr.Blocks(css=css) as demo:
     
     with gr.Column(elem_id="col-container"):
         gr.Markdown(f"""
-        # Text-to-Image Gradio Template
+        # Simple flux with lora
         """)
         
-        with gr.Row():
-            
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-            
-            run_button = gr.Button("Run", scale=0)
-        
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-            
-            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.Column():
+
             with gr.Row():
                 
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024, #Replace with defaults that work for your model
+                prompt = gr.Text(
+                    label="Prompt",
+                    show_label=False,
+                    max_lines=1,
+                    placeholder="Enter your prompt",
+                    container=False,
                 )
                 
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024, #Replace with defaults that work for your model
-                )
-            
-            with gr.Row():
+                lora_repo = gr.Text(
+                        label="Repo",
+                        max_lines=1,
+                        placeholder="Enter a lora repo",
+                        visible=False,
+                    )
                 
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0, #Replace with defaults that work for your model
-                )
-                
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2, #Replace with defaults that work for your model
-                )
-        
-        gr.Examples(
-            examples = examples,
-            inputs = [prompt]
-        )
+                lora_name = gr.Text(
+                        label="Weights",
+                        max_lines=1,
+                        placeholder="Enter a lora weights",
+                        visible=False,
+                    )
+                run_button = gr.Button("Run", scale=0)
+
+                with gr.Accordion("Advanced Settings", open=False):
+
+                    with gr.Row():
+                        seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0, randomize=True)
+                        randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+                        lora_scale = gr.Slider(label="LoRA Scale", minimum=0, maximum=3, step=0.01, value=0.95)
+
+                    with gr.Row():
+                        width = gr.Slider(label="Width", minimum=256, maximum=1536, step=64, value=1024)
+                        height = gr.Slider(label="Height", minimum=256, maximum=1536, step=64, value=1024)
+
+                    with gr.Row():
+                        cfg_scale = gr.Slider(label="CFG Scale", minimum=1, maximum=20, step=0.5, value=3.5)
+                        steps = gr.Slider(label="Steps", minimum=1, maximum=50, step=1, value=28)
+                    
+                    
+
+            with gr.Row():
+                progress_bar = gr.Markdown(elem_id="progress",visible=False)
+                result = gr.Image(label="Result", show_label=False)
+
     gr.on(
         triggers=[run_button.click, prompt.submit],
-        fn = infer,
-        inputs = [prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
-        outputs = [result, seed]
+        fn = run_lora,
+        inputs = [prompt, cfg_scale, steps, lora_repo, lora_name, randomize_seed, seed, width, height, lora_scale],
+        outputs=[result, seed, progress_bar]
     )
 
 demo.queue().launch()

live_preview_helpers.py

@@ -0,0 +1,166 @@
+import torch
+import numpy as np
+from diffusers import FluxPipeline, AutoencoderTiny, FlowMatchEulerDiscreteScheduler
+from typing import Any, Dict, List, Optional, Union
+
+# Helper functions
+def calculate_shift(
+    image_seq_len,
+    base_seq_len: int = 256,
+    max_seq_len: int = 4096,
+    base_shift: float = 0.5,
+    max_shift: float = 1.16,
+):
+    m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
+    b = base_shift - m * base_seq_len
+    mu = image_seq_len * m + b
+    return mu
+
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
+    **kwargs,
+):
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+    if timesteps is not None:
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+# FLUX pipeline function
+@torch.inference_mode()
+def flux_pipe_call_that_returns_an_iterable_of_images(
+    self,
+    prompt: Union[str, List[str]] = None,
+    prompt_2: Optional[Union[str, List[str]]] = None,
+    height: Optional[int] = None,
+    width: Optional[int] = None,
+    num_inference_steps: int = 28,
+    timesteps: List[int] = None,
+    guidance_scale: float = 3.5,
+    num_images_per_prompt: Optional[int] = 1,
+    generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+    latents: Optional[torch.FloatTensor] = None,
+    prompt_embeds: Optional[torch.FloatTensor] = None,
+    pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+    output_type: Optional[str] = "pil",
+    return_dict: bool = True,
+    joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+    max_sequence_length: int = 512,
+    good_vae: Optional[Any] = None,
+):
+    height = height or self.default_sample_size * self.vae_scale_factor
+    width = width or self.default_sample_size * self.vae_scale_factor
+
+    # 1. Check inputs
+    self.check_inputs(
+        prompt,
+        prompt_2,
+        height,
+        width,
+        prompt_embeds=prompt_embeds,
+        pooled_prompt_embeds=pooled_prompt_embeds,
+        max_sequence_length=max_sequence_length,
+    )
+
+    self._guidance_scale = guidance_scale
+    self._joint_attention_kwargs = joint_attention_kwargs
+    self._interrupt = False
+
+    # 2. Define call parameters
+    batch_size = 1 if isinstance(prompt, str) else len(prompt)
+    device = self._execution_device
+
+    # 3. Encode prompt
+    lora_scale = joint_attention_kwargs.get("scale", None) if joint_attention_kwargs is not None else None
+    prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(
+        prompt=prompt,
+        prompt_2=prompt_2,
+        prompt_embeds=prompt_embeds,
+        pooled_prompt_embeds=pooled_prompt_embeds,
+        device=device,
+        num_images_per_prompt=num_images_per_prompt,
+        max_sequence_length=max_sequence_length,
+        lora_scale=lora_scale,
+    )
+    # 4. Prepare latent variables
+    num_channels_latents = self.transformer.config.in_channels // 4
+    latents, latent_image_ids = self.prepare_latents(
+        batch_size * num_images_per_prompt,
+        num_channels_latents,
+        height,
+        width,
+        prompt_embeds.dtype,
+        device,
+        generator,
+        latents,
+    )
+    # 5. Prepare timesteps
+    sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
+    image_seq_len = latents.shape[1]
+    mu = calculate_shift(
+        image_seq_len,
+        self.scheduler.config.base_image_seq_len,
+        self.scheduler.config.max_image_seq_len,
+        self.scheduler.config.base_shift,
+        self.scheduler.config.max_shift,
+    )
+    timesteps, num_inference_steps = retrieve_timesteps(
+        self.scheduler,
+        num_inference_steps,
+        device,
+        timesteps,
+        sigmas,
+        mu=mu,
+    )
+    self._num_timesteps = len(timesteps)
+
+    # Handle guidance
+    guidance = torch.full([1], guidance_scale, device=device, dtype=torch.float32).expand(latents.shape[0]) if self.transformer.config.guidance_embeds else None
+
+    # 6. Denoising loop
+    for i, t in enumerate(timesteps):
+        if self.interrupt:
+            continue
+
+        timestep = t.expand(latents.shape[0]).to(latents.dtype)
+
+        noise_pred = self.transformer(
+            hidden_states=latents,
+            timestep=timestep / 1000,
+            guidance=guidance,
+            pooled_projections=pooled_prompt_embeds,
+            encoder_hidden_states=prompt_embeds,
+            txt_ids=text_ids,
+            img_ids=latent_image_ids,
+            joint_attention_kwargs=self.joint_attention_kwargs,
+            return_dict=False,
+        )[0]
+        # Yield intermediate result
+        latents_for_image = self._unpack_latents(latents, height, width, self.vae_scale_factor)
+        latents_for_image = (latents_for_image / self.vae.config.scaling_factor) + self.vae.config.shift_factor
+        image = self.vae.decode(latents_for_image, return_dict=False)[0]
+        yield self.image_processor.postprocess(image, output_type=output_type)[0]
+        latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+        torch.cuda.empty_cache()
+        
+
+    # Final image using good_vae
+    latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
+    latents = (latents / good_vae.config.scaling_factor) + good_vae.config.shift_factor
+    image = good_vae.decode(latents, return_dict=False)[0]
+    self.maybe_free_model_hooks()
+    torch.cuda.empty_cache()
+    yield self.image_processor.postprocess(image, output_type=output_type)[0]

requirements.txt

@@ -1,6 +1,6 @@
-accelerate
-diffusers
-invisible_watermark
 torch
+git+https://github.com/huggingface/diffusers@3b604e8c384631e1f66a4fd9076ed5e7e2b08686
+spaces
 transformers
-xformers
+peft
+sentencepiece