clean deploy HF

.gitignore

@@ -0,0 +1,2 @@
+.idea
+.venv

README.md

@@ -5,9 +5,11 @@ colorFrom: indigo
 colorTo: yellow
 sdk: gradio
 sdk_version: 5.29.0
-app_file: app.py
+app_file: app_hf.py
 pinned: false
 license: other
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+Flex.2-Preview experiment by [JustLab.ai](https://justlab.ai)
+
+Credits to [Ostris](https://huggingface.co/ostris) for this amazing all-in-one model

app.py

@@ -0,0 +1,196 @@
+import gradio as gr
+import torch
+from diffusers import DiffusionPipeline
+import gc
+from pipeline import Flex2Pipeline
+
+# Global variable to store the pipeline
+pipe = None
+
+def load_model(model_id="ostris/Flex.2-preview", device="cuda"):
+    """Load and cache the model to avoid reloading for each inference"""
+    global pipe
+    
+    if pipe is None:
+        print(f"Loading {model_id}...")
+        try:
+            # Load the model components directly using DiffusionPipeline
+            # This avoids trying to use custom_pipeline which is causing issues
+            components = DiffusionPipeline.from_pretrained(
+                model_id,
+                torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32
+            ).components
+            
+            # Create our custom Flex2Pipeline with the components
+            pipe = Flex2Pipeline(
+                scheduler=components["scheduler"],
+                vae=components["vae"],
+                text_encoder=components["text_encoder"],
+                tokenizer=components["tokenizer"],
+                text_encoder_2=components["text_encoder_2"],
+                tokenizer_2=components["tokenizer_2"],
+                transformer=components["transformer"],
+            )
+            
+            # Move to device
+            pipe = pipe.to(device)
+            print("Model loaded successfully!")
+        except Exception as e:
+            print(f"Error loading model: {e}")
+            return None
+    
+    # Enable TF32 precision if available
+    if torch.cuda.is_available():
+        torch.backends.cuda.matmul.allow_tf32 = True
+        torch.backends.cudnn.allow_tf32 = True
+    
+    return pipe
+
+def clear_gpu_memory():
+    """Clear GPU memory"""
+    global pipe
+    if pipe is not None:
+        del pipe
+        pipe = None
+    gc.collect()
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+        torch.cuda.ipc_collect()
+    return "GPU memory cleared"
+
+def generate_image(
+    prompt,
+    prompt_2=None,
+    inpaint_image=None,
+    inpaint_mask=None,
+    control_image=None,
+    control_strength=1.0,
+    control_stop=1.0,
+    height=1024,
+    width=1024,
+    num_inference_steps=28,
+    guidance_scale=3.5,
+    seed=-1,
+    progress=gr.Progress()
+):
+    """Generate image using Flex2Pipeline"""
+    global pipe
+    
+    # Load model if not already loaded
+    pipe = load_model()
+    if pipe is None:
+        return None, "Error: Failed to load the model. Please check logs."
+    
+    # Prepare generator for deterministic output
+    generator = None
+    if seed != -1:
+        generator = torch.Generator("cuda" if torch.cuda.is_available() else "cpu").manual_seed(seed)
+    else:
+        # Generate a random seed
+        seed = torch.randint(0, 2**32 - 1, (1,)).item()
+        generator = torch.Generator("cuda" if torch.cuda.is_available() else "cpu").manual_seed(seed)
+    
+    # Create callback for progress updates
+    def callback_on_step_end(pipe, i, t, callback_kwargs):
+        progress((i + 1) / pipe._num_timesteps)
+        return callback_kwargs
+    
+    try:
+        # Run the pipeline
+        output = pipe(
+            prompt=prompt,
+            prompt_2=prompt_2 if prompt_2 and prompt_2.strip() else None,
+            inpaint_image=inpaint_image,
+            inpaint_mask=inpaint_mask,
+            control_image=control_image,
+            control_strength=float(control_strength),
+            control_stop=float(control_stop),
+            height=height,
+            width=width,
+            num_inference_steps=num_inference_steps,
+            guidance_scale=guidance_scale,
+            generator=generator,
+            callback_on_step_end=callback_on_step_end,
+        )
+        
+        # Return the generated image and success message
+        return output.images[0], f"Successfully generated image with seed: {seed}"
+    
+    except Exception as e:
+        error_message = f"Error during image generation: {str(e)}"
+        print(error_message)
+        return None, error_message
+
+# Create Gradio Interface
+with gr.Blocks() as demo:
+    gr.Markdown("# Flex.2 Image Generator")
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            # Input parameters
+            prompt = gr.Textbox(label="Prompt", placeholder="Enter your prompt here...", lines=3)
+            prompt_2 = gr.Textbox(label="Secondary Prompt (Optional)", placeholder="Optional secondary prompt...", lines=2)
+            
+            with gr.Accordion("Image Settings", open=True):
+                with gr.Row():
+                    height = gr.Slider(minimum=256, maximum=1536, value=1024, step=64, label="Height")
+                    width = gr.Slider(minimum=256, maximum=1536, value=1024, step=64, label="Width")
+                with gr.Row():
+                    num_inference_steps = gr.Slider(minimum=1, maximum=100, value=28, step=1, label="Inference Steps")
+                    guidance_scale = gr.Slider(minimum=1.0, maximum=20.0, value=3.5, step=0.1, label="Guidance Scale")
+                seed = gr.Number(label="Seed (-1 for random)", value=-1)
+            
+            with gr.Accordion("Control Settings", open=False):
+                control_image = gr.Image(label="Control Image (Optional)", type="pil")
+                with gr.Row():
+                    control_strength = gr.Slider(minimum=0.0, maximum=2.0, value=1.0, step=0.05, label="Control Strength")
+                    control_stop = gr.Slider(minimum=0.0, maximum=1.0, value=1.0, step=0.05, label="Control Stop")
+            
+            with gr.Accordion("Inpainting Settings", open=False):
+                inpaint_image = gr.Image(label="Initial Image for Inpainting", type="pil")
+                inpaint_mask = gr.Image(label="Mask Image (White areas will be inpainted)", type="pil")
+            
+            # Generate button
+            generate_button = gr.Button("Generate Image", variant="primary")
+            
+            # Clear GPU memory button
+            clear_button = gr.Button("Clear GPU Memory")
+            
+            # Status message
+            status_message = gr.Textbox(label="Status", interactive=False)
+        
+        with gr.Column(scale=1):
+            # Output image
+            output_image = gr.Image(label="Generated Image")
+    
+    # Connect buttons to functions
+    generate_button.click(
+        fn=generate_image, 
+        inputs=[
+            prompt, prompt_2, inpaint_image, inpaint_mask, control_image, 
+            control_strength, control_stop, height, width, 
+            num_inference_steps, guidance_scale, seed
+        ],
+        outputs=[output_image, status_message]
+    )
+    
+    clear_button.click(fn=clear_gpu_memory, outputs=status_message)
+    
+    # Examples
+    gr.Examples(
+        [
+            ["A beautiful landscape with mountains and a lake", None, None, None, None, 1.0, 1.0, 1024, 1024, 28, 3.5, 42],
+            ["A cyberpunk cityscape at night with neon lights", "High quality, detailed", None, None, None, 1.0, 1.0, 1024, 1024, 28, 7.0, 1234],
+        ],
+        fn=generate_image,
+        inputs=[
+            prompt, prompt_2, inpaint_image, inpaint_mask, control_image, 
+            control_strength, control_stop, height, width, 
+            num_inference_steps, guidance_scale, seed
+        ],
+        outputs=[output_image, status_message],
+    )
+
+# Launch the app with queue enabled
+if __name__ == "__main__":
+    demo.queue(concurrency_count=1).launch(share=False)

app_hf.py

@@ -0,0 +1,194 @@
+import gradio as gr
+import spaces
+import torch
+from diffusers import DiffusionPipeline
+import gc
+from pipeline import Flex2Pipeline
+
+# Global variable to store the pipeline
+pipe = None
+
+@spaces.GPU
+def load_model(model_id="ostris/Flex.2-preview", device="cuda"):
+    """Load and cache the model to avoid reloading for each inference"""
+    global pipe
+    
+    if pipe is None:
+        print(f"Loading {model_id}...")
+        try:
+            # Load the model components directly using DiffusionPipeline
+            # This avoids trying to use custom_pipeline which is causing issues
+            components = DiffusionPipeline.from_pretrained(
+                model_id,
+                torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32
+            ).components
+            
+            # Create our custom Flex2Pipeline with the components
+            pipe = Flex2Pipeline(
+                scheduler=components["scheduler"],
+                vae=components["vae"],
+                text_encoder=components["text_encoder"],
+                tokenizer=components["tokenizer"],
+                text_encoder_2=components["text_encoder_2"],
+                tokenizer_2=components["tokenizer_2"],
+                transformer=components["transformer"],
+            )
+            
+            # Move to device
+            pipe = pipe.to(device)
+            print("Model loaded successfully!")
+        except Exception as e:
+            print(f"Error loading model: {e}")
+            return None
+    
+    # Enable TF32 precision if available
+    if torch.cuda.is_available():
+        torch.backends.cuda.matmul.allow_tf32 = True
+        torch.backends.cudnn.allow_tf32 = True
+    
+    return pipe
+
+def clear_gpu_memory():
+    """Clear GPU memory"""
+    global pipe
+    if pipe is not None:
+        del pipe
+        pipe = None
+    gc.collect()
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+        torch.cuda.ipc_collect()
+    return "GPU memory cleared"
+
+@spaces.GPU
+def generate_image(
+    prompt,
+    prompt_2=None,
+    inpaint_image=None,
+    inpaint_mask=None,
+    control_image=None,
+    control_strength=1.0,
+    control_stop=1.0,
+    height=1024,
+    width=1024,
+    num_inference_steps=28,
+    guidance_scale=3.5,
+    seed=-1,
+    progress=gr.Progress()
+):
+    """Generate image using Flex2Pipeline"""
+    global pipe
+    
+    # Load model if not already loaded
+    pipe = load_model()
+    if pipe is None:
+        return None, "Error: Failed to load the model. Please check logs."
+    
+    # Prepare generator for deterministic output
+    generator = None
+    if seed != -1:
+        generator = torch.Generator("cuda" if torch.cuda.is_available() else "cpu").manual_seed(seed)
+    else:
+        # Generate a random seed
+        seed = torch.randint(0, 2**32 - 1, (1,)).item()
+        generator = torch.Generator("cuda" if torch.cuda.is_available() else "cpu").manual_seed(seed)
+    
+    # Create callback for progress updates
+    def callback_on_step_end(pipe, i, t, callback_kwargs):
+        progress((i + 1) / pipe._num_timesteps)
+        return callback_kwargs
+    
+    try:
+        # Run the pipeline
+        output = pipe(
+            prompt=prompt,
+            prompt_2=prompt_2 if prompt_2 and prompt_2.strip() else None,
+            inpaint_image=inpaint_image,
+            inpaint_mask=inpaint_mask,
+            control_image=control_image,
+            control_strength=float(control_strength),
+            control_stop=float(control_stop),
+            height=height,
+            width=width,
+            num_inference_steps=num_inference_steps,
+            guidance_scale=guidance_scale,
+            generator=generator,
+            callback_on_step_end=callback_on_step_end,
+        )
+        
+        # Return the generated image and success message
+        return output.images[0], f"Successfully generated image with seed: {seed}"
+    
+    except Exception as e:
+        error_message = f"Error during image generation: {str(e)}"
+        print(error_message)
+        return None, error_message
+
+# Create Gradio Interface
+with gr.Blocks() as demo:
+    gr.Markdown("# Flex.2 generator by [JustLab.ai](https://justlab.ai)")
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            # Input parameters
+            prompt = gr.Textbox(label="Prompt", placeholder="Enter your prompt here...", lines=3)
+            prompt_2 = gr.Textbox(label="Secondary Prompt (Optional)", placeholder="Optional secondary prompt...", lines=2)
+            
+            with gr.Accordion("Image Settings", open=True):
+                with gr.Row():
+                    height = gr.Slider(minimum=256, maximum=1536, value=1024, step=64, label="Height")
+                    width = gr.Slider(minimum=256, maximum=1536, value=1024, step=64, label="Width")
+                with gr.Row():
+                    num_inference_steps = gr.Slider(minimum=1, maximum=100, value=28, step=1, label="Inference Steps")
+                    guidance_scale = gr.Slider(minimum=1.0, maximum=20.0, value=3.5, step=0.1, label="Guidance Scale")
+                seed = gr.Number(label="Seed (-1 for random)", value=-1)
+            
+            with gr.Accordion("Control Settings (You can use LoRA generated images.)", open=False):
+                control_image = gr.Image(label="Control Image (Optional)", type="pil")
+                with gr.Row():
+                    control_strength = gr.Slider(minimum=0.0, maximum=2.0, value=1.0, step=0.05, label="Control Strength")
+                    control_stop = gr.Slider(minimum=0.0, maximum=1.0, value=1.0, step=0.05, label="Control Stop")
+            
+            with gr.Accordion("Inpainting Settings", open=False):
+                inpaint_image = gr.Image(label="Initial Image for Inpainting", type="pil")
+                inpaint_mask = gr.Image(label="Mask Image (White areas will be inpainted)", type="pil")
+            
+            # Generate button
+            generate_button = gr.Button("Generate Image", variant="primary")
+            
+            # Status message
+            status_message = gr.Textbox(label="Status", interactive=False)
+        
+        with gr.Column(scale=1):
+            # Output image
+            output_image = gr.Image(label="Generated Image")
+    
+    # Connect buttons to functions
+    generate_button.click(
+        fn=generate_image, 
+        inputs=[
+            prompt, prompt_2, inpaint_image, inpaint_mask, control_image, 
+            control_strength, control_stop, height, width, 
+            num_inference_steps, guidance_scale, seed
+        ],
+        outputs=[output_image, status_message]
+    )
+    
+    # Examples
+    gr.Examples(
+        [
+            ["A beautiful landscape with mountains and a lake", None, None, None, None, 1.0, 1.0, 1024, 1024, 28, 3.5, 42],
+            ["A cyberpunk cityscape at night with neon lights", "High quality, detailed", None, None, None, 1.0, 1.0, 1024, 1024, 28, 7.0, 1234],
+        ],
+        fn=generate_image,
+        inputs=[
+            prompt, prompt_2, inpaint_image, inpaint_mask, control_image, 
+            control_strength, control_stop, height, width, 
+            num_inference_steps, guidance_scale, seed
+        ],
+        outputs=[output_image, status_message],
+    )
+
+# Configure for HF Spaces
+# Disable API endpoints
+demo.queue().launch(show_api=False)

pipeline.py

@@ -0,0 +1,440 @@
+from diffusers import FluxControlPipeline, FluxTransformer2DModel
+from typing import Any, Callable, Dict, List, Optional, Union
+import torch
+
+from diffusers.image_processor import PipelineImageInput
+import numpy as np
+import torch.nn.functional as F
+from diffusers.pipelines.flux.pipeline_output import FluxPipelineOutput
+from diffusers.pipelines.flux.pipeline_flux import calculate_shift, retrieve_timesteps, XLA_AVAILABLE
+
+
+class Flex2Pipeline(FluxControlPipeline):
+    def __init__(
+        self,
+        scheduler,
+        vae,
+        text_encoder,
+        tokenizer,
+        text_encoder_2,
+        tokenizer_2,
+        transformer,
+    ):
+        super().__init__(scheduler, vae, text_encoder, tokenizer, text_encoder_2, tokenizer_2, transformer)
+    
+    def check_inputs(
+        self,
+        prompt,
+        prompt_2,
+        height,
+        width,
+        prompt_embeds=None,
+        pooled_prompt_embeds=None,
+        callback_on_step_end_tensor_inputs=None,
+        max_sequence_length=None,
+        inpaint_image=None,
+        inpaint_mask=None,
+        control_image=None,
+    ):
+        super().check_inputs(
+            prompt,
+            prompt_2,
+            height,
+            width,
+            prompt_embeds=prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
+            max_sequence_length=max_sequence_length,
+        )
+        if inpaint_image is not None and inpaint_mask is None:
+            raise ValueError(
+                "If `inpaint_image` is passed, `inpaint_mask` must be passed as well. "
+                "Please make sure to pass both `inpaint_image` and `inpaint_mask`."
+            )
+        if inpaint_mask is not None and inpaint_image is None:
+            raise ValueError(
+                "If `inpaint_mask` is passed, `inpaint_image` must be passed as well. "
+                "Please make sure to pass both `inpaint_image` and `inpaint_mask`."
+            )
+    
+    @torch.no_grad()
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        prompt_2: Optional[Union[str, List[str]]] = None,
+        inpaint_image: Optional[PipelineImageInput] = None,
+        inpaint_mask: Optional[PipelineImageInput] = None,
+        control_image: Optional[PipelineImageInput] = None,
+        control_strength: Optional[float] = 1.0,
+        control_stop: Optional[float] = 1.0,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 28,
+        sigmas: Optional[List[float]] = 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,
+        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        max_sequence_length: int = 512,
+        **kwargs,
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                will be used instead
+            inpaint_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+                    `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+                The image to be inpainted.
+            inpaint_mask (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+                    `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+                A black and white mask to be used for inpainting. The white pixels are the areas to be inpainted, while the
+                black pixels are the areas to be kept.
+            control_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+                    `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+                The control image (line, depth, pose, etc.) to be used for the generation. The control image
+            control_strength (`float`, *optional*, defaults to 1.0):
+                The strength of the control image. The higher the value, the more the control image will be used to
+                guide the generation. The lower the value, the less the control image will be used to guide the
+                generation.
+            control_stop (`float`, *optional*, defaults to 1.0):
+                The percentage of the generation to drop out the control. 0.0 to 1.0.  0.5 mean the control will be dropped
+                out at 50% of the generation.
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The height in pixels of the generated image. This is set to 1024 by default for the best results.
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The width in pixels of the generated image. This is set to 1024 by default for the best results.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
+            guidance_scale (`float`, *optional*, defaults to 3.5):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+                to make generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.flux.FluxPipelineOutput`] instead of a plain tuple.
+            joint_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            callback_on_step_end (`Callable`, *optional*):
+                A function that calls at the end of each denoising steps during the inference. The function is called
+                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+                `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
+            max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.
+
+        Examples:
+
+        Returns:
+            [`~pipelines.flux.FluxPipelineOutput`] or `tuple`: [`~pipelines.flux.FluxPipelineOutput`] if `return_dict`
+            is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the generated
+            images.
+        """
+
+        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. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            prompt_2,
+            height,
+            width,
+            prompt_embeds=prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
+            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
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        device = self._execution_device
+
+        # 3. Prepare text embeddings
+        lora_scale = (
+            self.joint_attention_kwargs.get("scale", None) if self.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
+        
+        # only prepare latents for non controls
+        # (16 + 1 + 16 )
+        num_control_channels = 33
+        num_channels_latents = num_channels_latents - num_control_channels
+        
+        control_latents = None
+        inpaint_latents = None
+        inpaint_latents_mask = None
+        
+        latent_height = height // self.vae_scale_factor
+        latent_width = width // self.vae_scale_factor
+        
+        # process the control and inpaint channels
+        
+        if control_image is None:
+            control_latents = torch.zeros(
+                batch_size * num_images_per_prompt,
+                16,
+                latent_height,
+                latent_width,
+                device=device,
+                dtype=self.vae.dtype,
+            )
+        else:
+            control_image = self.prepare_image(
+                image=control_image,
+                width=width,
+                height=height,
+                batch_size=batch_size * num_images_per_prompt,
+                num_images_per_prompt=num_images_per_prompt,
+                device=device,
+                dtype=self.vae.dtype,
+            )
+            control_image = self.vae.encode(control_image).latent_dist.sample(generator=generator)
+            control_latents = (control_image - self.vae.config.shift_factor) * self.vae.config.scaling_factor
+        
+        # apply control strength
+        control_latents = control_latents * control_strength
+        
+        if inpaint_image is None and inpaint_mask is None:
+            inpaint_latents = torch.zeros(
+                batch_size * num_images_per_prompt,
+                16,
+                latent_height,
+                latent_width,
+                device=device,
+                dtype=self.vae.dtype,
+            )
+            inpaint_latents_mask = torch.ones(
+                batch_size * num_images_per_prompt,
+                1,
+                latent_height,
+                latent_width,
+                device=device,
+                dtype=self.vae.dtype,
+            )
+        else:
+            inpaint_image = self.prepare_image(
+                image=inpaint_image,
+                width=width,
+                height=height,
+                batch_size=batch_size * num_images_per_prompt,
+                num_images_per_prompt=num_images_per_prompt,
+                device=device,
+                dtype=self.vae.dtype,
+            )
+            inpaint_image = self.vae.encode(inpaint_image).latent_dist.sample(generator=generator)
+            inpaint_latents = (inpaint_image - self.vae.config.shift_factor) * self.vae.config.scaling_factor
+            height_inpaint_image, width_inpaint_image = control_image.shape[2:]
+            
+            inpaint_mask = self.prepare_image(
+                image=inpaint_mask,
+                width=width,
+                height=height,
+                batch_size=batch_size * num_images_per_prompt,
+                num_images_per_prompt=num_images_per_prompt,
+                device=device,
+                dtype=self.vae.dtype,
+            )
+            # mask is 3 ch -1 to 1. make it 1ch, 0 to 1
+            inpaint_mask = inpaint_mask[:, 0:1, :, :] * 0.5 + 0.5
+            # resize to match height_inpaint_image and width_inpaint_image
+            inpaint_latents_mask = F.interpolate(inpaint_mask, size=(height_inpaint_image, width_inpaint_image), mode="bilinear", align_corners=False)
+        
+        # apply inverted mask to inpaint latents
+        inpaint_latents = inpaint_latents * (1 - inpaint_latents_mask)
+        
+        # concat the latent controls on the channel dimension every step  
+        latent_controls = torch.cat([inpaint_latents, inpaint_latents_mask, control_latents], dim=1)
+        latent_no_controls = torch.cat([inpaint_latents, inpaint_latents_mask, torch.zeros_like(control_latents)], dim=1)
+        
+        # pack the controls
+        height_latent_controls, width_latent_controls = latent_controls.shape[2:]
+        packed_latent_controls = self._pack_latents(
+            latent_controls,
+            batch_size * num_images_per_prompt,
+            num_control_channels,
+            height_latent_controls,
+            width_latent_controls,
+        )
+        packed_latent_no_controls = self._pack_latents(
+            latent_no_controls,
+            batch_size * num_images_per_prompt,
+            num_control_channels,
+            height_latent_controls,
+            width_latent_controls,
+        )
+
+        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) if sigmas is None else sigmas
+        image_seq_len = latents.shape[1]
+        mu = calculate_shift(
+            image_seq_len,
+            self.scheduler.config.get("base_image_seq_len", 256),
+            self.scheduler.config.get("max_image_seq_len", 4096),
+            self.scheduler.config.get("base_shift", 0.5),
+            self.scheduler.config.get("max_shift", 1.15),
+        )
+        timesteps, num_inference_steps = retrieve_timesteps(
+            self.scheduler,
+            num_inference_steps,
+            device,
+            sigmas=sigmas,
+            mu=mu,
+        )
+        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+        self._num_timesteps = len(timesteps)
+
+        # handle guidance
+        if self.transformer.config.guidance_embeds:
+            guidance = torch.full([1], guidance_scale, device=device, dtype=torch.float32)
+            guidance = guidance.expand(latents.shape[0])
+        else:
+            guidance = None
+            
+        control_cutoff = int(len(timesteps) * control_stop)
+
+        # 6. Denoising loop
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if self.interrupt:
+                    continue
+                
+                control_latents = packed_latent_controls if i < control_cutoff else packed_latent_no_controls
+
+                latent_model_input = torch.cat([latents, control_latents], dim=2)
+
+                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+                timestep = t.expand(latents.shape[0]).to(latents.dtype)
+
+                noise_pred = self.transformer(
+                    hidden_states=latent_model_input,
+                    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]
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents_dtype = latents.dtype
+                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+
+                if latents.dtype != latents_dtype:
+                    if torch.backends.mps.is_available():
+                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                        latents = latents.to(latents_dtype)
+
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
+        if output_type == "latent":
+            image = latents
+        else:
+            latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
+            latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor
+            image = self.vae.decode(latents, return_dict=False)[0]
+            image = self.image_processor.postprocess(image, output_type=output_type)
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (image,)
+
+        return FluxPipelineOutput(images=image)
+
+    

requirements.txt

@@ -0,0 +1,14 @@
+torch>=2.0.0
+transformers>=4.30.0
+diffusers>=0.24.0
+accelerate>=0.20.0
+gradio>=5.29.0
+pillow>=9.0.0
+numpy>=1.20.0
+safetensors>=0.3.0
+ftfy>=6.1.0
+xformers>=0.0.20
+huggingface-hub>=0.19.0
+torchao>=0.1.0
+spaces
+sentencepiece