Image generation with Flux.1 and OpenVINO#

This Jupyter notebook can be launched after a local installation only.

Flux is a AI image generation model developed by Black Forest Labs. It represents a significant advancement in AI-generated art, utilizing a hybrid architecture of multimodal and parallel diffusion transformer blocks and scaled to 12B parameter. The model offers state-of-the-art performance image generation with top of the line prompt following, visual quality, image detail and output diversity.More details about model can be found in blog post and original repo

In this tutorial we consider how to convert and optimized Flux.1 model using OpenVINO.

Note: Some demonstrated models can require at least 32GB RAM for conversion and running.

Table of contents:

Prerequisites
Select model
Convert model with OpenVINO
Compress model weights
Run OpenVINO model inference
Interactive demo

Installation Instructions#

This is a self-contained example that relies solely on its own code.

We recommend running the notebook in a virtual environment. You only need a Jupyter server to start. For details, please refer to Installation Guide.

Prerequisites#

%pip install -q "gradio>=4.19" "torch>=2.1"  "transformers" "nncf>=2.12.0" "diffusers>=0.30.0" "opencv-python" "pillow" "peft>=0.7.0" --extra-index-url https://download.pytorch.org/whl/cpu
%pip install -q "sentencepiece" "protobuf"
%pip install -qU --pre "openvino" --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly

Note: you may need to restart the kernel to use updated packages.
Note: you may need to restart the kernel to use updated packages.
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
openvino-dev 2024.3.0 requires openvino==2024.3.0, but you have openvino 2024.4.0.dev20240827 which is incompatible.
Note: you may need to restart the kernel to use updated packages.

import requests
from pathlib import Path

if not Path("flux_helper.py").exists():
    r = requests.get(url="https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/notebooks/flux.1-image-generation/flux_helper.py")
    open("flux_helper.py", "w").write(r.text)

if not Path("gradio_helper.py").exists():
    r = requests.get(url="https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/notebooks/flux.1-image-generatio/gradio_helper.py")
    open("gradio_helper.py", "w").write(r.text)

if not Path("notebook_utils.py").exists():
    r = requests.get(url="https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/utils/notebook_utils.py")
    open("notebook_utils.py", "w").write(r.text)

Select model#

To strike a balance between accessibility and model capabilities, FLUX.1 comes in three variants: FLUX.1-pro, FLUX.1-dev and FLUX.1-schnell: * FLUX.1-pro: The best of FLUX.1, offering state-of-the-art performance image generation with top of the line prompt following, visual quality, image detail and output diversity, but not available for public usage. * FLUX.1-dev: FLUX.1-dev is an open-weight, guidance-distilled models. Directly distilled from FLUX.1-pro, FLUX.1-dev obtains similar quality and prompt adherence capabilities, while being more efficient than a standard model of the same size. FLUX.1-dev weights are available on HuggingFace. * FLUX.1-schnell: the fastest model from Flux family is tailored for local development and personal use. FLUX.1-schnell is openly available under an Apache2.0 license. Similar, FLUX.1-dev, weights are available on HuggingFace.

Be default, we will use FLUX.1-schnell model, but you can switch to FLUX.1-dev version using widget bellow.

from flux_helper import get_model_selector

model_selector = get_model_selector()
model_selector

2024-08-28 02:20:17.242597: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable TF_ENABLE_ONEDNN_OPTS=0.
2024-08-28 02:20:17.277034: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 AVX512F AVX512_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-08-28 02:20:17.935317: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT

 Dropdown(description='Model:', options=('black-forest-labs/FLUX.1-schnell', 'black-forest-labs/FLUX.1-dev'), v…



Note: run Flux.1-dev model with notebook, you will need to accept
license agreement. You must be a registered user in Hugging Face
Hub. Please visit HuggingFace model
card,
carefully read terms of usage and click accept button. You will need
to use an access token for the code below to run. For more
information on access tokens, refer to this section of the
documentation.
You can login on Hugging Face Hub in notebook environment, using
following code:

# uncomment these lines to login to huggingfacehub to get access to pretrained model

# from huggingface_hub import notebook_login, whoami

# try:
#     whoami()
#     print('Authorization token already provided')
# except OSError:
#     notebook_login()

Convert model with OpenVINO#

Starting from 2023.0 release, OpenVINO supports PyTorch models directly via Model Conversion API. ov.convert_model function accepts instance of PyTorch model and example inputs for tracing and returns object of ov.Model class, ready to use or save on disk using ov.save_model function.

The pipeline consists of four important parts:

Clip and T5 Text Encoders to create condition to generate an image from a text prompt.
Transformer for step-by-step denoising latent image representation.
Autoencoder (VAE) for decoding latent space to image.

We will use convert_flux helper function defined in flux_helper.py that create original PyTorch model and convert each part of pipeline using ov.convert_model.

from flux_helper import convert_flux

# uncomment the line to see model conversion code
# ??convert_flux

model_dir = convert_flux(model_selector.value)

Loading pipeline components...:   0%|          | 0/7 [00:00<?, ?it/s]

You set add_prefix_space. The tokenizer needs to be converted from the slow tokenizers

Loading checkpoint shards:   0%|          | 0/2 [00:00<?, ?it/s]

⌛ Transformer model conversion started
WARNING:tensorflow:Please fix your imports. Module tensorflow.python.training.tracking.base has been moved to tensorflow.python.trackable.base. The old module will be deleted in version 2.11.

[ WARNING ]  Please fix your imports. Module %s has been moved to %s. The old module will be deleted in version %s.

✅ Transformer model conversion finished
⌛ Clip Text encoder conversion started

/opt/home/k8sworker/ci-ai/cibuilds/ov-notebook/OVNotebookOps-761/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/transformers/modeling_utils.py:4713: FutureWarning: _is_quantized_training_enabled is going to be deprecated in transformers 4.39.0. Please use model.hf_quantizer.is_trainable instead
  warnings.warn(
/opt/home/k8sworker/ci-ai/cibuilds/ov-notebook/OVNotebookOps-761/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/transformers/modeling_attn_mask_utils.py:86: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if input_shape[-1] > 1 or self.sliding_window is not None:
/opt/home/k8sworker/ci-ai/cibuilds/ov-notebook/OVNotebookOps-761/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/transformers/modeling_attn_mask_utils.py:162: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if past_key_values_length > 0:

✅ Clip Text encoder conversion finished
⌛ T5 Text encoder conversion started
✅ T5 Text encoder conversion finished
⌛ VAE decoder conversion started

/opt/home/k8sworker/ci-ai/cibuilds/ov-notebook/OVNotebookOps-761/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/diffusers/models/upsampling.py:146: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert hidden_states.shape[1] == self.channels
/opt/home/k8sworker/ci-ai/cibuilds/ov-notebook/OVNotebookOps-761/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/diffusers/models/upsampling.py:162: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if hidden_states.shape[0] >= 64:

✅ VAE decoder onversion finished
✅ black-forest-labs/FLUX.1-schnell successfully converted and can be found in FLUX.1-schnell

from flux_helper import TRANSFORMER_PATH, VAE_DECODER_PATH, TEXT_ENCODER_PATH, TEXT_ENCODER_2_PATH

model_dict = {
    "transformer": model_dir / TRANSFORMER_PATH,
    "text_encoder": model_dir / TEXT_ENCODER_PATH,
    "text_encoder_2": model_dir / TEXT_ENCODER_2_PATH,
    "vae": model_dir / VAE_DECODER_PATH,
}

Compress model weights#

For reducing model memory consumption we will use weights compression. The Weights Compression algorithm is aimed at compressing the weights of the models and can be used to optimize the model footprint and performance of large models where the size of weights is relatively larger than the size of activations, for example, Large Language Models (LLM). Compared to INT8 compression, INT4 compression improves performance even more, but introduces a minor drop in prediction quality. We will use NNCF for weight compression.

from flux_helper import weight_compression_widget

to_compress = weight_compression_widget()

to_compress

Checkbox(value=True, description='Weight compression')

import nncf
import openvino as ov
import gc

compression_args = {"mode": nncf.CompressWeightsMode.INT4_SYM, "group_size": 64, "ratio": 1.0}

int4_model_dict = {}

if to_compress.value:
    core = ov.Core()

    for model_name, model_path in model_dict.items():
        int4_path = model_path.parent / (model_path.stem + "_int4.xml")
        if not int4_path.exists():
            print(f"⌛ {model_path.stem} compression started")
            print(
                f"Compression parameters:\n\tmode = {compression_args['mode']}\n\tratio = {compression_args['ratio']}\n\tgroup_size = {compression_args['group_size']}"
            )
            model = core.read_model(model_path)
            compressed_model = nncf.compress_weights(model, **compression_args)
            ov.save_model(compressed_model, int4_path)
            print(f"✅ {model_path.stem} compression finished")
            del compressed_model
            del model
            gc.collect()
        print(f"Compressed {model_path.stem} can be found in {int4_path}")
        int4_model_dict[model_name] = int4_path

INFO:nncf:NNCF initialized successfully. Supported frameworks detected: torch, tensorflow, onnx, openvino
⌛ transformer compression started
Compression parameters:
    mode = int4_sym
    ratio = 1.0
    group_size = 64
INFO:nncf:Statistics of the bitwidth distribution:
┍━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┑
│   Num bits (N) │ % all parameters (layers)   │ % ratio-defining parameters (layers)   │
┝━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┥
│              8 │ 0% (1 / 502)                │ 0% (0 / 501)                           │
├────────────────┼─────────────────────────────┼────────────────────────────────────────┤
│              4 │ 100% (501 / 502)            │ 100% (501 / 501)                       │
┕━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┙

Output()

✅ transformer compression finished
Compressed transformer can be found in FLUX.1-schnell/transformer/transformer_int4.xml
⌛ text_encoder compression started
Compression parameters:
    mode = int4_sym
    ratio = 1.0
    group_size = 64
INFO:nncf:Statistics of the bitwidth distribution:
┍━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┑
│   Num bits (N) │ % all parameters (layers)   │ % ratio-defining parameters (layers)   │
┝━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┥
│              8 │ 33% (3 / 74)                │ 0% (0 / 71)                            │
├────────────────┼─────────────────────────────┼────────────────────────────────────────┤
│              4 │ 67% (71 / 74)               │ 100% (71 / 71)                         │
┕━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┙

Output()

✅ text_encoder compression finished
Compressed text_encoder can be found in FLUX.1-schnell/text_encoder/text_encoder_int4.xml
⌛ text_encoder_2 compression started
Compression parameters:
    mode = int4_sym
    ratio = 1.0
    group_size = 64
INFO:nncf:Statistics of the bitwidth distribution:
┍━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┑
│   Num bits (N) │ % all parameters (layers)   │ % ratio-defining parameters (layers)   │
┝━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┥
│              8 │ 4% (3 / 170)                │ 0% (0 / 167)                           │
├────────────────┼─────────────────────────────┼────────────────────────────────────────┤
│              4 │ 96% (167 / 170)             │ 100% (167 / 167)                       │
┕━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┙

Output()

✅ text_encoder_2 compression finished
Compressed text_encoder_2 can be found in FLUX.1-schnell/text_encoder_2/text_encoder_2_int4.xml
⌛ vae_decoder compression started
Compression parameters:
    mode = int4_sym
    ratio = 1.0
    group_size = 64
INFO:nncf:Statistics of the bitwidth distribution:
┍━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┑
│   Num bits (N) │ % all parameters (layers)   │ % ratio-defining parameters (layers)   │
┝━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┥
│              8 │ 98% (36 / 39)               │ 0% (0 / 3)                             │
├────────────────┼─────────────────────────────┼────────────────────────────────────────┤
│              4 │ 2% (3 / 39)                 │ 100% (3 / 3)                           │
┕━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┙

Output()

✅ vae_decoder compression finished
Compressed vae_decoder can be found in FLUX.1-schnell/vae/vae_decoder_int4.xml

Run OpenVINO model inference#

OVFluxPipeline class defined in flux_helper.py provides convenient way for running model. It accepts directory with converted model and inference device as arguments.

from flux_helper import get_pipeline_selection_option

use_compressed = get_pipeline_selection_option(int4_model_dict)
use_compressed

Checkbox(value=True, description='Use compressed models')

from flux_helper import OVFluxPipeline, init_pipeline  # noqa: F401

# uncomment the line to see model pipeline
# ??OVFluxPipeline

from notebook_utils import device_widget

device = device_widget(default="CPU", exclude=["NPU"])
device

Dropdown(description='Device:', options=('CPU', 'AUTO'), value='CPU')

ov_pipe = init_pipeline(model_dir, model_dict if not use_compressed.value else int4_model_dict, device.value)

Models compilation
✅ transformer - Done!
✅ text_encoder - Done!
✅ text_encoder_2 - Done!
✅ vae - Done!

import torch

prompt = "A cat holding a sign that says hello OpenVINO"
image = ov_pipe(
    prompt, guidance_scale=0.0, num_inference_steps=4, max_sequence_length=256, generator=torch.Generator("cpu").manual_seed(0), height=256, width=256
).images[0]

image

0%|          | 0/4 [00:00<?, ?it/s]

../_images/flux.1-image-generation-with-output_20_1.png

Interactive demo#

from gradio_helper import make_demo

demo = make_demo(ov_pipe)

# if you are launching remotely, specify server_name and server_port
#  demo.launch(server_name='your server name', server_port='server port in int')
# if you have any issue to launch on your platform, you can pass share=True to launch method:
# demo.launch(share=True)
# it creates a publicly shareable link for the interface. Read more in the docs: https://gradio.app/docs/
try:
    demo.launch(debug=False)
except Exception:
    demo.launch(debug=False, share=True)

Running on local URL:  http://127.0.0.1:7860

To create a public link, set share=True in launch().