ROS Package bob_flux2 

The bob_flux2 package provides a ROS 2 node for generating images using the FLUX.2 model. It supports both pure text-to-image (TTI) and image-to-image (ITTI) generation, optimized for memory efficiency by sequentially loading model components.

Features

Memory Efficient: Loads and unloads the text encoder and transformer sequentially to stay within VRAM limits.
High Quality: Quantized 4-bit support for high-quality generation on consumer GPUs.
Flexible Filenaming: Automatic filename generation with counters and random suffixes.
Image Publishing: Publishes generated images directly to the generated_image topic using cv_bridge.
GPU Optimized: Specifically optimized for consumer GPUs with 24GB VRAM (e.g., NVIDIA RTX 4090), ensuring smooth execution within these hardware limits.

Setup and Installation

1. Clone the Repository

Navigate to your ROS 2 workspace src directory and clone the package:

cd ~/ros2_ws/src
git clone https://github.com/bob-ros2/bob_flux2.git

2. Install Dependencies

It is highly recommended to use a Python virtual environment (venv) to avoid library version conflicts and maintain a clean system environment.

Once your environment is active, install the required packages:

pip install -r requirements.txt

[!NOTE] Ensure cv_bridge and sensor_msgs are present in your ROS 2 environment. These are usually included in the desktop installation or can be installed via apt install ros-<distro>-cv-bridge.

3. Build the Package

From the root of your ROS 2 workspace:

colcon build --packages-select bob_flux2
source install/setup.bash

Parameters

Parameter	Type	Default	Env Variable	Description
`repo_id`	string	`diffusers/FLUX.2-dev-bnb-4bit`	`FLUX2_REPO_ID`	The Hugging Face repository ID.
`model_dir`	string	`./models`	`FLUX2_MODEL_DIR`	Directory to cache models.
`device`	string	`cuda:0`	`FLUX2_DEVICE`	Torch device to use (e.g., `cuda:0` or `cpu`).
`prompt`	string	`''`	`FLUX2_PROMPT`	Initial prompt to run at startup.
`input_image`	string	`''`	`FLUX2_INPUT_IMAGE`	Path or URL to an input image for ITTI.
`once`	bool	`false`	`FLUX2_ONCE`	If `true`, the node exits after one generation.
`image_path`	string	`''`	`FLUX2_IMAGE_PATH`	Save path. Ends with `auto` for autogeneration.
`seed`	int	`-1`	`FLUX2_SEED`	Seed for generation (`-1` for random).
`image_counter_start`	int	`1`	`FLUX2_IMAGE_COUNTER_START`	Starting value for the filename counter.

Topics

Subscribed Topics

prompt (std_msgs/String): Receives text prompts for image generation.

Published Topics

generated_image (sensor_msgs/Image): Publishes the generated image.

Usage

Using the Example Configuration

The package includes a sample configuration file at config/tti.yaml. This is the easiest way to start the node with customized settings.

ros2 run bob_flux2 flux2_node --ros-args --params-file src/bob_flux2/config/tti.yaml

Command Line Examples

Simple Text-to-Image

ros2 run bob_flux2 flux2_node --ros-args -p prompt:="A futuristic city in the style of cyberpunk"

Save with Auto-filenaming

Setting image_path to a path ending in auto will generate files like auto_0001_rtRfTDrt.png.

ros2 run bob_flux2 flux2_node --ros-args \
    -p image_path:=/path/to/output/auto \
    -p image_counter_start:=1

Image-to-Image (ITTI)

ros2 run bob_flux2 flux2_node --ros-args \
    -p input_image:=/path/to/input.png \
    -p prompt:="Make it look like a oil painting"

ROS Package bob_flux2