Standalone FM NewGens

Generate NewGen images for Football Manager using local Stable Diffusion

🖥️ GUI Available! - Now includes a user-friendly graphical interface

Intro

• "There are 10000s of faces out there already"
• "There are tools that work the same already?"
• "Have you got nothing better to do with your time?"

Example video here

These are some of the questions I think you are going to ask me, but I present to you something I have been working on, Standalone FM NewGens

The Idea

Taking inspiration from fm-ai-face-generator and NewGAN-Manager, this tool generates unique face images for Football Manager NewGens using local Stable Diffusion instead of requiring a separate ComfyUI server.

Thanks to both original projects for the inspiration!

So what does this do? Why would you want to use it? (you probably won't but personally, I like a bit more personalization from my NewGens, some I get quite attached to :) )

• This will create images for your NewGens, based off of information provided from the FM database, making a player's image more unique to that player
• Generate images locally using Stable Diffusion (no external server required)
• Download the images to the specified output directory, all processed players images will be saved as their uid from the game
• Remove all the background from the generated faces
• Create the needed config file for FM to load the faces
• You can then rename the generated folder and place it in your graphics folder as you would normally and voila!

Install Guide

Things you will need that I will not be going over on how to setup.

• Python 3.8+ installed
• Git installed
• At least 4GB of RAM (8GB+ recommended for better performance)
• CUDA-compatible GPU (optional, but highly recommended for faster generation)

Setup Instructions

• Use Git to checkout this repo
• You need to get a rtf file of the players you want to add images for, to do this you need the view and filter supplied with the project
• Copy the filters and views folder over to your Football Manager 2024 data folder in Documents, this may create these folders, or may just add the contained files into your existing folders
  • You can use python INSTALL_VIEW_AND_FILTER.py to do this automatically
• Included is the original is newgen filter created by the NewGAN-Manager team and a new view created by myself to get the needed data
• If you follow this video it will show you how to export the rtf file, you want to use our view, not the view in the video
• Once you have your rtf file add it to the root of the current repo
• Create a python virtual environment python -m venv venv
• Activate the venv .\venv\Scripts\activate (Windows) or source venv/bin/activate (Linux/Mac)
• Install the requirements pip install -r requirements.txt
• Copy the user_config.cfg.sample to user_config.cfg and make the needed changes
  • football_manager_version - Version of FM to generate for. Defaults to 2024
  • output_dir - Where to save the generated set. Defaults to ./generated_images/
  • rtf_file - Path to your exported RTF file with player data
  • model_name - Hugging Face model to use (see Model Selection section below)
  • model_dir - Custom directory to store models (optional)

Usage Options

🖥️ GUI Mode (Recommended)

For the best user experience, use the graphical interface:

# Test if GUI works on your system
python test_gui.py

# Launch the full GUI
python run_gui.py

GUI Features:

• Easy configuration with file browsers
• Real-time progress monitoring
• Live image preview
• Start/stop controls
• Comprehensive logging

📝 Command Line Mode

For advanced users or automation:

python comfy_fm_newgen.py

Command Line Features:

• Fast batch processing
• Resume capability
• Detailed logging
• Script automation

Model Selection

The tool uses Hugging Face models for image generation. You can specify which model to use in your user_config.cfg:

Recommended Models for Face Generation:

• SG161222/Realistic_Vision_V6.0_B1 - Best choice for realistic faces (default)
• digiplay/AbsoluteReality - Photorealistic faces and people
• stabilityai/stable-diffusion-2-1 - General purpose, good for faces
• runwayml/stable-diffusion-v1-5 - Classic SD 1.5, widely supported

Model Storage:

• Default: Models are stored in Hugging Face cache (~/.cache/huggingface/)
• Custom Directory: Set model_dir in config to store models in a specific folder
• Model Size: Most face models are 2-5GB, ensure you have sufficient disk space
• First Run: Model downloads automatically when you first run the tool

Changing Models:

1. Edit model_name in your user_config.cfg
2. Delete the old model from your cache/directory if desired
3. Restart the tool - it will download the new model automatically

System Requirements

• CPU Mode: Requires at least 8GB RAM, expect 2-5 minutes per image
• GPU Mode: Requires CUDA-compatible GPU with at least 4GB VRAM, expect 10-30 seconds per image
• First run will download the Stable Diffusion model (~5GB), so ensure you have sufficient disk space

Features

Core Features

• Local Generation: No external servers or API keys required
• Model Selection: Choose from various face generation models via configuration
• Automatic Processing: Batch process multiple players from your RTF file
• Background Removal: Automatically removes backgrounds from generated faces
• XML Configuration: Generates proper Football Manager configuration files
• Resume Support: Can resume from previous runs without regenerating existing images
• Model Caching: Downloads and caches models locally for faster subsequent runs

🖥️ GUI Features

• User-Friendly Interface: Modern tabbed interface with intuitive controls
• File Browsers: Easy selection of RTF files and output directories
• Real-Time Progress: Live progress bars and status updates
• Image Preview: View generated images as they're created
• Configuration Management: Save and load different configurations
• Error Handling: Clear error messages and troubleshooting tips
• Multi-threaded: Non-blocking interface during generation

GUI Troubleshooting

Common Issues:

GUI won't start:

• Ensure Python 3.8+ is installed
• Install requirements: pip install -r requirements.txt
• Check for missing dependencies: python -c "import tkinter"

Dependency installation fails:

• Try installing packages individually: pip install torch torchvision diffusers transformers
• For Python 3.11+, use: pip install rembg==2.0.59 (instead of 2.0.60)
• If striprtf fails, try: pip install striprtf==0.0.28
• Use alternative requirements: pip install -r requirements-gui.txt for GUI-only

Numba/rembg compatibility issues:

• The application now handles missing rembg gracefully
• Background removal is optional - images will work without it
• Test GUI functionality: python test_gui.py
• If you encounter Numba compilation errors, the GUI will still work

Alternative installation methods:

• Conda: conda install pytorch torchvision torchaudio cpuonly -c pytorch
• GPU Support: Replace cpuonly with cudatoolkit=11.8 for CUDA support
• Individual packages: Install problematic packages one by one to identify issues

Model download fails:

• Check internet connection
• Verify disk space (models are 2-5GB)
• Try a different model from the recommended list

Images not appearing in preview:

• Check output directory permissions
• Ensure images are being generated successfully
• Try refreshing the preview tab

Configuration not saving:

• Check file permissions in the project directory
• Ensure no other instances are running
• Try running as administrator

Performance Tips:

• First run takes longer due to model download
• Close other applications during generation for better performance
• Use SSD storage for faster model loading
• Monitor system resources - generation is CPU/GPU intensive

Command Line Usage

For advanced users or automation, you can still use the original command-line interface:

python comfy_fm_newgen.py

The GUI and command-line versions share the same configuration files and generated data.

---

You should get some console output of the progress, good luck!

Open an issue here if you're having problems

Thanks