GenMoji

Generate AI-transformed emoji sets locally using FLUX.2 [klein] 4B.

Author: cskillzmartin

Features

• Unicode emoji rendering to base image, then prompt-guided image editing
• FLUX.2 [klein] 4B backend (black-forest-labs/FLUX.2-klein-4B)
• WinForms desktop app
• Batch generation for all emojis or selected emoji input
• Output naming by Unicode codepoints + seed, e.g. emoji_1F600_s42.png
• Optional background removal with rembg
• Settings UI includes model download from Hugging Face and local model folder selection
• Real-time preview grid and run logging in the WinForms app

Repository Layout

src/EmojiForge.WinForms/     # Desktop UI
src/emojiforge_backend/      # Python backend (diffusion + emoji rendering + IPC)
scripts/install_deps.bat     # One-click backend dependency setup
docs/SETUP.md                # Additional setup notes

Requirements

• Windows 10/11 (WinForms target)
• Python 3.12+
• .NET 10 SDK
• NVIDIA GPU recommended (8 GB+ VRAM), CUDA-enabled PyTorch wheels are used by default
• git on PATH (required because diffusers is installed from GitHub source)

Setup

1. Clone this repo.
2. Run:
   • scripts\install_deps.bat
3. This creates src\emojiforge_backend\.venv and installs backend dependencies:
   • torch/torchvision/torchaudio (CUDA 12.8 wheels)
   • diffusers (GitHub source)
   • transformers, accelerate, Pillow, emoji, rembg, and others

Run

WinForms

dotnet run --project src/EmojiForge.WinForms

Settings and Behavior

• Default model: black-forest-labs/FLUX.2-klein-4B
• Device: cuda by default, with optional CPU offload in settings
• Generation controls: strength, steps, CFG scale, seed, output size
• Successful generations clean up intermediate debug artifacts (*.base_rgba.png, *.conditioning.png)
• If generation fails, debug artifacts are preserved for troubleshooting

Output

• Each run writes to a timestamped folder under your output directory.
• Generated files are named with Unicode codepoints and seed:
  • emoji_1F600_s123456789.png
• run_metadata.json tracks prompt, settings, progress, and failures.

Troubleshooting

• If CUDA is unavailable, verify your Python environment has CUDA-enabled Torch.
• If you hit OOM, reduce output size/steps, enable CPU offload, or switch device to cpu.
• If emoji render appears blank, verify your emoji font path (default: C:\Windows\Fonts\seguiemj.ttf).

License

• Project code: MIT (see LICENSE)
• Model weights: Apache 2.0 (FLUX.2 [klein] 4B)