TextFlux: An OCR-Free DiT Model for High-Fidelity Multilingual Scene Text Synthesis

English | 中文简体

Thanks to ComfyUI's quantization optimizations for Flux, this model can be run directly on consumer-grade GPUs like the 4090.

How to Use

1. You need to manually download the Flux-related model weights. Following the official Flux workflow, place the models in the ComfyUI/models directory. The required weights are listed below. Only the LoRA weights are an additional download. You can also download the full parameter model textflux to replace flux1-fill-dev and the LoRA for an even better text editing experience.

   • Diffusion Model: diffusion_models/flux1-fill-dev.safetensors
   • Text Encoder: clip/t5xxl_fp16.safetensors
   • VAE: ae/ae.safetensors
   • LoRA: lora/pytorch_lora_weights.safetensors

2. Clone this project. You can directly use the workflow from example_workflows/textflux_lora_inpaint_example.json.

   Bash

   # Navigate to your ComfyUI custom nodes directory
   cd ComfyUI/custom_nodes
   
   # Clone the repository
   git clone https://github.com/yyyyyxie/textflux_comfyui.git
   
   # Optional: Copy the example workflow to your ComfyUI user workflows
   # Note: Adjust the path if your user workflow directory is different
   cp example_workflows/textflux_lora_inpaint_example.json ComfyUI/user/default/workflows
   

3. Start ComfyUI !