Automated ComfyUI workflow that creates a custom LoRA model from just one reference image.
- LLM Prompt Generation: Gemini analyzes your image and generates 20 diverse prompts
- Image Generation: FLUX.1 Kontext creates 20 consistent variations using those prompts
- Dataset Creation: Images are automatically saved with caption files
- LoRA Training: Integrated training directly in ComfyUI
Install via ComfyUI Manager:
ComfyUI-Impact-PackWAS Node SuiteComfyUI_Fill-NodesComfyUI-KJNodescomfyui_tinyterranodesComfyUI-FluxTrainerComfyUI-SP-NodesComfyUI-Custom-Scripts
- FLUX.1 Kontext model (
flux1-dev-kontext_fp8_scaled.safetensors) - FLUX VAE (
ae.safetensors) - FLUX CLIP encoders (
clip_l.safetensors,t5xxl_fp8_e4m3fn_scaled.safetensors) - Training models (
flux1-dev.safetensors,t5xxl_fp16.safetensors)
- Load the
workflow.jsonfile into ComfyUI - Connect your reference image to the Load Image node
- Set your Gemini API key in the FL_GeminiVideoCaptioner node
- Configure:
- Trigger word: Unique word for your LoRA
- Folder name: Where to save dataset and LoRA
- Output path: Path to ComfyUI output directory
- Run the workflow
Your trained LoRA will be saved in the specified folder.