restore_photo_v4.1-lora

This LoRA model is designed for photo restoration and colorization tasks, fine-tuned on the FLUX.1-dev base model. It excels at enhancing old or damaged photos, making them photorealistic while preserving original details.

Trained using the AI Toolkit by Ostris, which supports efficient fine-tuning of diffusion models on consumer-grade hardware.

Recommended Prompts

For optimal results, use these prompts into your generation workflow:

• Primary prompt: restore and colorize this photo, photorealistic
• Alternatives:
  • restore this photo
  • restore and colorize this photo

Recommended Flux Guidance

For best result with the lora use a FluxGuidance of 4.

Demo Video

Watch a full-length demonstration of the LoRA in action:

YouTube Demo

Download and Usage

Download the model weights in Safetensors format from the Files & versions tab. It's compatible with popular interfaces like ComfyUI, AUTOMATIC1111, SD.Next, and InvokeAI.

Integration with Diffusers Library

You can easily load and use this LoRA with the Hugging Face 🧨 diffusers library. Here's an example script:

from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-Kontext-dev', torch_dtype=torch.bfloat16).to('cuda')
pipeline.load_lora_weights('cyburn/restore_photo_v4.1-lora', weight_name='restore_photo_v4.1_000002000.safetensors')

# Example: Restore and colorize a described photo
image = pipeline('restore and colorize this photo, photorealistic').images[^0]
image.save("restored_photo.png")

For advanced options like weighting, merging, or fusing LoRAs, refer to the diffusers documentation on loading adapters. Ensure you have access to the base model on Hugging Face and sufficient GPU resources (at least 24GB VRAM recommended for FLUX.1 models). ^1

⁂