Transform Your Photos into Breathtaking Oil Paintings with FLUX Model System

CN
ComfyUI.org
2025-04-29 08:23:57

1. Workflow Overview

ma28pqx7pfkgc0bvd2a4f0ded05c17b1c5d70457994e009ace9b22e8be628a4c6c2e9a7af4f3c79a435.png

This is a Rembrandt-style portrait repainting workflow based on FLUX model system, featuring:

  • Transforms photos into hyper-realistic oil paintings

  • Facial detail preservation (PULID) + pose control (Openpose ControlNet)

  • Auto 4x upscaling (Anime6B model) and resizing

2. Core Models

Model

Function

Source

Base Algorithm_F.1

Core UNet model

Built-in

FLUX-ControlNet-Union-Pro

Pose control

Manual download required

PULID_Flux_v0.9.1

Facial identity preservation

Separate download

Oil Painting LoRA

Rembrandt style transfer

Built-in weights

3. Key Nodes

Node

Purpose

Installation

ApplyPulidFlux

Facial feature fusion

Requires ComfyUI-PULID

ControlNetApplyAdvanced

Pose guidance

Needs Advanced-ControlNet

ImageScaleByAspectRatio

Smart resizing

Built-in

R-ESRGAN_4x

Super-resolution

Via ComfyUI-Manager

4. Workflow Stages

Phase 1: Preprocessing

  • Input: Portrait photo (via LoadImage)

  • Processing:

    • Standardize size (1280xauto)

    • Openpose extraction

    • PULID facial encoding

Phase 2: Painting Generation

  • Key Parameters:

    • Sampler: Euler

    • Steps: 25

    • Prompt: "Masterpiece, oil painting, Rembrandt lighting"

    • ControlNet strength: 0.75

Phase 3: Post-Processing

  • 4x upscale → 50% resize (quality preservation) → Final output

5. Inputs & Outputs

Input Parameters:

  • Image: Portrait ratio recommended (e.g. 968x1280)

  • Prompt: Edit art style in CR Text node

  • Seed: Default 702652529958285

Output:

  • PNG with metadata

6. Notes

  1. VRAM: 16GB+ GPU recommended (for 1280px processing)

  2. Critical Adjustments:

    • Style intensity: Modify 油画,著作_001 LoRA weight (default 1.2)

    • Facial fidelity: Adjust identity_strength in ApplyPulidFlux (default 0.85)

  3. Troubleshooting:

    • Face distortion: Reduce ControlNet strength or increase PULID weight

    • Weak style: Boost LoRA weight to 1.5