1. Workflow Overview

Purpose:
This workflow (named FLUX丨Kontext) enables AI-powered image editing via natural language commands (e.g., "Replace the goose head with a snake head"), combining inpainting, semantic understanding, and multilingual support.

Core Models:

• FluxKontextMaxImageNode: Edits images based on text prompts (object/background replacement).

• Stable Diffusion (implicit): Powers underlying image generation.

• LibLibTranslate: Translates non-English prompts (e.g., Chinese→English).

2. Critical Nodes

Model Dependencies:

• Download .safetensors checkpoint (e.g., SD 1.5/XL) to models/checkpoints.

3. Workflow Structur

graph LR
  A[Input Image] --> B[Resolution Adjust]
  B --> C[Flux Image Edit]
  D[Text Prompt] --> C
  C --> E[Output Image]

Group Breakdown:

1. Input Group:

   • LoadImage: Loads source image (e.g., 1152x1536 PNG).

   • Get_提示词: Accepts text commands (e.g., "Change background to flower field").

2. Process Group:

   • ImageScaleByAspectRatio: Scales image to 1024x1024 (maintains aspect ratio).

   • FluxKontextMaxImageNode: Executes edits (uses prompt + seed).

3. Output Group:

   • PreviewImage: Displays final result.

4. Inputs & Outputs

Input Parameters:

• Image: PNG/JPG, recommended resolution ≥1024x1024.

• Text Prompt: Supports multilingual input (auto-translated).

• Seed: Fixed (e.g., 944134171880156) or randomized.

Output:

• Edited image (retains original dimensions/metadata).

5. Important Notes

⚠️ Hardware:

• Requires NVIDIA GPU (≥8GB VRAM). RTX 3060+ recommended.

• High-resolution images may cause OOM; pre-scale inputs.

⚠️ Installation:

• FluxKontextMaxImageNode must be manually installed from GitHub to custom_nodes.

• LibLibTranslate may require API key (some versions are paywalled).

⚠️ Troubleshooting:

• If output is unchanged, verify prompt translation to English.

• Disable mask input in ImageScaleByAspectRatio if unused.