02. Image Editing (Image Editing)

What is Image Editing?

If image generation is the task of “creating something from nothing,” then Image Editing is the task of “modifying an existing image in a desired direction.”

AI-based image editing can automatically perform complex modifications using only text prompts or masks.

The main editing types supported in ComfyUI are as follows:

Editing Type	Representative Use Cases
Inpainting	Removing unwanted objects, face correction, outfit replacement
Outpainting	Background expansion
Relighting	Day-to-night conversion, changing lighting direction
Context-aware Editing	Character style transfer, product mockups
Instruction-based Editing	Simple object replacement and modification

---

Detailed Explanation by Editing Type

2.1 Inpainting

Inpainting is a technique where you designate a specific area of an image with a Mask, and only that area is filled with new content. A mask is like an instruction saying “please redraw this part.”

How It Works

Original image + Mask (area to edit) + Text
-> Only the masked area is regenerated
-> The rest of the image is preserved

Representative Workflow: Flux.1 Fill Inpainting

This is an inpainting workflow using the Flux.1 Fill Dev model.

Key Components:

Node	Role
DualCLIPLoader	Loads two text encoders, clip_l and t5xxl, with the “flux” type
Load Diffusion Model	Loads the flux1-fill-dev.safetensors model
FluxGuidance	Sets prompt adherence strength (default value 30)
InpaintModelConditioning	Passes the original image and mask to the model

Tip

Setting the mask area slightly wider than the part you want to edit will produce more natural results.

---

2.2 Outpainting

Outpainting is a technique that extends the canvas beyond the image boundaries to generate new content. It naturally fills surrounding areas while maintaining the style and content of the original image.

How It Works

Original image
-> ImagePadForOutpaint (how much to extend in each direction)
-> Generate a mask for the extended area
-> Generate the extended area based on text

Representative Workflow: Flux.1 Outpainting

This uses the same flux1-fill-dev.safetensors model as inpainting, but with additional padding and blending nodes.

Key Nodes:

Node	Role
ImagePadForOutpaint	Sets padding size and feathering for each direction. Extension pixels can be specified individually per direction
DifferentialDiffusion	Smoothly blends the boundary between the original and extended areas

Tip

The feathering value (default 24) controls the smoothness of the boundary between the original and extended areas. Higher values produce more natural boundaries, but may slightly alter the edges of the original image.

Tip

Outpainting reuses the inpainting model. It works by treating the area to be extended as an “empty mask region” and inpainting it.

---

2.3 Relighting

Relighting is a technique for changing the lighting of an image. You can control the lighting direction and intensity via text, such as changing from daytime to nighttime, or switching from side lighting to backlighting.

Representative Workflow: Qwen Image Relight

This workflow applies a Relight LoRA to the Qwen Edit 2509 model.

Key Components:

Item	Details
Model	qwen_image_edit_2509_fp8_e4m3fn.safetensors
LoRA	Lightning 4-step LoRA + Relight LoRA
Trigger Word	Chinese trigger word “重新照明,” (meaning “relight”)
Prompt Format	”重新照明,” + desired lighting description (in English)

Caution

In the relighting workflow, you must prepend the Chinese trigger word “重新照明,” to your prompt. This trigger word tells the model that the task is relighting.

---

2.4 Text-based Editing

Instruction-based Editing is a method of modifying images using only text, without masks. It uses intuitive commands such as “remove the text.”

Representative Workflow: Qwen Image Edit

This is the most versatile instruction-based editing workflow.

• Example prompt: “Remove all UI text elements”

Tip

The ImageScaleToTotalPixels node automatically adjusts the input image to 1.5 megapixels. This ensures the optimal resolution at which the model operates efficiently.

---

2.5 Context-aware Editing

Context-aware Editing is a technique for changing the style, background, attributes, and more of a character or object while preserving its appearance. Unlike regular editing, it “remembers” the key features of the input image to ensure consistency.

Representative Workflow: Flux Kontext Dev

The Flux.1 Kontext Dev model is an editing model specialized in character consistency.

Key Features:

• Can accept 1-2 reference images as input
• Character consistency: Transform the same character into different poses/backgrounds
• Style Transfer: Change the artistic style of an image
• Text editing: Modify text within an image
• Object modification: Change attributes of specific objects

Tip

When you input 2 images into Flux Kontext, cross-editing becomes possible — for example, placing the character from the first image into the style/background of the second image.

---

Workflow Comparison Table

Workflow	Editing Type	Model	Input	Features
Flux.1 Fill Inpaint	Inpainting	Flux.1 Fill Dev	Image+Mask+Text	Text-based fill of masked area
Flux.1 Outpaint	Outpainting	Flux.1 Fill Dev	Image+Text	Canvas extension, padding/feathering settings
Qwen Relight	Relighting	Qwen Edit 2509	Image+Text	Lighting change, trigger word required
Qwen Image Edit	General Edit	Qwen Image Edit	Image+Text	General-purpose editing, turbo mode
Flux Kontext	Context Edit	Flux.1 Kontext	Image (1-2)+Text	Character consistency, style transfer

---

General Tips for Editing Tasks

Tip

If you are not satisfied with the editing result, try changing the seed value to explore different results. Even with the same text, results can vary significantly depending on the seed.

Tip

The more specific your text, the better the results. “Change the blue shirt to a red shirt” produces more accurate results than just “change it.”

Tip

For complex edits, break them into steps rather than doing everything at once. For example, applying background changes first and then lighting adjustments sequentially will produce more natural results.

Tip

If the resolution of your editing result is insufficient, you can connect the 04. Upscale/Enhancement chapter’s workflow after editing to increase the resolution.

---