Edit Image

Apply AI-powered image operations including upscaling, background removal, and context-aware editing to an existing image.

Note: This page documents the edit-image API endpoint / legacy node. In the editor, this functionality has been split into dedicated nodes: Upscale Image (see upscale-image.md), Remove Background (see remove-background.md), and Modify Image (see modify-image.md).

Overview

Edit Image takes an existing image as input and applies a transformation operation. It supports AI upscaling (Recraft, Topaz, and Grok), background removal (Recraft), context-aware prompt-based editing (Nano Banana Edit), and the Grok Imagine 2 task-chained operations — a free segment map and region-targeted prompt edits of a prior Grok 2 generation (see Grok Imagine 2 Task-Chained Editing). Unlike Generate Image or Image to Image, this node focuses on non-destructive enhancement and utility operations rather than creative generation. The default operation is Recraft Upscale.

Configuration

Field Type Default Description
Provider / Operation select recraft-upscale The edit operation to apply (5 options, see providers table)
Prompt text "" Edit instructions (only used by nano-banana-edit)
Style select "" Style preset or custom text (only used by nano-banana-edit)
Negative Prompt text "" Elements to exclude (only used by nano-banana-edit)
Upscale Factor select 2 Topaz only: 1x, 2x, 4x
Aspect Ratio select – Available for nano-banana-edit
Seed number – Reproducibility seed (nano-banana-edit)
Connected Media Order list – Order of connected input media assets

Inputs & Outputs

Inputs:

Outputs:

Inpainting Mask

Edit Image’s backend forwards an attached mask to the provider as the mask_url parameter for workflow use – wire the mask output of a Generate Mask or hand-painted Paint Mask node into this node’s mask input, and the mask will be passed through at execution time.

The interactive Mask Painter UI (Paint Mask / Edit Mask buttons, brush/eraser/lasso tools, overlay/mask/source view toggle) currently lives in the Modify Image node when the Ideogram Edit provider is selected. White areas of the mask are re-rendered by the model; black areas are preserved untouched from the source image. Once a mask is saved on a Modify Image node, re-opening the painter pre-seeds the canvas with the existing mask so further edits build on prior work rather than starting blank.

Supported Providers

Provider Label Description
recraft-upscale Recraft Upscale AI-powered upscaling and enhancement. Simple one-click operation with no additional configuration.
topaz-image-upscale Topaz Upscale Advanced upscaling with a configurable factor (1x/2x/4x).
recraft-remove-bg Recraft Remove BG Removes the background and outputs a transparent PNG. No additional configuration needed.
nano-banana-edit Nano Banana Edit Context-aware image editing using a text prompt. Supports style presets, negative prompts, aspect ratio, seed, and character/asset references.
grok-upscale Grok Upscale AI upscaling via Grok. Takes a prior Grok generation’s taskId instead of an image URL.
grok-2-edit Grok Imagine 2 Edit Prompt-based edit of a prior Grok 2 generation (taskId + prompt required). Optional maskIndexes restrict the edit to named regions from a segment map. Priced the same as a Grok 2 generation.
grok-2-segment Grok Imagine 2 Segment Map Free. Named segment-mask map of a prior Grok 2 generation (taskId required, no prompt).

Grok Imagine 2 Task-Chained Editing

The three grok-* operations don’t take an image URL. They reference a prior Grok generation on the provider side via taskId — the task id is returned in the source generation job’s output as kieTaskId (fetch the job via the API/SDK/MCP get_job and read output_data.kieTaskId). Only Grok generations can be chained; images from other providers or uploads cannot.

The full region-editing flow with grok-2:

  1. Generate an image with the grok-2 provider (Generate Image node or POST /v1/generate-image). Note the completed job’s kieTaskId.
  2. Segment (optional, free) — POST /v1/edit-image with { "provider": "grok-2-segment", "taskId": "<kieTaskId>" }. output_data.segments lists { index, name } pairs (e.g. 0 = sky, 1 = person), order-aligned with the job’s output images — which are ~128×128 alpha-masked cutouts of each region (a bounding-box crop of the region’s own pixels), not full-frame binary masks. Additionally pass imageUrl (the source generation’s image) and each segment also gains a normalized bbox: { x, y, w, h } — its recovered position in the source image (Grok returns no geometry; the backend recovers it by template-matching the cutout against the source; absent when matching wasn’t confident).
  3. Edit — POST /v1/edit-image with { "provider": "grok-2-edit", "taskId": "<kieTaskId>", "prompt": "make the sky stormy", "maskIndexes": [0] }. Omit maskIndexes to let the prompt apply to the whole image.

maskIndexes entries are the segment map’s index values passed through verbatim (0-based in practice, whatever the segment map returned). A grok-2-edit run costs the same as a grok-2 generation; the segment map costs nothing. Note the edit is region-driven rather than pixel-locked: Grok may adjust global lighting to keep the scene consistent (e.g. a night-sky edit also relights the water).

In the editor, this flow is built into the Generate Image node: with the grok-2 provider selected, its config panel shows a Refine Regions section (detect regions → tick named chips → prompt edit) — see Generate Image → Refine regions. The API/MCP flow above is the same machinery.

Best Practices

Common Use Cases

Tips