Practical field guide

Krea2 Identity Edit: The Complete Guide

Krea2 Identity Edit is a community LoRA workflow for changing an image while anchoring the referenced person's likeness. It accepts a source image plus a natural-language edit instruction. Use it for restaging, local replacement, compositing, inpainting, and virtual try-on without rebuilding the subject from scratch.

What you need before starting

For the online workflow, you only need a modern browser and a source image you have permission to edit. Prepare the following before entering the shared GPU queue.

01

A clear source

Use a sharp, well-lit image with one primary subject and a face large enough to inspect.
02

One intended change

Decide what must change and what must remain untouched before you write the prompt.
03

A comparison habit

Keep the source visible so you can check eyes, face shape, skin details, clothing marks, and unintended changes.

Workflow

Step-by-step Krea2 Identity Edit tutorial

Follow this six-step loop for a reliable first result. Change one variable at a time so you know what improved or damaged likeness.

  1. 01

    Open the online workspace

    Go to the identity editor and wait for the embedded Space to finish loading. Open it full-screen if the page is narrow.
  2. 02

    Upload the source image

    Choose a clear portrait or character image. Avoid heavy compression, motion blur, and multiple similar faces.
  3. 03

    Write a concrete prompt

    Use a trained action such as replace, recolor, restage, add, or remove. Mention only what you want changed.
  4. 04

    Choose a practical starting recipe

    For common restages and replacements, start with a short Turbo-style run. Use the exposed defaults before tuning advanced controls.
  5. 05

    Generate and audit identity

    Compare the face, hairline, skin marks, silhouette, and any detail you expected to remain fixed.
  6. 06

    Refine one variable

    Shorten the prompt if the edit spreads too far. Increase reference strength carefully if likeness drifts, then generate again.

Prompt examples that work

These templates are deliberately concrete. Replace the bracketed detail, keep one main verb, and leave preserved details out of the instruction.

Use casePrompt template
BackgroundCreate a photo of this person at a night market with warm lantern light.
Studio restagePlace this person in a minimal photo studio with a soft gray backdrop.
Outfit colorRecolor the jacket to rust orange while keeping its material and shape.
Virtual try-onReplace the shirt with a tailored navy blazer over a white T-shirt.
Object replacementReplace the coffee cup with a small ceramic bowl.
Object additionAdd a black leather shoulder bag to this person.
Season changeRestage this person on a snowy street in soft morning light.
Cinematic lightRelight the scene with a cool window key light and warm practical lights.
Pose and framingCreate a waist-up editorial portrait of this person looking over their shoulder.
Character compositePlace the person from image 2 naturally into the street scene from image 1.
Hair editChange the hairstyle to a short textured bob while keeping the same face.
Clean backgroundReplace the background with a clean off-white wall and soft floor shadow.

FAQ

Common problems and fixes

Most failed edits come from an overloaded prompt, a weak source, or a setting outside the model's comfortable range.

The face looks like a relative, not the same person.

Use a sharper source, shorten the prompt, and increase reference fidelity gradually. Extreme changes in angle can still alter facial geometry.

The image turns into a split composition.

Reference grounding may be too strong or the input order may be wrong. Return to defaults and verify scene-first, person-second ordering for two-reference composites.

The edit changes too much.

Ask for one local change with a direct replacement or recolor verb. Remove style language that affects the entire frame.

Removing an object keeps failing.

Removal is harder than adding or replacing. Try the Raw-style recipe when available, simplify the surrounding scene, and expect a few reruns.

Try it online without ComfyUI

Use the browser editor first to validate your source image and prompt. You can move to a local ComfyUI workflow later if you need pinned weights, batch output, or stacked LoRAs.

Krea2 vs Krea AI

The names are easy to confuse. Our upcoming comparison separates the open Krea 2 model stack, the community Identity Edit workflow, and Krea's official web product.

Preview the Krea2 vs Krea AI comparison