Summary
This in-depth tutorial from the Veteran AI channel tests Krea Two Identity Edit, a ComfyUI LoRA that adds character-consistent image editing capabilities to the Krea 2 model. Over 48 tests conducted in ComfyUI — with an accessible mirror workflow built on RunningHub — the creator systematically evaluates what the model can and cannot do, presenting both successful results and clear failure cases rather than cherry-picking outputs.
Two core workflows are covered: Single Reference editing (modifying one image while preserving identity) and Two Reference editing (placing a character from one image into a scene from another). Key parameters are explained throughout — LoRA strength at 1.0, Turbo mode using 8-step sampling with CFG 1.0 for speed, and a switch to Raw mode with 20+ steps and CFG 3.0 specifically for object deletion tasks. The comfyui-krea2edit extension is required and installs without additional Python dependencies. A practical prompt formula is introduced: editing action + specific object description + target result + explicit preservation instructions, which the creator demonstrates across clothing changes, accessory additions, expression tweaks, and multi-character position swaps.
Test categories are divided into local attribute edits (stable results), object addition/deletion/replacement (reliable with mode adjustments), and large-scale scene or pose changes (notably unstable). Multi-character editing, including swapping two characters’ positions while maintaining correct prop-binding, is shown to work well with position-plus-clothing prompt identifiers. The tutorial is practical for AI artists and ComfyUI users evaluating identity-consistent editing tools for character design and creative production workflows.
📺 Source: Veteran AI · Published July 13, 2026
🏷️ Format: Tutorial Demo
