Functional LoRAs Expand Local Krea 2 Editing

A small wave of community LoRA releases has expanded what Krea 2 can do in local inference, especially for practical image-editing workflows. Instead of adding new styles, these components target harder problems such as preserving a person’s identity during edits and extending an image beyond its original frame, which makes them more notable for production-oriented use.

Key details

The newly shared tools fall into three main categories. First, @Fishmongr posted Krea 2’s **Identity Edit v1.2 LoRA**, saying it can effectively turn **Krea 2 Turbo** into an image-editing model while keeping a subject’s identity consistent; the examples focus on editing clothing, pose, and related composition elements. Second, @Upbeat_Birthday_6123 open-sourced a **ReID Reference** LoRA for **local Krea 2 inference**. According to that post, users can provide a single identity reference image and then use prompts to freely change clothing, pose, composition, and background; the setup combines **Qwen3-VL** image conditioning. Third, the same release also included a **positional outpainting** LoRA aimed at image extension.

Workflow feedback

For the outpainting workflow, @linoy_tsaban highlighted a new Krea 2 outpaint LoRA as small in size, fast, and high in quality, and pointed to a demo. A repost from @multimodalart also amplified community reactions that Krea 2’s outpainting results look exceptionally strong when paired with this custom LoRA. Across the posts, the common theme is that these are lightweight, task-specific add-ons intended to make Krea 2 more usable for real editing work rather than cosmetic style changes.

2026-07-17 ~ 2026-07-18 · 6 related posts

Full story(3 episodes)→