Advanced reference

Krea2 Edit Technical Guide

This page collects the model, tuning, and local-workflow details behind the browser editor. It is written for users who want pinned weights, repeatable settings, lower-VRAM variants, or a local ComfyUI graph.

Model weight and version guidance

Prefer v1.2 when it is available. It is the recommended build on the model card and adds reference sheets, swaps, inpainting, outpainting, virtual try-on, and FIT geometry. Confirm the version pinned by the hosted Space before comparing results across sessions.

Recommended file

krea2_identity_edit_v1_2.safetensors

更改 Krea2 Edit 结果的设置

Treat these values as starting points from the public model card and node changelog, not universal guarantees. Change one control at a time and compare against the source image.

职位类型检查站步骤CFG
添加/重新着色/重新设计/重新舞台Krea 2 Turbo8–121.0
删除/大量删除Krea 2 Raw~203.0
  • ref_boost ≈ 4 — v1.2 节点上的强烈相似起点;高于 ~10 开始破坏移除;低于 1 会削弱创造力。
  • grounding_px — 较低的版本更一致地遵循编辑;更高的保护身份。重复的“分割图片”帧通常意味着该值远远高于训练范围(v1.1 文档引用 384-768 作为舒适带)。
  • LoRA strength 1.0 — 已发布默认值。当 CFG 高于 1 时,将负片与空提示加上相同的图像接地。
  • v1.2 的低 VRAM SVD 削减( r128 ~ 0.91 GB, r64 ~ 0.46 GB) 每张卡保留 >99% 的重量能量 — 当您将浏览器 Krea2 Edit 留给笔记本电脑 GPU 时非常有用。

浏览器 Krea2 Edit 与本地 ComfyUI

The hosted editor is best for quick prompt tests. A local graph is better when you need exact versions, repeatable seeds, batches, private infrastructure, or additional LoRAs.

托管 Space(本页)

  • 无检查点下载;在几分钟内尝试 Krea2 Edit
  • 适合 Krea2 Edit 上的快速研发和客户端演示
  • 取决于 Space 正常运行时间和共享 GPU 队列
  • 全屏打开: hf.space

本地 ComfyUI-Krea2Edit

  • 引脚 v1.2 权重和 FIT / ref_boost 节点版本
  • 在顶部堆叠字符、正文或样式 LoRA
  • 需要原生 Krea 2 + Qwen3-VL 编码器支持
  • ComfyUI-Krea2Edit on GitHub

Krea2 Edit 的限制、许可和负责任的使用

Krea2 Edit 上的相似性在上游被描述为纹理忠实且比例保守:皮肤和灯光通常转移得很好,而非常不寻常的面部几何形状可能看起来像近亲。服装交换是偶然的;删除需要 Raw 配方,但有时仍然会失败。极端的头发或身体类型可能需要在顶部放置主题 LoRA。

权重是下式下的导数模型 Krea 2 Community License。在公布的收入门槛下允许商业使用(模型卡摘要:低于 100 万美元/年);除此之外,请联系 Krea 了解企业条款。部署应适度内容; AI 披露可能适用。 LoRA 经过 SFW 训练 - 不会创建未经同意或有害的真人肖像。阅读我们的 隐私政策 服务条款.

Technical sources

Check the upstream model card and repositories for current files, workflows, changelogs, licenses, and compatibility notes.

FAQ

Technical FAQ

Version and local-workflow answers for advanced users.

Which model weight should I use?

Prefer v1.2 (krea2_identity_edit_v1_2.safetensors) when available, and confirm the exact version pinned by the hosted Space before comparing sessions.

What do I need for a local workflow?

Use ComfyUI with native Krea 2 and Qwen3-VL support, install the ComfyUI-Krea2Edit node pack, load the LoRA, and start from a workflow shipped by the project.

What settings should I start with?

For common additions, recolors, restyles, and restages, start with the Turbo recipe at 8–12 steps and CFG 1.0. For large removals, try the Raw recipe near 20 steps and CFG 3.0.