Krea2Turbo_Easy_I2I_Auto-scale / Fix-scale
Special thanks to iamddtla for the krea2turbo checkpoint.
Image editing workflow based on Krea2 Turbo + Identity Edit LoRA.
Designed to modify existing images of fictional characters (anime, manga, illustrations, etc.) while preserving identity, pose, and framing.
Possible uses: clothing changes, style transfer, manga colorization, targeted prompt-based edits, NSFW on fictional characters only.
Important: Intended for synthetic/fictional characters only. Not designed for real people.
Two versions available:
Auto-scale: Automatically adapts the output size to the input image. Convenient for quick tests.
Fix-scale: Manual fixed resolution (recommended 832x1280 x64 multiplier). Better control and usually higher quality.
Note: If the result quality is not good enough with the auto-scale version (blur, artifacts, loss of detail…), switch to the Fix-scale version.
Required Models & LoRAs
Checkpoint: Krea2 Turbo NSFW Pornmaster
Text Encoder: Qwen3-VL-4B FP8 (qwen3vl_4b_fp8_scaled.safetensors)
VAE: Included with Krea2
LoRA: Krea2 Identity Edit v1.2 (required)
Required Custom Nodes
Approximate Generation Time (832×1280, 8–10 steps + SageAttention)
RTX 5090: ~15–25 s
RTX 4090: ~25–40 s
RTX 3090: ~45–70 s
RTX 3060: ~80–120 s
Recommended Settings
Sampler: euler / simple
CFG: 1.0
Denoise: 1
ref_boost: 4–8
Steps: 4–10
Krea2_Easy_I2I_Auto-scale / Fix-scale
基于 Krea2 Turbo + Identity Edit LoRA 的图像编辑工作流。
专为修改现有虚构角色图片(动漫、漫画、插画等)设计,同时尽量保持角色身份、姿势和构图不变。
适用场景:更换服装、风格转换、漫画上色、通过提示词进行定向编辑、仅限虚构角色的 NSFW 内容。
重要说明:仅适用于合成/虚构角色。不适用于真实人物。
提供两个版本:
Auto-scale:自动根据输入图像调整输出尺寸,方便快速测试。
Fix-scale:手动固定分辨率(推荐 1024×1024)。控制更精确,通常画质更好。
注意:如果使用 Auto-scale 版本时画质不佳(模糊、伪影、细节丢失等),请改用 Fix-scale 版本。
所需模型与 LoRA
Checkpoint:Krea2 Turbo NSFW Pornmaster
Text Encoder:Qwen3-VL-4B FP8(文件:qwen3vl_4b_fp8_scaled.safetensors)
VAE:随 Krea2 提供
LoRA:Krea2 Identity Edit v1.2(必需)
所需自定义节点
大致生成时间(832×1280,8–10 steps + SageAttention)
RTX 5090:约 15–25 秒
RTX 4090:约 25–40 秒
RTX 3090:约 45–70 秒
RTX 3060:约 80–120 秒
推荐参数
Sampler:euler / simple
CFG:1.0
Denoise:1
ref_boost:4–8
Steps:4–10








