<h1 id="krea2turbo_easy_i2i_auto-scale-fix-scale">Krea2Turbo_Easy_I2I_Auto-scale / Fix-scale</h1>Special thanks to <a target="_blank" rel="ugc" href="https://civarchive.com/user/iamddtla">iamddtla</a> 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:<ul><li>Auto-scale: Automatically adapts the output size to the input image. Convenient for quick tests.</li><li>Fix-scale: Manual fixed resolution (recommended 832x1280 x64 multiplier). Better control and usually higher quality.</li></ul>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.<h2 id="required-models-and-loras">Required Models &amp; LoRAs</h2><ul><li>Checkpoint: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2735032/pornmaster-krea2?modelVersionId=3373630">Krea2 Turbo NSFW Pornmaster</a></li><li>Text Encoder: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen3-VL">Qwen3-VL-4B FP8</a> (qwen3vl_4b_fp8_scaled.safetensors)</li><li>VAE: Included with Krea2</li><li>LoRA: Krea2 Identity Edit v1.2 (required)</li></ul><h2 id="required-custom-nodes">Required Custom Nodes</h2><ul><li><a target="_blank" rel="ugc" href="https://github.com/Krea-AI/Krea2Edit">Krea2Edit</a></li><li><a target="_blank" rel="ugc" href="https://github.com/willmiao/ComfyUI-Lora-Manager">ComfyUI-LoraManager</a></li></ul><h2 id="approximate-generation-time-(8321280-8-10-steps-+-sageattention)">Approximate Generation Time (832×1280, 8–10 steps + SageAttention)</h2><ul><li>RTX 5090: ~15–25 s</li><li>RTX 4090: ~25–40 s</li><li>RTX 3090: ~45–70 s</li><li>RTX 3060: ~80–120 s</li></ul><h2 id="recommended-settings">Recommended Settings</h2><ul><li>Sampler: euler / simple</li><li>CFG: 1.0</li><li>Denoise: 1</li><li>ref_boost: 4–8</li><li>Steps: 4–10</li></ul><hr /><h1 id="krea2_easy_i2i_auto-scale-fix-scale">Krea2_Easy_I2I_Auto-scale / Fix-scale</h1>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：<ul><li>Auto-scale：自动根据输入图像调整输出尺寸，方便快速测试。</li><li>Fix-scale：手动固定分辨率（推荐 1024×1024）。控制更精确，通常画质更好。</li></ul>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。<h2 id="lora">所需模型与 LoRA</h2><ul><li>Checkpoint：<a target="_blank" rel="ugc" href="https://civarchive.com/models/2735032/pornmaster-krea2?modelVersionId=3373630">Krea2 Turbo NSFW Pornmaster</a></li><li>Text Encoder：<a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen3-VL">Qwen3-VL-4B FP8</a>（文件：qwen3vl_4b_fp8_scaled.safetensors）</li><li>VAE：随 Krea2 提供</li><li>LoRA：Krea2 Identity Edit v1.2（必需）</li></ul><h2>所需自定义节点</h2><ul><li><a target="_blank" rel="ugc" href="https://github.com/Krea-AI/Krea2Edit">Krea2Edit</a></li><li><a target="_blank" rel="ugc" href="https://github.com/willmiao/ComfyUI-Lora-Manager">ComfyUI-LoraManager</a></li></ul><h2 id="83212808-10-steps-+-sageattention">大致生成时间（832×1280，8–10 steps + SageAttention）</h2><ul><li>RTX 5090：约 15–25 秒</li><li>RTX 4090：约 25–40 秒</li><li>RTX 3090：约 45–70 秒</li><li>RTX 3060：约 80–120 秒</li></ul><h2>推荐参数</h2><ul><li>Sampler：euler / simple</li><li>CFG：1.0</li><li>Denoise：1</li><li>ref_boost：4–8</li><li>Steps：4–10</li></ul>

Files with SHA256: 78b016aa97677552f53c5a2781b93e1a3425e9a4c33fb405958a0e01e9ff18df