<div data-youtube-video="https://youtu.be/CU1UKxjWBRo"><iframe src="https://www.youtube.com/embed/CU1UKxjWBRo" width="853" height="480" allowfullscreen></iframe></div> 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2079889925343891458?inviteCode=rh-v1111
<h2>Krea 2 Identity Edit v1.2 - Single-Image Editor (ref_boost 4)</h2>
Purpose: a single-image Krea 2 editing graph built around the Identity Edit v1.2 LoRA. You load one image, write a plain-English instruction, and the graph edits that image while trying to preserve facial identity and the rest of the frame. This is the plain single-image mode; the optional second-reference group ships bypassed.
<h2>Graph (as saved)</h2>
24 nodes / 26 links / 6 groups: MODELS, 1 IMAGE TO EDIT, 2 SECOND REFERENCE (optional, bypassed), 3 INSTRUCTION, 4 FIDELITY, 5 GENERATE.
<ul>
<li>UNETLoader: <code>krea2_turbo_fp8_scaled.safetensors</code>.</li>
<li>LoRA: <code>krea2_identity_edit_v1_2_fp16.safetensors</code> at strength 1.</li>
<li>CLIPLoader: <code>qwen3vl_4b_fp8_scaled.safetensors</code>, type <code>krea2</code>. VAELoader: <code>qwen_image_HDR_vae_fp32_comfy.safetensors</code>.</li>
<li>KSampler: 10 steps, CFG 1, <code>euler</code>, <code>simple</code>, denoise 1 (turbo-matched; the workflow's own note says krea2_raw instead of turbo is 40 steps / CFG 3-4).</li>
<li>Krea2EditModelPatch: ref_boost 4, single-image mode, fit - the Fidelity dial ships at the recommended 4.0.</li>
<li>Krea2EditGroundedEncode: positive instruction (the saved example asks to redesign part of a product into a clean-beauty tube while preserving the model's face, smile, pose and lighting); negative left empty.</li>
<li>SaveImage prefix <code>krea2_identity_edit</code>.</li>
</ul>
<h2>How to use</h2>
Load your image in group 1, write the edit as a direct instruction in group 3, queue. Keep the Fidelity dial at 4 for likeness-critical edits; the graph's note documents 1.0 as the older v1.1 behaviour and warns that values above ~10 start over-copying and breaking removals. The second-reference group is disabled in this export: enable it deliberately if you want scene + person insertion, and put the scene in slot 1 and the person in slot 2.
<h2>Limits and not-verified items</h2>
No inference was run for this listing: no speed, VRAM, quality or likeness claim is made. The saved prompts and reference images are the author's demo content, not evidence of an output. The graph's notes are author notes, not measured results. Weights, LoRA and input media are not bundled; resolve the exact filenames in your own installation and check their licenses separately. The cover image is a promotional still shipped with the source video, not a frame produced by this graph.
Tutorial: <a href="https://www.youtube.com/watch?v=CU1UKxjWBRo" rel="ugc">YouTube</a> · <a href="https://www.bilibili.com/video/BV1bJgy6eEei" rel="ugc">Bilibili</a>. Workflows are supplied through AI-KSK; verify workflow, node, model, LoRA, source-image and cover rights separately. No blanket commercial or redistribution license is asserted.

Krea 2 Identity Edit v1.2 - Single-Image Editor (ref_boost 4)

<div data-youtube-video="https://youtu.be/CU1UKxjWBRo"></div> 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2079889925343891458?inviteCode=rh-v1111
Krea 2 Identity Edit v1.2 - Single-Image Editor (ref_boost 4)
Purpose: a single-image Krea 2 editing graph built around the Identity Edit v1.2 LoRA. You load one image, write a plain-English instruction, and the graph edits that image while trying to preserve facial identity and the rest of the frame. This is the plain single-image mode; the optional second-reference group ships bypassed.
Graph (as saved)
24 nodes / 26 links / 6 groups: MODELS, 1 IMAGE TO EDIT, 2 SECOND REFERENCE (optional, bypassed), 3 INSTRUCTION, 4 FIDELITY, 5 GENERATE.
<ul>
<li>UNETLoader: <code>krea2_turbo_fp8_scaled.safetensors</code>.</li>
<li>LoRA: <code>krea2_identity_edit_v1_2_fp16.safetensors</code> at strength 1.</li>
<li>CLIPLoader: <code>qwen3vl_4b_fp8_scaled.safetensors</code>, type <code>krea2</code>. VAELoader: <code>qwen_image_HDR_vae_fp32_comfy.safetensors</code>.</li>
<li>KSampler: 10 steps, CFG 1, <code>euler</code>, <code>simple</code>, denoise 1 (turbo-matched; the workflow's own note says krea2_raw instead of turbo is 40 steps / CFG 3-4).</li>
<li>Krea2EditModelPatch: ref_boost 4, single-image mode, fit - the Fidelity dial ships at the recommended 4.0.</li>
<li>Krea2EditGroundedEncode: positive instruction (the saved example asks to redesign part of a product into a clean-beauty tube while preserving the model's face, smile, pose and lighting); negative left empty.</li>
<li>SaveImage prefix <code>krea2_identity_edit</code>.</li>
</ul>
How to use
Load your image in group 1, write the edit as a direct instruction in group 3, queue. Keep the Fidelity dial at 4 for likeness-critical edits; the graph's note documents 1.0 as the older v1.1 behaviour and warns that values above ~10 start over-copying and breaking removals. The second-reference group is disabled in this export: enable it deliberately if you want scene + person insertion, and put the scene in slot 1 and the person in slot 2.
Limits and not-verified items
No inference was run for this listing: no speed, VRAM, quality or likeness claim is made. The saved prompts and reference images are the author's demo content, not evidence of an output. The graph's notes are author notes, not measured results. Weights, LoRA and input media are not bundled; resolve the exact filenames in your own installation and check their licenses separately. The cover image is a promotional still shipped with the source video, not a frame produced by this graph.
Tutorial: <a href="https://www.youtube.com/watch?v=CU1UKxjWBRo" rel="ugc">YouTube</a> · <a href="https://www.bilibili.com/video/BV1bJgy6eEei" rel="ugc">Bilibili</a>. Workflows are supplied through AI-KSK; verify workflow, node, model, LoRA, source-image and cover rights separately. No blanket commercial or redistribution license is asserted.

krea2IdentityEditV12Single_v10.json

v1.0

https://youtu.be/CU1UKxjWBRo 
 
This workflow is the standard single-image edit setup for Krea 2 Identity Edit v1.2. It is meant for cases where you want to load one image, describe a new visual direction, and let Krea 2 rebuild the picture while preserving the core subject and layout. Compared with a strict masked workflow, this version gives the model more room to reinterpret the whole frame. It is suitable for changing clothing, restaging a character, improving the mood, testing a new style, or turning an existing image into a more polished concept without starting from a blank prompt. 
 
The active model chain uses `krea2_turbo_fp8_scaled.safetensors` as the UNET, `qwen3vl_4b_fp8_scaled.safetensors` as the Krea2 CLIP/text encoder, and `qwen_image_HDR_vae_fp32_comfy.safetensors` as the VAE. The workflow loads `krea2_identity_edit_v1_2_fp16.safetensors` at strength 1.0, which is the key edit LoRA in this setup. The inspected output path includes a 1024 x 1024 EmptySD3LatentImage route plus a ResolutionSelector set to a square profile. The active KSampler is a 10-step euler/simple pass with CFG 1 and denoise 1, so this version behaves more like a direct image editing/rebuild graph than a tiny local retouch. 
 
Use this workflow when the edit can affect the whole image but the original subject should still be recognizable. It is especially useful for character portraits, fashion swaps, cinematic restaging, thumbnail variations, and quick visual direction tests. The bypass warning in the graph indicates that some reachable nodes are disabled, so the description only focuses on the active output chain and does not claim extra enhancement branches. 
 
Main features: 
 
- Single-image Krea 2 editing workflow 
- Krea 2 Turbo FP8 active UNET path 
- Krea 2 Identity Edit v1.2 LoRA enabled at strength 1.0 
- Qwen3VL 4B FP8 Krea2 text encoder route 
- Qwen Image HDR VAE decode path 
- Square 1024 x 1024 latent setup 
- ResolutionSelector square profile 
- 10-step euler/simple KSampler 
- CFG 1 and denoise 1 edit pass 
- Best for full-frame image edits and restaging 
 
Suggested workflow: 
 
Load a source image with a clear subject and write the edit as a complete instruction, such as changing the outfit, mood, scene, lighting, or pose direction. Start with a simple prompt and check whether the identity is preserved. If the result becomes too different, make the prompt more conservative and explicitly mention preserving the same person, face, pose, or composition. If the edit is too weak, add clearer target details instead of overloading the prompt. 
 
RunningHub Workflow 
 
Try the workflow online right now - no installation required. 
Workflow: https://www.runninghub.ai/post/2079889925343891458?inviteCode=rh-v1111 
 
If the results meet your expectations, you can later deploy it locally for customization. 
 
Fan Benefits: Register to get 1000 points + daily login 100 points - enjoy 4090 performance and 48 GB super power! 
 
Bilibili Updates (Mainland China &amp; Asia-Pacific) 
 
If you are in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown. 
Bilibili Video: https://www.bilibili.com/video/BV1bJgy6eEei/ 
 
Support Me on Ko-fi 
 
If you find my content helpful and want to support future creations, you can buy me a coffee. 
Every bit of support helps me keep creating. 
Ko-fi: https://ko-fi.com/aiksk 
 
Business Contact 
 
For collaboration or inquiries, please contact aiksk95 on WeChat. 
 
打开下方链接即可在线体验，无需安装。 
工作流：https://www.runninghub.ai/post/2079889925343891458?inviteCode=rh-v1111 
 
如果觉得效果理想，也可以在本地进行自定义部署。 
 
粉丝福利：注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48G 超级性能。 
 
Bilibili 更新（中国大陆及亚太地区） 
 
如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。 
B站视频：https://www.bilibili.com/video/BV1bJgy6eEei/ 
 
本期 Krea 2 相关模型资源： 
https://pan.quark.cn/s/07bdc81784ce?pwd=FpDf#/list/share 
 
这些资源主要面向本地用户，方便进行创作与学习。

Files with SHA256: 6740fe226220c962f12187744ee60b838fd5cd77d559c32d7f399d3d2a01352b