<div data-youtube-video="https://youtu.be/j6EZ2I07fYs"><iframe src="https://www.youtube.com/embed/j6EZ2I07fYs" width="853" height="480" allowfullscreen></iframe></div> 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2078868402176872450?inviteCode=rh-v1111
<h2>Krea 2 Advanced Cleanup - Single Image (Structure LoRA + Reverse Prompt)</h2>
Purpose: a single-image Krea 2 "cleanup" graph. It reverse-prompts the input image into a Krea 2 prompt, then rebuilds the image through depth-structure conditioning with a selectable number of render stages and a final 4x upscale. This is the single-image variant of the set.
<h2>Graph (as saved)</h2>
51 nodes / 89 links / 9 groups.
<ul>
<li>Model: UNETLoader <code>krea2_turbo_fp8.safetensors</code>; CLIPLoader <code>qwen3vl_4b_fp8_scaled.safetensors</code> (type <code>krea2</code>); VAELoader <code>qwen_image_HDR_vae_fp32_comfy.safetensors</code>.</li>
<li>Structure LoRA: one <code>Krea2ControlLoRALoader</code> on <code>depth-control-lora.safetensors</code> at strength 1.0 — a single shared structure LoRA, not a preset trio.</li>
<li>Three depth stages: three <code>Krea2ControlImageEncode</code> → <code>Krea2ControlApply</code> pairs (first / middle / final size), all <code>match_latent_size</code>, lanczos, center, grayscale, per-image min/max.</li>
<li>Sampling: three KSamplers at 8 steps / CFG 1 / <code>euler</code> / <code>simple</code> with denoise 0.55, 0.35 and 0.25; two <code>LatentUpscale</code> nodes (bislerp, 1024) saved disabled; final <code>UpscaleModelLoader</code> <code>4x-UltraSharp.pth</code> + <code>ImageUpscaleWithModel</code>.</li>
<li>Reverse prompt: <code>Qwen3VLProcessor</code> (open-source route) and <code>RH_LLMAPI_Pro_Node</code> (the closed route, model label GLM-4.6V) — the user console asks which source to use.</li>
<li>User console: render stages (two or three), final long edge in pixels (2048/3072/4096, aspect preserved and 64-aligned), reverse-prompt source. <code>ConditioningKrea2Rebalance</code> is saved bypassed here; the <code>ComfyUI-Krea2T-Enhancer</code> patch is saved enabled.</li>
<li>Output: SaveImage prefix <code>KSK_Krea2_ReverseDepthHires</code>.</li>
</ul>
<h2>How to use</h2>
Load the source image, pick two or three stages, set the final long edge, and choose the reverse-prompt route. Treat 2048 as the value-for-money point; the graph's own note says 3072/4096 increase VRAM and cost noticeably.
<h2>Limits and not-verified items</h2>
No inference was run for this listing: no speed, VRAM, fidelity or upscale-quality claim is made. The saved input image and notes are the author's demo content. Weights, LoRA, depth model, upscaler and the VLM routes are not bundled; check their licenses separately. The cover image is a promotional still, not a frame produced by this graph.
Tutorial: <a href="https://www.youtube.com/watch?v=j6EZ2I07fYs" rel="ugc">YouTube</a> · <a href="https://www.bilibili.com/video/BV1i7KU62Esg" rel="ugc">Bilibili</a>. Workflows supplied through AI-KSK; verify workflow, node, model, LoRA, depth model and cover rights separately. No blanket commercial or redistribution license is asserted.

Krea 2 Advanced Cleanup - Single Image (Structure LoRA + Reverse Prompt)

<div data-youtube-video="https://youtu.be/j6EZ2I07fYs"></div> 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2078868402176872450?inviteCode=rh-v1111
Krea 2 Advanced Cleanup - Single Image (Structure LoRA + Reverse Prompt)
Purpose: a single-image Krea 2 "cleanup" graph. It reverse-prompts the input image into a Krea 2 prompt, then rebuilds the image through depth-structure conditioning with a selectable number of render stages and a final 4x upscale. This is the single-image variant of the set.
Graph (as saved)
51 nodes / 89 links / 9 groups.
<ul>
<li>Model: UNETLoader <code>krea2_turbo_fp8.safetensors</code>; CLIPLoader <code>qwen3vl_4b_fp8_scaled.safetensors</code> (type <code>krea2</code>); VAELoader <code>qwen_image_HDR_vae_fp32_comfy.safetensors</code>.</li>
<li>Structure LoRA: one <code>Krea2ControlLoRALoader</code> on <code>depth-control-lora.safetensors</code> at strength 1.0 — a single shared structure LoRA, not a preset trio.</li>
<li>Three depth stages: three <code>Krea2ControlImageEncode</code> → <code>Krea2ControlApply</code> pairs (first / middle / final size), all <code>match_latent_size</code>, lanczos, center, grayscale, per-image min/max.</li>
<li>Sampling: three KSamplers at 8 steps / CFG 1 / <code>euler</code> / <code>simple</code> with denoise 0.55, 0.35 and 0.25; two <code>LatentUpscale</code> nodes (bislerp, 1024) saved disabled; final <code>UpscaleModelLoader</code> <code>4x-UltraSharp.pth</code> + <code>ImageUpscaleWithModel</code>.</li>
<li>Reverse prompt: <code>Qwen3VLProcessor</code> (open-source route) and <code>RH_LLMAPI_Pro_Node</code> (the closed route, model label GLM-4.6V) — the user console asks which source to use.</li>
<li>User console: render stages (two or three), final long edge in pixels (2048/3072/4096, aspect preserved and 64-aligned), reverse-prompt source. <code>ConditioningKrea2Rebalance</code> is saved bypassed here; the <code>ComfyUI-Krea2T-Enhancer</code> patch is saved enabled.</li>
<li>Output: SaveImage prefix <code>KSK_Krea2_ReverseDepthHires</code>.</li>
</ul>
How to use
Load the source image, pick two or three stages, set the final long edge, and choose the reverse-prompt route. Treat 2048 as the value-for-money point; the graph's own note says 3072/4096 increase VRAM and cost noticeably.
Limits and not-verified items
No inference was run for this listing: no speed, VRAM, fidelity or upscale-quality claim is made. The saved input image and notes are the author's demo content. Weights, LoRA, depth model, upscaler and the VLM routes are not bundled; check their licenses separately. The cover image is a promotional still, not a frame produced by this graph.
Tutorial: <a href="https://www.youtube.com/watch?v=j6EZ2I07fYs" rel="ugc">YouTube</a> · <a href="https://www.bilibili.com/video/BV1i7KU62Esg" rel="ugc">Bilibili</a>. Workflows supplied through AI-KSK; verify workflow, node, model, LoRA, depth model and cover rights separately. No blanket commercial or redistribution license is asserted.

krea2AdvancedCleanupSingle_v10.json

v1.0

<div data-youtube-video="https://youtu.be/j6EZ2I07fYs"><iframe src="https://www.youtube.com/embed/j6EZ2I07fYs" width="853" height="480" allowfullscreen></iframe></div> 
 
This workflow is the single-image version of the Krea2 refinement setup. It is designed for cases where you want to take one existing image, keep the original subject and composition recognizable, and let Krea2 rebuild the texture, lighting, detail, and final polish. It is a practical choice for creators who want a controlled redraw instead of a fully new generation: character illustrations, AI covers, anime portraits, product-style renders, and old outputs that need a cleaner second pass. 
 
The active model path uses `krea2_turbo_fp8.safetensors`, with `qwen3vl_4b_fp8_scaled.safetensors` loaded as the Krea2 CLIP route and `qwen_image_HDR_vae_fp32_comfy.safetensors` as the VAE. A standard LoadImage entry brings the source picture into the workflow, and VAEEncode sends it into the latent path. The graph also contains Qwen3VLProcessor and RH LLM API support, which can help with image understanding or prompt assistance depending on how you run the online version. The important structural control is the `depth-control-lora.safetensors` path at strength 1.0, combined with DA3 depth inference and multiple Krea2Control encode/apply stages. 
 
The sampling structure is a three-stage refinement chain. The inspected graph shows 8-step KSampler passes with CFG 1, euler/simple sampling, and denoise values around 0.55, 0.35, and 0.25. This is more conservative than a full redraw: the first pass is strong enough to clean and restyle the image, while the later passes focus on keeping the structure stable and recovering sharper detail. The workflow also includes latent resizing toward 1024 x 1024, `4x-UltraSharp.pth` for image upscaling, and a final 2048 x 2048 image scale path. Because the actual graph points to a 2048-square output route, this description avoids overclaiming a guaranteed 4K result. 
 
Main features: 
 
- Single-image Krea2 refinement workflow 
- LoadImage and VAEEncode input path 
- Krea2 Turbo FP8 UNET route 
- Qwen3VL Krea2 CLIP encoder 
- Qwen Image HDR VAE decode path 
- Optional Qwen3VL-assisted image understanding route 
- Depth-control LoRA enabled at strength 1.0 
- DA3 depth inference for structure preservation 
- Three Krea2Control encode/apply stages 
- Three 8-step euler/simple sampler passes 
- Conservative denoise sequence around 0.55, 0.35, and 0.25 
- 1024 latent refinement path 
- 4x-UltraSharp image upscale stage 
- Final 2048-square output scale path 
- Useful for controlled cleanup, redraw, and detail recovery 
 
Suggested workflow: 
 
Start with a single image that already has the composition you want to keep. Use a conservative prompt such as image restoration, preserved subject identity, clean texture, natural fine detail, coherent lighting, and realistic depth. If the output changes the face, pose, or main silhouette too much, lower the first-stage denoise or make the control guidance stricter. If the output feels too close to the original and lacks improvement, increase the first-stage freedom slightly before touching the later polish stages. 
 
Use the preview as a quality check before saving the final version. This workflow is especially useful when you are refining one important image rather than processing a whole folder, because you can tune the prompt and denoise balance more carefully for that single source. 
 
⚙️ RunningHub Workflow 
 
Try the workflow online right now — no installation required. 
👉 Workflow: https://www.runninghub.ai/post/2079074609300197378?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’re 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/BV1i7KU62Esg/ 
 
☕ 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 — just like a spark that can ignite a blazing flame. 
👉 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/2079074609300197378?inviteCode=rh-v1111 
如果觉得效果理想，你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 48 G 超级性能！ 
 
📺 Bilibili 更新（中国大陆及南亚太地区） 
 
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/BV1i7KU62Esg/ 
 
本期 Krea2 相关模型资源： 
👉 https://pan.quark.cn/s/07bdc81784ce 
这些资源主要面向本地用户，方便进行创作与学习。

Krea2 Single Image Refinement

<div data-youtube-video="https://youtu.be/j6EZ2I07fYs"></div> 
 
This workflow is the single-image version of the Krea2 refinement setup. It is designed for cases where you want to take one existing image, keep the original subject and composition recognizable, and let Krea2 rebuild the texture, lighting, detail, and final polish. It is a practical choice for creators who want a controlled redraw instead of a fully new generation: character illustrations, AI covers, anime portraits, product-style renders, and old outputs that need a cleaner second pass. 
 
The active model path uses `krea2_turbo_fp8.safetensors`, with `qwen3vl_4b_fp8_scaled.safetensors` loaded as the Krea2 CLIP route and `qwen_image_HDR_vae_fp32_comfy.safetensors` as the VAE. A standard LoadImage entry brings the source picture into the workflow, and VAEEncode sends it into the latent path. The graph also contains Qwen3VLProcessor and RH LLM API support, which can help with image understanding or prompt assistance depending on how you run the online version. The important structural control is the `depth-control-lora.safetensors` path at strength 1.0, combined with DA3 depth inference and multiple Krea2Control encode/apply stages. 
 
The sampling structure is a three-stage refinement chain. The inspected graph shows 8-step KSampler passes with CFG 1, euler/simple sampling, and denoise values around 0.55, 0.35, and 0.25. This is more conservative than a full redraw: the first pass is strong enough to clean and restyle the image, while the later passes focus on keeping the structure stable and recovering sharper detail. The workflow also includes latent resizing toward 1024 x 1024, `4x-UltraSharp.pth` for image upscaling, and a final 2048 x 2048 image scale path. Because the actual graph points to a 2048-square output route, this description avoids overclaiming a guaranteed 4K result. 
 
Main features: 
 
- Single-image Krea2 refinement workflow 
- LoadImage and VAEEncode input path 
- Krea2 Turbo FP8 UNET route 
- Qwen3VL Krea2 CLIP encoder 
- Qwen Image HDR VAE decode path 
- Optional Qwen3VL-assisted image understanding route 
- Depth-control LoRA enabled at strength 1.0 
- DA3 depth inference for structure preservation 
- Three Krea2Control encode/apply stages 
- Three 8-step euler/simple sampler passes 
- Conservative denoise sequence around 0.55, 0.35, and 0.25 
- 1024 latent refinement path 
- 4x-UltraSharp image upscale stage 
- Final 2048-square output scale path 
- Useful for controlled cleanup, redraw, and detail recovery 
 
Suggested workflow: 
 
Start with a single image that already has the composition you want to keep. Use a conservative prompt such as image restoration, preserved subject identity, clean texture, natural fine detail, coherent lighting, and realistic depth. If the output changes the face, pose, or main silhouette too much, lower the first-stage denoise or make the control guidance stricter. If the output feels too close to the original and lacks improvement, increase the first-stage freedom slightly before touching the later polish stages. 
 
Use the preview as a quality check before saving the final version. This workflow is especially useful when you are refining one important image rather than processing a whole folder, because you can tune the prompt and denoise balance more carefully for that single source. 
 
⚙️ RunningHub Workflow 
 
Try the workflow online right now — no installation required. 
👉 Workflow: https://www.runninghub.ai/post/2079074609300197378?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’re 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/BV1i7KU62Esg/ 
 
☕ 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 — just like a spark that can ignite a blazing flame. 
👉 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/2079074609300197378?inviteCode=rh-v1111 
如果觉得效果理想，你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 48 G 超级性能！ 
 
📺 Bilibili 更新（中国大陆及南亚太地区） 
 
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/BV1i7KU62Esg/ 
 
本期 Krea2 相关模型资源： 
👉 https://pan.quark.cn/s/07bdc81784ce 
这些资源主要面向本地用户，方便进行创作与学习。

Files with SHA256: e81cf4af534ef66eef96e04414eaec45478116a6fdfe0a11bb3c57afd2744c9d