<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 built for creators who need to refine many images with Krea2 without opening the graph for every single file. It is the batch-oriented version of the Krea2 image refinement setup: upload a ZIP package, let the workflow extract the image range, run the same restoration chain in a loop, and collect the outputs through RunningHub. It is useful for image cleanup, style-safe retouching, thumbnail batches, character card updates, and dataset preview passes where you want consistent processing rather than one-off manual tuning. 
 
The main model path uses `krea2_turbo_fp8.safetensors` through a UNET loader, paired with `qwen3vl_4b_fp8_scaled.safetensors` as the Krea2 CLIP/text encoder and `qwen_image_HDR_vae_fp32_comfy.safetensors` for decoding. The graph also includes a Qwen3VL processing route and an RH LLM API node, which helps generate or organize prompt guidance around the uploaded image content. The active structure-control side uses `depth-control-lora.safetensors` at strength 1.0, DA3 depth inference, Krea2ControlImageEncode, and Krea2ControlApply nodes. That means the workflow is not just doing blind upscaling: it tries to preserve the source image structure while giving Krea2 room to rebuild detail. 
 
The batch logic is visible in the graph through RHUploadZip, RHBatchImages, RH_ExtractImage, `easy forLoopStart`, `easy forLoopEnd`, and a batch-any output collection path. The image chain uses three KSampler stages with 8 steps, CFG 1, euler/simple sampling, and denoise values around 0.95, 0.35, and 0.25. The first stage gives the model enough freedom to rebuild the image; the later stages compress the structure back and polish the final output. The workflow also includes latent resizing toward 1024 x 1024, a `4x-UltraSharp.pth` image upscale model, and a final 2048 x 2048 image scale path. I describe it as a high-resolution refinement workflow rather than a guaranteed 4K/6K pipeline, because the confirmed graph ends around a 2048-square output path. 
 
Main features: 
 
- Krea2 Turbo FP8 image refinement chain 
- Online ZIP upload entry for batch image processing 
- RHBatchImages and loop-based image extraction 
- Qwen3VL-assisted prompt or image understanding route 
- Qwen Image HDR VAE decode path 
- Depth-control LoRA enabled at strength 1.0 
- DA3 depth inference for structure guidance 
- Three Krea2Control image encode/apply stages 
- Three 8-step euler/simple sampler passes 
- Denoise sequence around 0.95, 0.35, and 0.25 
- 1024 latent refinement path 
- 4x-UltraSharp upscaling stage 
- Final 2048-square image scale path 
- Multiple preview outputs for checking batch results 
- Best suited for consistent batch cleanup and retouching 
 
Suggested workflow: 
 
Prepare a ZIP that only contains supported image files such as JPG, JPEG, PNG, or WEBP. Keep the first batch small, then check whether the denoise balance preserves the source subject and composition. If the output becomes too different from the original, reduce the strength of the early rebuild stage or make the prompt more conservative. If the image keeps the structure but lacks detail, focus on the later refinement and upscale stages instead of forcing more variation at the beginning. 
 
For large batches, use the same prompt logic across the whole set and inspect several outputs before running everything. This workflow is strongest when the input images share a similar cleanup target, such as softer anime illustrations, character portraits, or concept art that needs sharper detail and more coherent texture. 
 
⚙️ RunningHub Workflow 
 
Try the workflow online right now — no installation required. 
👉 Workflow: https://www.runninghub.ai/post/2079074599203254274?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/2079074599203254274?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 Online Batch Refinement

<div data-youtube-video="https://youtu.be/j6EZ2I07fYs"></div> 
 
This workflow is built for creators who need to refine many images with Krea2 without opening the graph for every single file. It is the batch-oriented version of the Krea2 image refinement setup: upload a ZIP package, let the workflow extract the image range, run the same restoration chain in a loop, and collect the outputs through RunningHub. It is useful for image cleanup, style-safe retouching, thumbnail batches, character card updates, and dataset preview passes where you want consistent processing rather than one-off manual tuning. 
 
The main model path uses `krea2_turbo_fp8.safetensors` through a UNET loader, paired with `qwen3vl_4b_fp8_scaled.safetensors` as the Krea2 CLIP/text encoder and `qwen_image_HDR_vae_fp32_comfy.safetensors` for decoding. The graph also includes a Qwen3VL processing route and an RH LLM API node, which helps generate or organize prompt guidance around the uploaded image content. The active structure-control side uses `depth-control-lora.safetensors` at strength 1.0, DA3 depth inference, Krea2ControlImageEncode, and Krea2ControlApply nodes. That means the workflow is not just doing blind upscaling: it tries to preserve the source image structure while giving Krea2 room to rebuild detail. 
 
The batch logic is visible in the graph through RHUploadZip, RHBatchImages, RH_ExtractImage, `easy forLoopStart`, `easy forLoopEnd`, and a batch-any output collection path. The image chain uses three KSampler stages with 8 steps, CFG 1, euler/simple sampling, and denoise values around 0.95, 0.35, and 0.25. The first stage gives the model enough freedom to rebuild the image; the later stages compress the structure back and polish the final output. The workflow also includes latent resizing toward 1024 x 1024, a `4x-UltraSharp.pth` image upscale model, and a final 2048 x 2048 image scale path. I describe it as a high-resolution refinement workflow rather than a guaranteed 4K/6K pipeline, because the confirmed graph ends around a 2048-square output path. 
 
Main features: 
 
- Krea2 Turbo FP8 image refinement chain 
- Online ZIP upload entry for batch image processing 
- RHBatchImages and loop-based image extraction 
- Qwen3VL-assisted prompt or image understanding route 
- Qwen Image HDR VAE decode path 
- Depth-control LoRA enabled at strength 1.0 
- DA3 depth inference for structure guidance 
- Three Krea2Control image encode/apply stages 
- Three 8-step euler/simple sampler passes 
- Denoise sequence around 0.95, 0.35, and 0.25 
- 1024 latent refinement path 
- 4x-UltraSharp upscaling stage 
- Final 2048-square image scale path 
- Multiple preview outputs for checking batch results 
- Best suited for consistent batch cleanup and retouching 
 
Suggested workflow: 
 
Prepare a ZIP that only contains supported image files such as JPG, JPEG, PNG, or WEBP. Keep the first batch small, then check whether the denoise balance preserves the source subject and composition. If the output becomes too different from the original, reduce the strength of the early rebuild stage or make the prompt more conservative. If the image keeps the structure but lacks detail, focus on the later refinement and upscale stages instead of forcing more variation at the beginning. 
 
For large batches, use the same prompt logic across the whole set and inspect several outputs before running everything. This workflow is strongest when the input images share a similar cleanup target, such as softer anime illustrations, character portraits, or concept art that needs sharper detail and more coherent texture. 
 
⚙️ RunningHub Workflow 
 
Try the workflow online right now — no installation required. 
👉 Workflow: https://www.runninghub.ai/post/2079074599203254274?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/2079074599203254274?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: 7d9681db9a60cbbb83fdacf962289c9cce3d2369bda1dd71c5fb46c5592d1dc2