<div data-youtube-video="https://youtu.be/j6EZ2I07fYs"><iframe src="https://www.youtube.com/embed/j6EZ2I07fYs" width="853" height="480" allowfullscreen></iframe></div><br /><br />
<p>This workflow is local-only: it runs in your own ComfyUI installation and has no RunningHub workflow page in the source material. Files are in the Hugging Face / ModelScope package for episode 280.</p>
<h2>Krea 2 Advanced Cleanup - Local Folder Batch (Structure LoRA + Reverse Prompt)</h2>
<p>Purpose: the local folder-loop variant of the Krea 2 cleanup set. Point it at a folder of images; each image is reverse-prompted and rebuilt independently through depth-structure conditioning and the 4x upscale, then saved one by one.</p>
<h2>Graph (as saved)</h2>
<p>57 nodes / 98 links / 10 groups.</p>
<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.</li>
<li>Three depth stages: three <code>Krea2ControlImageEncode</code> → <code>Krea2ControlApply</code> pairs driven by DA3 depth maps.</li>
<li>Batch loop: <code>easy loadImagesForLoop</code> + <code>easy forLoopEnd</code> + <code>easy batchAnything</code>; each image is processed independently (the saved note is explicit that results are not reused across a batch).</li>
<li>Sampling: three KSamplers at 8 steps / CFG 1 / euler / simple with denoise 0.95, 0.35 and 0.25; two <code>LatentUpscale</code> nodes (bislerp, 1024) saved disabled; final 4x-UltraSharp upscale.</li>
<li>Reverse prompt: Qwen3VLProcessor (open route) and RH_LLMAPI_Pro_Node (closed route, GLM-4.6V label); <code>ConditioningKrea2Rebalance</code> is active in this variant.</li>
<li>Output: SaveImage prefix <code>Krea2_本地批量洗图/Krea2_batch</code>.</li>
<li>Saved-state inconsistency: the in-graph Note text still says "线上批量版" (online batch), but the actual nodes are a local folder loop and the output path is the local-batch path. The node graph is what runs; the note text is stale.</li>
</ul>
<h2>How to use</h2>
<p>Point the loop loader at your folder, set the render stages / final long edge / reverse-prompt source in the console, and queue. Because this workflow is local-only, expect local VRAM and time costs — the online variant exists as a separate graph.</p>
<h2>Limits and not-verified items</h2>
<p>No inference was run for this listing: no speed, VRAM, batch throughput or quality claim is made. The saved notes are the author's configuration, not measured results. Weights, LoRA, depth model, upscaler and VLM routes are not bundled; check their licenses separately. The cover image is a promotional still, not a frame produced by this graph.</p>
<p>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.</p>

Krea 2 Advanced Cleanup - Local Folder Batch (Structure LoRA + Reverse Prompt)

<div data-youtube-video="https://youtu.be/j6EZ2I07fYs"></div><br /><br />
<p>This workflow is local-only: it runs in your own ComfyUI installation and has no RunningHub workflow page in the source material. Files are in the Hugging Face / ModelScope package for episode 280.</p>
Krea 2 Advanced Cleanup - Local Folder Batch (Structure LoRA + Reverse Prompt)
<p>Purpose: the local folder-loop variant of the Krea 2 cleanup set. Point it at a folder of images; each image is reverse-prompted and rebuilt independently through depth-structure conditioning and the 4x upscale, then saved one by one.</p>
Graph (as saved)
<p>57 nodes / 98 links / 10 groups.</p>
<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.</li>
<li>Three depth stages: three <code>Krea2ControlImageEncode</code> → <code>Krea2ControlApply</code> pairs driven by DA3 depth maps.</li>
<li>Batch loop: <code>easy loadImagesForLoop</code> + <code>easy forLoopEnd</code> + <code>easy batchAnything</code>; each image is processed independently (the saved note is explicit that results are not reused across a batch).</li>
<li>Sampling: three KSamplers at 8 steps / CFG 1 / euler / simple with denoise 0.95, 0.35 and 0.25; two <code>LatentUpscale</code> nodes (bislerp, 1024) saved disabled; final 4x-UltraSharp upscale.</li>
<li>Reverse prompt: Qwen3VLProcessor (open route) and RH_LLMAPI_Pro_Node (closed route, GLM-4.6V label); <code>ConditioningKrea2Rebalance</code> is active in this variant.</li>
<li>Output: SaveImage prefix <code>Krea2_本地批量洗图/Krea2_batch</code>.</li>
<li>Saved-state inconsistency: the in-graph Note text still says "线上批量版" (online batch), but the actual nodes are a local folder loop and the output path is the local-batch path. The node graph is what runs; the note text is stale.</li>
</ul>
How to use
<p>Point the loop loader at your folder, set the render stages / final long edge / reverse-prompt source in the console, and queue. Because this workflow is local-only, expect local VRAM and time costs — the online variant exists as a separate graph.</p>
Limits and not-verified items
<p>No inference was run for this listing: no speed, VRAM, batch throughput or quality claim is made. The saved notes are the author's configuration, not measured results. Weights, LoRA, depth model, upscaler and VLM routes are not bundled; check their licenses separately. The cover image is a promotional still, not a frame produced by this graph.</p>
<p>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.</p>

Files with SHA256: dd9eea1b5d1da03560535acf8a1683eb9d0acde4c29cbac6daa627c410dd8c31