# ✏️ Sketch Enhancer## 📥 Download Workflow&gt; ✅ No custom nodes required!## 👋 IntroHi! Here I'm sharing a workflow I created while experimenting with a certain LoRA for Flux Klein 2.With this workflow, you can turn any (as far as I've tested) anime / manga / cartoon low-quality image into an enhanced, colored, detailed pencil version. 🎨Here's what you can expect it to do:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/8cacb149-c74a-4736-a6f5-693a0f175306/original=true/8cacb149-c74a-4736-a6f5-693a0f175306.jpeg" /> You can even colorize black-and-white manga pages / panels! 📖 <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/05f8efbc-3fd3-4feb-9b1d-2b43b7d15fdf/original=true/05f8efbc-3fd3-4feb-9b1d-2b43b7d15fdf.jpeg" /> &gt; ⚠️ Note: If the original aesthetic, facial, or body features are altered too much, you may need to add specific instructions to your prompt, or add certain LoRAs for specific image features (e.g. NSFW).---## ⚙️ Setup### 📦 Required Files&gt; 💾 Download links are included in the workflow file as notes.| File | Purpose ||---|---|| <code>RealESRGAN_x4plus_anime_6B.pth</code> | Upscale model || <code>Flux Klein 2 9B</code> | Diffusion model || <code>Hyperdetailed Colored Pencil</code> | LoRA || <code>Qwen 3 8B</code> | CLIP model || <code>Flux2</code> | VAE | ### 🖥️ ComfyUI VersionTested on ComfyUI v0.27.0---## 🧩 Workflow Explanation1. Load image2. Upscale it 4x its original resolution3. Downscale it to 1 megapixel4. Get image size and create an empty latent from it5. Load model, CLIP, VAE, and LoRA6. Write prompt with LoRA trigger words: <code>ArsMJStyle</code> OR <code>Colored pencil hyperdetailed realism</code> OR both7. Set up sampler:   - Sampler: [fill in]   - Scheduler: [fill in]   - Steps: <code>4</code> (range: 4–16)   - CFG: <code>1.0</code> (range: 0.8–2.0)   - Denoise: <code>1.0</code> (range: 0.85–1.0)8. Decode and preview the image---## 🏁 Final WordAs mentioned above, this workflow requires no custom nodes — only native ComfyUI nodes. 🙌My original, messy, non-post-friendly workflow uses a ton of custom nodes and subgraphs to make researching value/LoRA combinations more practical. Feel free to combine different LoRAs and sampler values to your heart's content — you can get even better results than my "default" settings shown here! 🔥### ⏱️ PerformanceWith the default step count of <code>4</code>, the workflow takes ~32–33 seconds on average.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a1548667-1da5-4830-8ce9-ff3883a6bba6/original=true/a1548667-1da5-4830-8ce9-ff3883a6bba6.jpeg" />- 🕒 <code>8</code> steps → ~66 seconds- 🕒 <code>16</code> steps → ~127 secondsMy hardware:- 🧠 Intel Core i7-8700K- 💾 32 GB DDR4 RAM- 🎮 NVIDIA RTX 3060 12GB VRAM---💬 Feedback, questions, and results are always welcome — drop a comment below!

Files with SHA256: 08d53590f3b9e97232ade10270da77f31a3bdffe14f06e1aec635351d5a11418