# Anima EyesHQ — Hi-Res Detailer V2I think I can finally call this workflow actually finished? maybe....A lot of testing went into improving the overall detail quality, especially faces and eyes, while keeping it usable on an RTX 4070 8GB.⏱️ Around 150–200s per generation on my laptopAt this point, I’m mostly updating it because I apparently don’t know how to stop tweaking things💀Quality first. Always !<h3 id="required-custom-nodes">Required Custom Nodes</h3><ul><li>Impact Pack — <a target="_blank" rel="ugc" href="https://github.com/comfyorg/comfyui-impact-pack">https://github.com/comfyorg/comfyui-impact-pack</a></li><li>Impact Subpack — necessary for <code>UltralyticsDetectorProvider</code></li><li>rgthree-comfy — <a target="_blank" rel="ugc" href="https://github.com/rgthree/rgthree-comfy">https://github.com/rgthree/rgthree-comfy</a></li><li>JPS Nodes — <a target="_blank" rel="ugc" href="https://github.com/JPS-GER/ComfyUI_JPS-Nodes">https://github.com/JPS-GER/ComfyUI_JPS-Nodes</a></li><li>UltimateSDUpscale — <a target="_blank" rel="ugc" href="https://github.com/ssitu/ComfyUI_UltimateSDUpscale">https://github.com/ssitu/ComfyUI_UltimateSDUpscale</a></li><li>NVIDIA RTX Nodes — <a target="_blank" rel="ugc" href="https://github.com/Comfy-Org/Nvidia_RTX_Nodes_ComfyUI">https://github.com/Comfy-Org/Nvidia_RTX_Nodes_ComfyUI</a></li></ul><h3 id="required-models">Required Models</h3><ul><li>Anima Aesthetic 1.0 — <code>anima_aestheticV10.safetensors</code></li><li>Qwen 3 0.6B Base — <code>qwen_3_06b_base.safetensors</code></li><li>Qwen Image VAE — <code>qwen_image_vae.safetensors</code></li><li>4x-UltraSharp — <code>4x-UltraSharp.pth</code></li><li>SAM ViT-B — <code>sam_vit_b_01ec64.pth</code></li><li>Anzhc Face v4 — <code>Anzhc_Face_v4_y11n.pt</code></li><li>Eyeful v2 Paired — <code>Eyeful_v2-Paired.pt</code></li><li>NTD11 Anime NSFW Segmentation — <code>ntd11_anime_nsfw_segm_v5-variant1.pt</code></li></ul><pre><code>UltralyticsDetectorProvider is no longer included directly in Impact Pack, you need to get Impact Subpack separately for recent versions</code></pre>⚠️ RTX Nodes are optional if you don't have a compatible NVIDIA RTX GPU. Simply bypass the RTX upscale group :)

Anima Aesthetic Hi-Res Detailer V2 — EyesHQ + detailer 🌙

The workflow is finally finished.After many tests, several suspiciously smooth images, and a HiRes pass that briefly forgot how details work, this should be the final version.This workflow uses Anima with a clean anime style, Grain Guard refinement, two HiRes passes, and a dedicated eye detailer to keep the eyes sharp instead of letting them turn into decorative soup.What it does<ul><li>Generates anime-style images with Anima.</li><li>Supports multiple resolutions.</li><li>Uses <code>er_sde</code> with the <code>simple</code> scheduler.</li><li>Includes Grain Guard to reduce unwanted grain.</li><li>Applies a pre-detailer HiRes pass.</li><li>Detects and refines the eyes with YOLO, SAM, and Impact Detailer.</li><li>Applies a final HiRes pass.</li><li>Supports optional LoRAs.</li><li>Keeps the workflow organized so you can hopefully understand it without opening twelve browser tabs.</li></ul>Recommended settings<ul><li>Steps: 35</li><li>CFG: 4</li><li>Sampler: <code>er_sde</code></li><li>Scheduler: <code>simple</code></li><li>Grain Guard: 10 steps, <code>dpmpp_2m</code>, <code>karras</code>, denoise 0.15</li><li>Eye Detailer guide size: 768</li><li>Eye Detailer max size: 1024</li><li>Eye Detailer denoise: 0.32</li><li>HiRes: 50% of a 4x upscale, effectively 2x</li></ul>Required models<ul><li>Anima model ( aesthetic 1.0 the best )</li><li>Qwen text encoder</li><li>Qwen image VAE</li><li><code>Eyeful_v2-Paired.pt</code></li><li><code>sam_vit_b_01ec64.pth</code></li><li><code>4x-UltraSharpV2.pth</code></li></ul>Required custom nodes<ul><li>ComfyUI-Easy-Use</li><li>rgthree-comfy</li><li>ComfyUI-Impact-Pack</li><li>ComfyUI-Impact-Subpack</li><li>JPS Nodes</li></ul>If something does not work, check the model paths first. If everything is installed correctly and it still breaks, blame ComfyUI!!

Files with SHA256: 5d139ed35fe90ba6ff07b01a60ab46bd7b506280803a409d6c070e4481fb01dd