Optimized Low-VRAM Workflow for Z-Image-Turbo (GGUF) with CacheDiT Acceleration Roughly based on WikkedAI’s WikkedZITv4 – refined and enhanced by Experimental_dAIverThis workflow delivers the full power of Z-Image-Turbo in GGUF format, specially optimized for GPUs with less than 8 GB VRAM, like my RTX 4050 with only 6 GB. The integrated CacheDiT_Model_Optimizer and SageAttention2 (both optional!) provides a noticeable 1.4–1.6× speed boost with almost no quality loss. Two intelligent upscaling stages, automatic trigger-word integration via the Super LoRA Loader, and an extended save node complete this elegant setup.Version 1.5 brings significant improvements in speed, usability, and upscaling quality — while remaining extremely VRAM-efficient (tested on RTX 4050 with only 6 GB).What’s new in v1.5:<ul><li>PatchSageAttentionKJ integration for automatic Sage Attention optimization and faster sampling on supported hardware</li><li>selectLatentSizePlus — intuitive aspect-ratio and resolution selector with beautiful presets (including 7:12 Tall Vista and other golden-ratio-friendly options) plus easy orientation swap</li><li>Full SEEDVR2 Video Upscaler Subgraph — powerful DiT-based (.safetensors or GGUF) high-end upscaler that delivers stunning 4K+ results with intelligent resolution handling, Lab color correction, and temporal settings. Works exceptionally well on still images too, producing superior detail and coherence</li><li>Main model updated to the higher-quality z-image-turbo-Q8_0.gguf</li><li>CLIP switched to the abliterated Qwen3-4B-Instruct-2507.Q5_K_S.gguf (lumina2 type)</li><li>Improved workflow organization, expanded notes, and more robust saving options</li></ul><hr /><h3 id="required-custom-nodes-(updated-for-v1.5):">Required Custom Nodes (updated for v1.5):</h3><ul><li>ComfyUI-GGUF - <a target="_blank" rel="ugc" href="https://github.com/city96/ComfyUI-GGUF">https://github.com/city96/ComfyUI-GGUF</a> - UnetLoaderGGUF + CLIPLoaderGGUF</li><li>ComfyUI-CacheDiT - <a target="_blank" rel="ugc" href="https://github.com/Jasonzzt/ComfyUI-CacheDiT">https://github.com/Jasonzzt/ComfyUI-CacheDiT</a> - CacheDiT_Model_Optimizer – the turbo boost for DiT models</li><li>nd-super-nodes - <a target="_blank" rel="ugc" href="https://github.com/HenkDz/nd-super-nodes">https://github.com/HenkDz/nd-super-nodes</a> - NdSuperLoraLoader with tags, trigger words &amp; beautiful UI</li><li>save-image-extended-comfyui - <a target="_blank" rel="ugc" href="https://github.com/thedyze/save-image-extended-comfyui">https://github.com/thedyze/save-image-extended-comfyui</a> - Advanced saving with metadata &amp; dynamic filenames</li><li>ComfyUi-MzMaXaM - <a target="_blank" rel="ugc" href="https://github.com/MzMaXaM/ComfyUi-MzMaXaM">https://github.com/MzMaXaM/ComfyUi-MzMaXaM</a> - selectLatentSizePlus</li><li>ComfyUI-SeedVR2_VideoUpscaler - <a target="_blank" rel="ugc" href="https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler">https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler</a> - SEEDVR2 Video Upscaler Subgraph</li></ul><hr /><h3 id="models-and-downloads-(exact-paths)">Models &amp; Downloads (exact paths)</h3>The following list explains the base models I am most frequently using with this workflow. The list as well explains where to put each file after you downloaded it.1. Main Model (Diffusion Model)<ul><li>File: z_image_turbo-Q8_0.gguf (higher quality Q8_0)</li><li>Download: <a target="_blank" rel="ugc" href="https://huggingface.co/jayn7/Z-Image-Turbo-GGUF/resolve/main/z_image_turbo-Q8_0.gguf">https://huggingface.co/jayn7/Z-Image-Turbo-GGUF/resolve/main/z_image_turbo-Q8_0.gguf</a></li><li>Target folder: ComfyUI/models/diffusion_models/</li></ul>2. Text Encoder (CLIP)<ul><li>File: Qwen3-4B-Instruct-2507-abliterated.Q5_K_S.gguf</li><li>Download: <a target="_blank" rel="ugc" href="https://huggingface.co/mradermacher/Huihui-Qwen3-4B-Instruct-2507-abliterated-GGUF/resolve/main/Huihui-Qwen3-4B-Instruct-2507-abliterated.Q5_K_S.gguf">https://huggingface.co/mradermacher/Huihui-Qwen3-4B-Instruct-2507-abliterated-GGUF/resolve/main/Huihui-Qwen3-4B-Instruct-2507-abliterated.Q5_K_S.gguf</a></li><li>Target folder: ComfyUI/models/text_encoders/ (or clip/)</li></ul>3. VAE<ul><li>File: ae.safetensors (~335 MB)</li><li>Download: Usually included with Z-Image-Turbo setups or available here: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensors?download=true">https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensors</a></li><li>Target folder: ComfyUI/models/vae/</li></ul>4. Upscalers<ul><li>4× Upscaler: 4xLSDIRplusN.pth (variant of 4x-UltraSharp) → <a target="_blank" rel="ugc" href="https://civarchive.com/models/116225/4x-ultrasharp">https://civarchive.com/models/116225/4x-ultrasharp</a></li><li>1× Skin-Contrast Upscaler: 1xSkinContrast-High-SuperUltraCompact.pth Download: <a target="_blank" rel="ugc" href="https://huggingface.co/notkenski/upscalers/blob/main/1xSkinContrast-High-SuperUltraCompact.pth">https://huggingface.co/notkenski/upscalers/blob/main/1xSkinContrast-High-SuperUltraCompact.pth</a></li><li>Target folder: ComfyUI/models/upscale_models/</li></ul>5. SEEDVR2 Models (for the new high-end upscaler – optional but recommended):<ul><li>DiT Model: seedvr2_ema_3b-Q8_0.gguf</li><li>VAE: ema_vae_fp16.safetensors</li><li>Download from the official ComfyUI-SeedVR2_VideoUpscaler repository or Hugging Face and place in the folders required by the custom node.</li></ul><hr /><h3 id="key-nodes-and-their-functions">Key Nodes &amp; Their Functions</h3><ul><li>CacheDiT_Model_Optimizer + PathchSageAttentionKJ → next-generation turbo boost</li><li>NdSuperLoraLoader with automatic trigger-word detection and clean tag interface</li><li>selectLatentSizePlus → effortless aspect-ratio and resolution control</li><li>KSamplerAdvanced with proven settings</li><li>Two-stage classic upscaler (4× LSDIR + 1× Skin-Contrast) in its own subgraph, or</li><li>New SEEDVR2 Video Upscaler Subgraph → for ultimate quality (optional, easily bypassed)</li><li>SaveImageExtended with full metadata and dynamic filenames</li></ul><hr /><h3 id="recommended-settings-(already-set-in-the-workflow)">Recommended Settings (already set in the workflow)</h3><ul><li>Sampler: res_multistep or dpmpp_sde</li><li>Scheduler: beta (or ddim_uniform)</li><li>Steps: 8–11</li><li>CFG Scale: 1.0–1.5</li><li>Shift: 4–7</li><li>Resolution: 864×1280 (portrait) – perfectly balanced for the golden ratio and typical ZIT outputs - the upscaler will automatically upscale by a factor of 4</li></ul><hr /><h3 id="how-to-use-the-workflow">How to Use the Workflow</h3><ol><li>Install all required custom nodes</li><li>Load the workflow</li><li>Enter your positive prompt (NdSuperLoraLoader automatically adds trigger words)</li><li>Adjust the negative prompt</li><li>Choose your desired aspect ratio and resolution in the Size Selector</li><li>Keep CacheDiT and Sage Attention enabled → Generate</li><li>Optionally run the SEEDVR2 upscaler for breathtaking 4K+ results</li><li>Done — images saved with complete metadata</li></ol>Special thanks to @LumaRift who provided the SeedVR2 subworkflow and some good advise on optimizing my setup.

Files with SHA256: d7f0e13c62fe479601ef952f00342c0d22ef562414a21aec1d7812a120c1a36b