<div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/K1AWKd1_ibY?rel=1" start="0"></iframe></div><h1 id="minimax-h3-ref2va-ultra-detail-workflow-(merged-model-and-dual-sampler)">MiniMax-H3 Ref2VA Ultra-Detail Workflow (Merged Model &amp; Dual-Sampler)</h1>This is currently what I consider the best workflow for MiniMax-H3 multi-reference video generation! Designed specifically to overcome high-motion artifacting while significantly boosting video clarity, this workflow features two key innovations:<ul><li>Dual-Stage Resampling via <code>ExtendIntermediateSigmas</code>: Fixes video artifacts and injects rich details under high-motion conditions by adding linear noise interpolation to the final 2 steps.</li><li>Custom Merged Diffusion Model: Powered by a hybrid checkpoint merging <code>fl2va</code> (for superior image quality) and <code>ref2va</code> (for temporal stability), giving you pristine visual fidelity while fully retaining reference video editing and audio capabilities.</li></ul><h3 id="try-it-online-(free-test-drive)">☁️ Try It Online (Free Test Drive)</h3>Test this workflow directly in your browser on RunningHub before downloading:👉 <a target="_blank" rel="ugc" href="https://www.runninghub.ai/workflow/2089055364653268993?inviteCode=rh-v1559">Run on RunningHub</a><h3 id="recommended-model-configuration">⚙️ Recommended Model Configuration</h3>1. Diffusion Model (Base)<ul><li>Model: <code>minimax_h3_ref2va_XUELUO_int8_convrot</code> (High-fidelity merged Ref model)</li><li>Download: <a target="_blank" rel="ugc" href="https://civarchive.com/">Civitai Model Page</a> (Currently uploading; link will be live ~1 hour post-release)</li><li>Download2:<a target="_blank" rel="ugc" href="https://huggingface.co/WarmBloodAban/Minimax_h3_ref2va_XUELUO_int8_convrot">WarmBloodAban/Minimax_h3_ref2va_XUELUO_int8_convrot · Hugging Face</a></li></ul>2. Text Encoder, VAE &amp; Acceleration LoRA<ul><li>Text Encoder: <code>qwen3vl_32b_minimax_h3_int8_convrot</code></li><li>4-Step Acceleration LoRA: <code>minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16</code></li><li>VAE: <code>minimax_h3_video_vae_int8_convrot</code></li><li>Download Sources:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3">HuggingFace: Comfy-Org/MiniMax-H3</a></li><li><a target="_blank" rel="ugc" href="https://www.google.com/search?q=https://huggingface.co/Kijai/MiniMax-H3-Playground">HuggingFace: Kijai/MiniMax-H3-Playground</a></li></ul></li></ul><h3 id="generation-and-tuning-tips">💡 Generation &amp; Tuning Tips</h3><ul><li>Standard Speed Run (Recommended): Set <code>steps</code> to 4. The workflow will automatically append 2 additional interpolation steps at the end to refine details and smooth out noise.</li><li>Ultra-High Quality Setup:<ul><li>Increase initial <code>steps</code> to 6–8 (or bypass the 4-step LoRA completely and run at 20 steps).</li><li>Scale up the <code>scale_by</code> parameter in the Upscale Latent By node from <code>1.2</code> to a higher value (Note: requires significantly higher VRAM / GPU compute).</li></ul></li></ul>If you find this workflow helpful, please consider giving it a 👍 and sharing your generated runs in the gallery!

Files with SHA256: 646e8ff973708ff6cfbe76cabacc4f3c2f4e3e2948135082f1acbe28633422f3