<h1 id="minimax-h3-turbo-lora">⚡ MINIMAX H3 TURBO LoRA</h1><h2 id="4-step-audio-+-video-acceleration">4-STEP AUDIO + VIDEO ACCELERATION</h2><blockquote>EXTREME SPEED // NATIVE AUDIO // COMFYUI READY</blockquote>MiniMax H3 Turbo LoRA is an experimental acceleration adapter designed to drastically reduce MiniMax H3 video generation time while preserving its native synchronized stereo audio capabilities.It reduces sampling from the usual higher step count down to only 4 steps, providing extremely fast previews and significantly shorter generation times.The LoRA supports both:<pre><code>OFFICIAL EKKIVOK WORKFLOW :</code></pre><a target="_blank" rel="ugc" href="https://civarchive.com/models/2838258/minimax-h3-4-steps-turbo-video-aio-workflow">https://civarchive.com/models/2838258/minimax-h3-4-steps-turbo-video-aio-workflow</a><pre><code>TEXT TO VIDEO
IMAGE TO VIDEO
FIRST / LAST FRAME VIDEO</code></pre><hr /><h1 id="early-experimental-release">⚠️ EARLY EXPERIMENTAL RELEASE</h1>This LoRA is currently an experimental and under-trained prototype.The 4-step mode provides impressive speed, but it may produce:<ul><li>Softer image quality</li><li>Reduced fine details</li><li>Unstable textures</li><li>Motion ghosting</li><li>Motion trails</li><li>Lower overall quality compared with the full MiniMax H3 model</li></ul>For improved visual quality, using 6 to 8 steps may produce cleaner results while still remaining much faster than standard generation.<hr /><h1 id="recommended-settings">🚀 RECOMMENDED SETTINGS</h1><pre><code></code></pre><pre><code>LoRA Strength : 1.0
Steps : 6–8 recommended
Fast Preview : 4 steps
Scheduler : simple
CFG : 1.0 or workflow default
Frame Rate : 24 FPS
Resolution : multiples of 32
Short Edge : approximately 768 px</code></pre><h3 id="recommended-lora-file">Recommended LoRA file</h3><pre><code></code></pre><pre><code>minimax_h3_turbo_4step_ckpt500.safetensors</code></pre>The non-EMA <code>ckpt500</code> version is currently the recommended checkpoint.The EMA version may produce softer results and can be more prone to ghosting or visible motion trails.<hr /><h1 id="native-synchronized-audio">🔊 NATIVE SYNCHRONIZED AUDIO</h1>MiniMax H3 generates video and audio together:<pre><code></code></pre><pre><code>VIDEO STREAM
+
32 kHz STEREO AUDIO</code></pre>The audio is generated natively and synchronized directly with the video.A dedicated Turbo sampler is required because MiniMax H3 uses different sampling trajectories for video and audio:<pre><code></code></pre><pre><code>Video Flow Shift : 12
Audio Flow Shift : 3</code></pre>Using a standard ComfyUI sampler at only 4 steps may incorrectly sample the audio component, resulting in distorted, saturated, or unusable sound.<hr /><h1 id="required-comfyui-custom-node">🧩 REQUIRED COMFYUI CUSTOM NODE</h1>Install the following custom node through ComfyUI Manager:<pre><code></code></pre><pre><code>ComfyUI-MiniMax-H3-Turbo</code></pre>Official repository:<a target="_blank" rel="ugc" href="https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo">https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo</a>The node pack includes:<pre><code></code></pre><pre><code>MiniMax-H3 Turbo LoRA
MiniMax-H3 Turbo Sampler</code></pre>The custom node also supports pruned MiniMax H3 checkpoints by automatically applying the required AdaLN LoRA layers during generation.<hr /><h1 id="base-model-compatibility">🧠 BASE MODEL COMPATIBILITY</h1>The Turbo LoRA is compatible with several MiniMax H3 model formats:<pre><code></code></pre><pre><code>BF16
INT8 ConvRot
Pruned INT8
Pruned FP8</code></pre>The dedicated Turbo node automatically detects the model structure and applies the LoRA appropriately.<hr /><h1 id="installation">📂 INSTALLATION</h1>Place the LoRA file inside:<pre><code></code></pre><pre><code>ComfyUI/models/loras/</code></pre>Recommended workflow structure:<pre><code></code></pre><pre><code>Load Diffusion Model
 ↓
MiniMax-H3 Turbo LoRA
 ↓
MiniMax-H3 Turbo Sampler
 ↓
Video and Audio Output</code></pre>Recommended sampling configuration:<pre><code></code></pre><pre><code>Scheduler : simple
Steps : 4–8</code></pre>Do not use a standard sampler for the 4-step audio workflow.<hr /><h1 id="original-creator">👤 ORIGINAL CREATOR</h1><h2 id="larryvrh">Larryvrh</h2>This LoRA was originally created and trained by Larryvrh.<h3 id="original-hugging-face-repository">Original Hugging Face repository</h3><a target="_blank" rel="ugc" href="https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora">https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora</a><h3 id="official-comfyui-custom-node">Official ComfyUI custom node</h3><a target="_blank" rel="ugc" href="https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo">https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo</a>All credit for the LoRA training, development, research, and dedicated Turbo sampler belongs to Larryvrh.<hr /><h1 id="civitai-upload-information">📤 CIVITAI UPLOAD INFORMATION</h1><pre><code></code></pre><pre><code>Original Creator : Larryvrh
Civitai Uploader : EKKIVOK</code></pre>This Civitai page is a community re-upload intended to make the model easier to discover, download, and use with ComfyUI.I am not the original creator of this LoRA.Please support and follow the original creator through the official repositories linked above.<hr /><h1 id="speed-and-quality-guide">⚡ SPEED AND QUALITY GUIDE</h1><pre><code></code></pre><pre><code>4 STEPS = MAXIMUM SPEED / PREVIEW QUALITY
6 STEPS = BALANCED TURBO MODE
8 STEPS = BEST CURRENT TURBO QUALITY</code></pre>The Turbo LoRA is especially useful for:<ul><li>Rapid prompt testing</li><li>Motion previews</li><li>Seed comparison</li><li>Fast workflow testing</li><li>Video drafts</li><li>Quickly testing image-to-video inputs</li><li>Reducing MiniMax H3 sampling time</li></ul>For maximum final quality, the standard MiniMax H3 model without the Turbo LoRA and with a higher step count may still produce better results.<hr /><blockquote>CREATED BY LARRYVRH UPLOADED TO CIVITAI BY EKKIVOK MINIMAX H3 // TURBO MODE // FUTURE IN MOTION</blockquote>

MINIMAX H3 TURBO LoRA ⚡

⚡ MiniMax H3 Turbo — V4 Step600🚀 Faster MiniMax H3 generation with a better quality / speed balanceThis LoRA is designed to make MiniMax H3 significantly faster while preserving as much of the original model quality as possible.Compared to earlier Turbo experiments, V4 Step600 aims for a cleaner and more refined result, especially on:✨ Faces ✨ Fine details ✨ Skin rendering ✨ Motion consistency ✨ Overall image cleanlinessIt is a great choice if you love the quality of MiniMax H3 but want to reduce generation time for everyday use.🎯 What can you expect?With the right workflow, this LoRA can generate strong results at around:⚡ 6–8 steps recommendedThis makes it very useful for:✅ Faster generations ✅ Prompt testing ✅ Rapid iteration ✅ Cinematic scenes ✅ Character animation ✅ General H3 video generation ✅ Testing multiple seeds quicklyFor many scenes, 6 steps can already look very good, while 8 steps is usually safer for more complex motion.⚙️ Recommended settingsLoRA Strength: <code>1.0</code> Steps: <code>6–8</code> Scheduler: <code>Simple</code> Base Model: Compatible MiniMax H3 model💡 If your scene contains complex movement, fast camera motion or several interacting subjects, try using 8 steps instead of 6.⚠️ Known limitationsThis is still a few-step acceleration LoRA, so some compromises can appear compared to a full-step H3 generation.Possible issues include:⚠️ Motion smear during very fast movement ⚠️ Temporal instability in complex scenes ⚠️ Hands or fingers may become less consistent ⚠️ Small moving objects can lose detail ⚠️ Multiple-character interactions may be less stable ⚠️ Very aggressive camera motion can reduce coherence ⚠️ Audio quality may vary depending on the workflowIn most cases, increasing the step count slightly can help improve stability.💡 Recommended useThis version is especially good if you want:🔥 Fast H3 generations 🔥 Good visual quality 🔥 Better refinement than older Turbo versions 🔥 A strong balance between speed and qualityFor final renders with lots of movement, 8 steps is generally the safest starting point.👤 CreditsOriginal creator: Larryvrh Uploaded by: EKKIVOK❤️ Please support the original creator if you enjoy this LoRA.

Files with SHA256: 5f3a626cd72c93a8b9318d6760c510bc5092d2ab13aaba1f932c5bab07a416d3