<h1 id="qw21-turbo-0.1-by-viggle">QW21-Turbo-0.1 by <a target="_blank" rel="ugc" href="https://viggle.ai/">Viggle</a></h1><h3 id="steps-4-16-cfg1-t2ii2i">Steps 4-16 CFG=1 T2i/i2i</h3>演示图片同时使用了 Viggle/Qwen-Image-2.1-viggle-turbo 和 REDQW21 UNLOCKED LoRA (with NV DLSS_NR) 仅作为[ 研究或评估 ]用途内使用 / only for the research or evaluation purposesThe demo images uses both Viggle/Qwen-Image-2.1-viggle-turbo and REDQW21 UNLOCKED LoRA simultaneously Built with Qwen, Applicable to Qwen-Image-2.1Viggle/Qwen-Image-2.1-viggle-turbo source repo: <a target="_blank" rel="ugc" href="https://huggingface.co/Viggle/Qwen-Image-2.1-viggle-turbo">https://huggingface.co/Viggle/Qwen-Image-2.1-viggle-turbo</a>All credit belongs to the original model author. License is the same as the original model.Model: <a target="_blank" rel="ugc" href="https://huggingface.co/Viggle/Qwen-Image-2.1-viggle-turbo">Viggle/Qwen-Image-2.1-viggle-turbo</a> · Built with Qwen — distilled from <a target="_blank" rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1">Qwen/Qwen-Image-2.1</a>, which is released under the <a target="_blank" rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE">Qwen RESEARCH LICENSE AGREEMENT</a> (non-commercial: research or evaluation purposes only).Built with Qwen. A 4-step distilled student of <a target="_blank" rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1">Qwen/Qwen-Image-2.1</a>, trained by Viggle with Distribution Matching Distillation. It does both text-to-image and instruction-driven editing with 1–3 reference images in 4 transformer passes instead of 40, with no classifier-free guidance.Two students are shipped — pick one:<ul><li><code>transformer/</code> — full fine-tuned transformer (bf16, 14.2 GB). Replaces the base transformer; exact, no adapter. This is what the <a target="_blank" rel="ugc" href="https://huggingface.co/spaces/Viggle/Qwen-Image-2.1-viggle-turbo">demo Space</a> runs by default and the one we currently recommend — in our qualitative comparison it edits more faithfully than the LoRA.</li><li><code>Qwen-Image-2.1-viggle-turbo-4step-lora-r64.safetensors</code> — LoRA adapter (rank 64, 340 MB) loaded on top of the base transformer at runtime. Smaller download, slightly weaker.</li></ul>Status: v0.1 preview, work in progress — this release still falls short of the base model. On complicated image editing (multi-reference composition, face swaps, identity-preserving edits, instructions with several constraints) it is clearly worse than the 40-step base model. Text-to-image at 4 steps is usable. We are still working on it and will update this repository as the distillation improves; treat the current weights as a preview, not a replacement for the base model.<hr /><h1 id="minimax-h3-turbo-by-lightx2v"><a target="_blank" rel="ugc" href="https://github.com/ModelTC/Minimax-H3-Turbo">Minimax-H3-Turbo by LightX2V</a></h1>Minimax-H3-Turbo provides batch MiniMax-H3 inference and NFE/LoRA comparisons.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/85a4f8c1-9688-4bd7-ac02-0bb3af81bdf7/original=true/85a4f8c1-9688-4bd7-ac02-0bb3af81bdf7.jpeg" /><h2 id="repos">Repos</h2>Kijai/MiniMax-H3_comfy <a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/MiniMax-H3_comfy/tree/main/loras">Minimax-H3-Turbo ComfyUI v0.1</a>lightx2v/Minimax-h3-Turbo <a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Minimax-h3-Turbo/tree/main">Minimax-h3-Turbo Diffusers/ComfyUI v1.0</a><h2 id="note">Note</h2>Its current visual quality can be reviewed in the linked examples, and the image details still need improvement. An enhanced version is in development.<h3 id="this-edition-is-compatible-with-the-comfyui-pruned">This Edition is compatible with the ComfyUI-pruned</h3><h3 id="minimax-h3-fl2va-model20b.">MiniMax H3 FL2VA model（20B）.</h3><h2 id="roadmap">Roadmap</h2><ol><li>Improve the visual details and overall quality of FL2V Turbo.</li><li>Develop distillation based on Ref2V.</li></ol><h2 id="preview">Preview</h2><a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Minimax-h3-Turbo/discussions/2">https://huggingface.co/lightx2v/Minimax-h3-Turbo/discussions/2</a>Apache-2.0 license

[ Accelerator LoRAs ] Applicable to QW21 | Minimax H3 | HYVideo TURBO/LIGHTNING/DMD/FAST

<a target="_blank" rel="ugc" href="https://github.com/ModelTC/Minimax-H3-Turbo">Minimax-H3-Turbo by LightX2V</a>Minimax-H3-Turbo provides batch MiniMax-H3 inference and NFE/LoRA comparisons.Note<a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Minimax-h3-Turbo/discussions/2">v0.1 is a preview release</a>. Its current visual quality can be reviewed in the linked examples, and the image details still need improvement. An enhanced version is in development.This Edition is compatible with the ComfyUI-pruned MiniMax H3 FL2VA model（20B）.Roadmap<ol><li>Improve the visual details and overall quality of FL2V Turbo.</li><li>Develop distillation based on Ref2V.</li></ol>Preview<a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Minimax-h3-Turbo/discussions/2">https://huggingface.co/lightx2v/Minimax-h3-Turbo/discussions/2</a>Apache-2.0 license

minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors

minimax_h3_fl2v_lightx2v_turbo_4step_v1.0_768p_resized_avg_rank_31_bf16.safetensors

minimax_h3_fl2v_turbo_4step_v1.0_768p_comfyui_bf16.safetensors

minimax_h3_fl2v_lightx2v_turbo_8step_v1.0_resized_avg_rank_24_bf16.safetensors

minimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors

minimax_h3_fl2v_lightx2v_turbo_4step_v0.1_comfy_resized_avg_rank_21_bf16.safetensors

minimax_h3_fl2v_turbo_4step_v0.1_768p_sla_comfyui_bf16.safetensors

lightx2v / Minimax-h3 1.0

TURBO(Viggle)QW21 v0.1

Llavallama3 FP8+VAE FP16

Hunyuan FASTVIDEO 2024

taomate_fl2va_3step_ema uploaded Current best settings: 4-6 steps, er_sde/euler, simple, Strength 0.75 minimax_h3_ref2v_turbo_8step_v1.0_768p_comfyui_bf16: Current best settings: 8-12 steps, er_sde/res_multistep, simple, Strength 1.0 lightx2v_4step_768p_v1.1+ v1.2: Current best settings: 5-8 steps, er_sde/res_multistep, simple. Strength 1.0 larryvrh_v4_600_ema: Current best settings: 6-8 steps, er_sde/res_multistep, simple. Strength 1.0 All examples are generated with 1.0 strengthI did not train, distill, or create the original Turbo LoRA weights. This model page provides only modified compatibility versions intended to let the remaining compatible LoRA adapters load with ComfyUI's built-in LoRA loader when using the pruned/curve-form model. Full credit for the original h3_turbo_4step_pruned belongs to <a target="_blank" rel="ugc" href="https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora">larryvrh</a>. And <a target="_blank" rel="ugc" href="https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI">drbaph</a> for the pruned comfy-compatible version Full credit for the original lightx2v_4step_turbo_v0.1 belongs to the <a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Minimax-h3-Turbo">lightx2v</a> team and <a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/MiniMax-H3_comfy">kijai</a> for the comfy-compatible version Full credit for the original taomate_fl2va_3step_ema belongs to the <a target="_blank" rel="ugc" href="https://github.com/TaoLiveAIGC">TaoLiveAIGC</a> team. And <a target="_blank" rel="ugc" href="https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI">drbaph</a> for the comfy-compatible version I'll be uploading different Minimax H3 Turbo LoRAs here and crediting the original creators

Minimax H3 Turbo LoRas

lightx2v_8step_v1.0

taomate_fl2va_3step_ema

lightx2v_4step_768p_v1.2

lightx2v_ref2v_8step_v1.0

lightx2v_4step_768p_v1.1

lightx2v_4step_768p_v1.0

lightx2v_4step_v0.1

larryvrh_4step_pruned

<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/C78AQztbwEo" start="0"></iframe></div><h2 id="fan-registration-bonus">🎁 Fan Registration Bonus  </h2>Register to receive 1000 RH coins:  🔗 <a target="_blank" rel="ugc" href="https://www.runninghub.ai/?inviteCode=rh-v1325">https://www.runninghub.ai/?inviteCode=rh-v1325</a><h2 id="workflow">⚙️ Workflow  </h2>Ultimate Simplified MiniMax H3 Full-Feature Auto Prompt Workflow (Lenient moderation, NSFW content allowed)  🔗 Experience link: <a target="_blank" rel="ugc" href="https://www.runninghub.ai/post/2088114005419282433/?inviteCode=rh-v1325">https://www.runninghub.ai/post/2088114005419282433/?inviteCode=rh-v1325</a>🔗 Civitai free download link: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2882357/">https://civarchive.com/models/2882357/</a><h2 id="models">📦 Models  </h2>🔗 <a target="_blank" rel="ugc" href="https://civarchive.com/models/2879272/minimax-h3-remix">https://civarchive.com/models/2879272/minimax-h3-remix</a><h2 id="turbo-lora">⚡ Turbo LoRA  </h2>🔗 <a target="_blank" rel="ugc" href="https://civarchive.com/api/download/models/3257181?fileId=3140963">https://civarchive.com/api/download/models/3257181?fileId=3140963</a><h2 id="clip-(uncensored">🎬 Clip (uncensored） </h2>🔗 <a target="_blank" rel="ugc" href="https://civarchive.com/api/download/models/3254025?fileId=3137729">https://civarchive.com/api/download/models/3254025?fileId=3137729</a>🛠️ Required Nodes  - Easy-H3: <a target="_blank" rel="ugc" href="https://github.com/FX-FeiHou/ComfyUI-FeiHou-Easy-H3">https://github.com/FX-FeiHou/ComfyUI-FeiHou-Easy-H3</a>  - FeiHou-Toolbox: <a target="_blank" rel="ugc" href="https://github.com/FX-FeiHou/ComfyUI-FeiHou-Toolbox">https://github.com/FX-FeiHou/ComfyUI-FeiHou-Toolbox</a><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2c2c0111-648a-4b8b-82cc-7a5528120973/original=true/2c2c0111-648a-4b8b-82cc-7a5528120973.jpeg" /><h1 id="minimax-h3-remix">🌟 MiniMax-H3 Remix</h1>Dual‑H3 Hierarchical‑Recombination Merge Dual‑H3 Hierarchical‑Recombination Merge is a merged checkpoint built with a custom hierarchical recombination technical pipeline based on two H3 base models. It adopts targeted layer‑rearrangement instead of naive weight averaging, preserving the solid stability of the base model while importing artistic bias, dynamic expression and fine‑grained detail features from the overlay model. LoRA weights are statically fused internally, requiring no external LoRA file during inference.Two precision variants are released for different hardware conditions: full‑precision BF16 for quality‑oriented workflows, and ComfyUI‑native INT8 ConvRot for low‑VRAM deployment scenarios.<blockquote>Recommended to use alongside the FeiHou Easy H3 custom nodes and workflow: <a target="_blank" rel="ugc" href="https://github.com/FX‑FeiHou/ComfyUI‑FeiHou‑Easy‑H3">https://github.com/FX‑FeiHou/ComfyUI‑FeiHou‑Easy‑H3</a></blockquote>⚠️ Please follow local laws and exercise responsible usage when generating content.<h2 id="model-variants">📦 Model Variants</h2><h3 id="bf16-variant-(full-precision)">🔹 BF16 Variant (Full Precision)</h3>Optimized for complete numerical precision and maximum detail retention. Best for high‑quality local generation, rendering and art creation.<ul><li>Installation: Place the safetensors file under your <code>ComfyUI\models\diffusion_models</code> folder</li><li>Usage: Load via FeiHou Easy H3 workflow</li><li>Recommend VRAM: ≥24 GB</li></ul><h3 id="int8-convrot-variant-(comfyuinative)">🔹 INT8 ConvRot Variant (ComfyUI‑Native)</h3>8‑bit quantized release with native ConvRot implementation for ComfyUI. Prioritizes VRAM reduction and practical deployment convenience.<ul><li>Installation: Place the safetensors file under your <code>ComfyUI\models\diffusion_models</code> folder</li><li>Usage: Load via FeiHou Easy H3 workflow; no extra quantization plugin required</li><li>Recommend VRAM: ≥14 GB</li></ul>⚠️ Except for numerical precision and memory footprint, both variants share identical layer structure, merge logic and inherent model characteristics. ⚠️ LoRA effects are statically baked into both checkpoints; external LoRA files are not needed.<h3 id="merge-technical-notes">🎉 Merge Technical Notes</h3>This checkpoint is produced through dual‑H3 hierarchical recombination workflow:<ul><li>One H3 model acts as the stable base foundation</li><li>The second H3 model performs weight overlay within designated response blocks</li><li>LoRA effects are statically fused directly into checkpoint weights</li><li>Targeted re‑orchestration is applied to key modulation layers including AdaLN, instead of simple weight averaging</li><li>Retains base‑model stability while inheriting overlay‑model visual tendency, dynamic performance and fine details</li></ul><h2 id="core-capabilities">✨ Core Capabilities</h2><ul><li>Stable Base Generation Reliable baseline output inherited from the primary H3 base model, avoiding common merge‑induced artifacts and model collapse.</li><li>Enhanced Visual &amp; Dynamic Features Absorbs painting tendency, dynamic expressiveness and subtle detail features brought by the overlay H3 model via block‑level targeted recombination.</li><li>Zero Extra LoRA Dependency All LoRA contributions are statically fused into checkpoint weights. No additional LoRA needs to be loaded during inference.</li><li>Flexible Hardware Adaptation Choose BF16 for supreme quality or INT8 ConvRot to run on mid‑range GPUs with limited video memory.</li><li>Optimized for FeiHou Easy H3 Designed for the FeiHou Easy H3 node suite and workflow for best generation performance.</li></ul><h2 id="recommended-settings">⚙️ Recommended Settings</h2><ul><li>Sampler: <code>res_multistep</code></li><li>Scheduler: <code>simple</code></li><li>Steps: 8‑12</li><li>Turbo LoRA weight: <code>0.75</code></li></ul>This combination balances generation stability, fine‑detail preservation and visual consistency, suitable for most creative workflows.<h2 id="use-cases">🎯 Use Cases</h2><ul><li>General‑purpose text‑to‑image generation</li><li>Dynamic‑oriented artwork and illustration creation</li><li>Complex‑detail‑heavy scene rendering</li><li>ComfyUI local deployment on high‑end or mid‑range consumer GPUs</li><li>Workflows expecting both stable baseline and enhanced stylistic features</li><li>Iterative creation, concept art, character illustration</li></ul><h2 id="known-tips">📌 Known Tips</h2><ol><li>Do not load additional matching LoRA together with this checkpoint, since LoRA effects are already statically fused. Applying extra LoRA may cause over‑saturation or distorted outputs.</li><li>INT8 ConvRot variant has minor precision loss compared to BF16. If you observe obvious detail degradation on key creative projects, switch over to the BF16 version.</li><li>This is a Checkpoint Merge, not a native trained H3 checkpoint. Some edge‑case rare concepts may behave differently from original base H3 models.</li></ol>

Files with SHA256: 2339acdf19bfe123f46b971ea35d367a84adb85de43627e1eceafa5a5b2b111e