2.0 updatedDetail could be found in the link below:新版本更新詳細可以參考下方連結<a target="_blank" rel="ugc" href="https://civarchive.com/articles/33990/minimax-h3-simple-workflow-multiref-for-low-vram-updated-20">https://civarchive.com/articles/33990/minimax-h3-simple-workflow-multiref-for-low-vram-updated-20</a>This is a simple workflow for VoxCPM2, featuring a single-speaker setup and a multi-speaker setup. 這是作為輸助語音生成VoxCPM2的Workflow<a target="_blank" rel="ugc" href="https://civarchive.com/models/2865730/voxcpm2-simple-workflow-for-low-v-ram">https://civarchive.com/models/2865730/voxcpm2-simple-workflow-for-low-v-ram</a>I've also written an article about VoxCPM2—feel free to check it out if you're interested!還有關於VoxCPM2的文章, 有興趣可以看看<a target="_blank" rel="ugc" href="https://civarchive.com/articles/34023/why-do-we-still-need-a-audio-generation-model-even-minimax-h3-could-produce-audio">https://civarchive.com/articles/34023/why-do-we-still-need-a-audio-generation-model-even-minimax-h3-could-produce-audio</a><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3228b02e-09cd-43eb-85fb-58ac3a44a3c2/original=true/3228b02e-09cd-43eb-85fb-58ac3a44a3c2.jpeg" alt="image.png" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/132c753c-eba0-40b4-abc4-6337dfbc23b7/original=true/132c753c-eba0-40b4-abc4-6337dfbc23b7.jpeg" alt="image.png" />This is the simplest Minimax H3 Multi Ref workflow, and you can easily convert it to Image to Video or First-Last Frame to Video if needed. Detailed instructions can be found in the article linked below, so I will not repeat them here.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1e8d7610-4a30-4f49-936c-9959e658745d/original=true/1e8d7610-4a30-4f49-936c-9959e658745d.jpeg" /><hr />這是最簡單的 Minimax H3 Multi Ref Workflow, 如果想轉成image to Video 或 First Last Frame to Video就可以了詳細說明可以參考下方連結的文章, 這兒就不重覆了<a target="_blank" rel="ugc" href="https://civarchive.com/articles/33959/for-someone-whom-want-to-use-minimax-h3-in-low-vram">https://civarchive.com/articles/33959/for-someone-whom-want-to-use-minimax-h3-in-low-vram</a><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3740a775-1db3-497a-b26e-9a033ed11580/original=true/3740a775-1db3-497a-b26e-9a033ed11580.jpeg" />📊 Hardware Requirements<ul><li>8GB VRAM is usable but suboptimal.</li><li>12GB VRAM minimum is highly recommended.</li><li>32GB+ System RAM is required.</li><li>Note: VRAM will inevitably fill up, forcing the system to offload data to system RAM.</li></ul>➡️ Model &amp; Text Encoder Selection<ul><li>Model: Choose the INT8 convrot version (Civitai Link). Avoid the GGUF model due to excessive quality degradation.</li><li>Text Encoder: Stick to INT8. Using GGUF doubles the generation time and carries a very high risk of OOM (Out of Memory) errors.</li><li>CLIP: GGUF is perfectly fine here; Q4KM is the top choice.</li><li>Model Versions:<ul><li>fl2v version: For Image to Video / First Image / First-Last Image tasks.</li><li>ref2va version: For Multi-reference tasks (Video / Dialogue Audio / BGM Audio).</li></ul></li></ul>💡 LoRA Settings<ul><li>Low VRAM Adjustments: You can consider 4 steps, but it is not recommended. It degrades the quality too much, turning H3's best features—the dialogue and sound effects—into complete garbage.</li><li>Recommended: Use 8 steps if your hardware allows.</li><li>768p Version: Even for the 4-step 768p model, the creator still recommends 6–8 steps. This specific 768p version delivers highly impressive image and audio quality.</li><li>Note: LoRAs are also split into fl2v and ref2v versions. Download the one that matches your specific workflow needs.</li></ul>🗒 VAE Configuration<ul><li>Video VAE: Consider using the INT8 version. It has a slightly smaller file size with negligible difference in final quality.</li><li>Audio VAE: The file size is already inherently small, so the version choice is inconsequential.</li></ul>🛠 Workflow &amp; Speed Optimization<ul><li>Keep it Simple: Due to Low VRAM limitations, keep your main workflow minimal. Move non-generation nodes like Prompt Enhancer and Upscale into a separate standalone workflow to prevent OOM errors.</li><li>Prompt Enhancer (Chinese): Chinese language users can consider T8. It supports running local models natively without requiring an API.</li><li>Sage Attention: It is strongly recommended to integrate Sage attention into your workflow, as it drastically improves generation speed. A tutorial link is provided below.</li><li>Easy Cache Node: This node yields marginal performance gains when paired with 4–8 step LoRAs. However, it is retained in the workflow since it introduces zero visual degradation.</li></ul>📥 Workflow Downloads &amp; Resources<ul><li>Sample Workflow: The attached file contains a simple workflow that I am currently using. Feel free to download and use it if needed.</li><li>This Workflow: This is the simplest Minimax H3 Multi Ref workflow. You can easily convert it to Image to Video or First-Last Frame to Video based on your needs.</li><li>Detailed Guide: For a comprehensive breakdown, please refer to the article linked below, so I will not repeat the details here.</li></ul>🌟 Minimax H3 完整設定與優化指南📊 硬體需求<ul><li>8GB VRAM：勉強可用，但體驗不佳。</li><li>建議最低 12GB VRAM：強烈推薦的硬體門檻。</li><li>系統 RAM 32GB 以上：不可或缺的配置。</li><li>備註：不論如何調整，VRAM 遲早都會裝不下，系統最後還是會自動 offload（分流）到系統 RAM。</li></ul><hr />➡️ 模型與 Text Encoder 選擇<ul><li>主要模型：請認準並選用 INT8 convrot 版本（Civitai 連結）。請別選 GGUF 格式，因為畫面劣化實在太嚴重。</li><li>Text Encoder：同樣請堅持使用 INT8。如果換成 GGUF，生成時間足足會慢上一倍，而且有極大機率會直接爆顯存（OOM）。</li><li>CLIP：這邊可以使用 GGUF，首選為 Q4KM。</li><li>模型版本對應：<ul><li>fl2v 版本：適用於 Image to Video（圖生影片）/ First Image（首幀）/ First Last Image（首尾幀）任務。</li><li>ref2va 版本：適用於 Multi-reference（多重參考）任務（如：影片 / 對白音訊 / BGM 背景音樂）。</li></ul></li></ul><hr />💡 LoRA 設定<ul><li>低顯存（Low VRAM）調整：雖然可以考慮設定為 4 steps，但非常不建議。因為劣化太過嚴重，會導致 H3 最強大的音效與對白直接變成垃圾。</li><li>推薦設定：在硬體條件允許的情況下，最好使用 8 steps。</li><li>768p 版本：就算是 4 steps 的 768p 版本，原創作者的官方建議依舊是 6–8 steps。其中這個 768p 版本的畫質與聲音表現都非常出色。</li><li>備註：LoRA 同樣有區分 fl2v 和 ref2v 兩種版本，請務必根據自己的工作流需求對應下載。</li></ul><hr />🗒 VAE 配置<ul><li>Video VAE（影片 VAE）：可以考慮使用 INT8 版本。檔案體積略小一些，且與原版相比幾乎沒有顯著差別。</li><li>Audio VAE（音訊 VAE）：因為這個元件本來的體積就非常小，所以使用什麼版本都無所謂。</li></ul><hr />🛠 工作流與速度優化<ul><li>保持簡單：由於是低顯存（Low VRAM）環境，主要工作流請盡量保持精簡。像 Prompt Enhancer（提示詞增強）、Upscale（放大） 這類不涉及核心生成的節點，請獨立拆分到另一個工作流中，否則極度容易引發 OOM 錯誤。</li><li>Prompt Enhancer（中文使用者推薦）：中文使用者可以考慮使用 T8 這個 Prompt Enhancer，它支援本地模型運作，不需要另外串接 API。</li><li>Sage Attention：強烈建議在工作流中加入並啟用 Sage attention，這會讓整體的生成速度產生極其明顯的差距。如果需要教學教程，可以參考下方提供的連結。</li><li>Easy Cache 節點：這個節點在搭配 4–8 steps 的 LoRA 時提速效果雖然沒有很顯著，但因為它不會造成圖片劣化，所以依然選擇保留在工作流中。</li></ul><hr />📥 工作流下載與資源<ul><li>附帶工作流：附件中包含了我目前正在使用的簡單工作流，有需要的人可以自行下載使用。</li><li>本工作流說明：這是目前最精簡的 Minimax H3 Multi Ref（多重參考）工作流，如果後續想要自行轉換成 Image to Video 或 First Last Frame to Video 也完全沒問題。</li><li>詳細教學：更多細節說明與圖文教學可以參考下方連結的文章，此處就不再重複贅述。</li></ul><hr /><hr />

Minimax H3 Multi Ref workflow for low VRAM

Basic simple workflow of H3.Use Set Get Node to avoid spaghettiAnd hide most of the node in subgraph to keep tidy.Saperate the Prompt Enhancer in different workflow for optional use.

Files with SHA256: 204012b91b3475b5c7a26b66f0370110d33b4d7aea199075b01a4932faf60c84