MiniMax

HyperFlow started with a question we keep exploring: how far can a focused training loop take a strong base model? We took ＠MiniMax H3, ran data-free flow self-distillation on it. The model generates its own outputs, filters for quality, and teaches itself. No external data. No human labels. Better camera control. Better consistency. Better material detail. Significantly lower inference cost. The same distillation method that powers RSI in our own models, now proven on an open foundation. The technique travels. Open source. Go build.

@video_rebirth

We're open-sourcing H3 HyperFlow. > A data-free flow self-distillation technology built on @MiniMax_AI H3. No external training data. Significantly reduces inference cost while preserving frontier model quality. > Full demo:

• videorebirth.com

Been great building this together. The Ref speedup speaks for itself. Thank you for the partnership.

@xieenze_jr

Day-0 Sol Engine + Sol-Attn for @video_rebirth HyperFlow. > 8-step, data-free distill of MiniMax-H3. One LoRA: T2V / FL / Ref. > Sol-Attn is the training-free sparse path — near-lossless vs dense attention, no extra fine-tune. > On 8×B200, same HyperFlow recipe before vs after Sol: > T2V 5s 9.89s → 2.75s (3.6×) T2V 15s 36.93s → 11.85s (3.1×) Ref 5s 15.27s → 2.97s (5.1×) >

• github.com

MiniMax-H3 😊Singularity GGUF (Q3-Q8) -reduce VRAM requirements for use with ComfyUI -specialized in enhanced visual fidelity — HDR clarity, distant-face restoration, cleaner skin, stronger action motion, and improved VFX. 👇

• huggingface.co

Kijai's 😃 Minimax H3 Vae use less VRAM that not only speeds up the int8 VAE even more, but also the original fp16 VAE. 👇

• huggingface.co

We're proud to contribute as an AI partner to the newly launched @Singtel AI Pass, supporting Singapore's SkillsFuture AI Subscription initiative. 🇸🇬 MiniMax H3 (@Hailuo_AI), MiniMax Agent (@MiniMaxAgent) and MiniMax Audio are all part of the program, giving eligible learners across 200+ SWDA-supported AI courses hands-on access to premium AI tools and the opportunity to build practical, real-world AI skills. This partnership brings our mission, "Intelligence with Everyone" to life, and we're excited to see learners across Singapore turn curiosity into skills, ideas into action, and build confidently with AI. Learn more:

• singtel.com

MiniMax Code 研发负责人何涛宣布，MiniMax Code 将在今晚开源。 MiniMax Code 是 MiniMax 的桌面 AI 编程 Agent，支持直接操作项目文件、运行命令和代码审查，也能让多个 Agent 分工协作，手机端还可以远程查看任务进度和批准操作。 MiniMax 其实早就建好了 minimax-code GitHub 仓库，但此前主要用来收集 Bug 和功能建议，目前仓库里仍只有 README 和 Issue 模板，还没有真正的客户端源码。

@Ronny_MiniMax

MiniMax Code is going open source in the next few days. > Built with developers. Now, built in the open. > See you on GitHub.

🎉 MiniMax Design Update 💸 Save 20% on annual plans. 🛠️ Custom model support is now live. Create with your preferred models now>> #MiniMaxDesign #MiniMax

• design.minimax.io

Three paths to faster video attention: compute the same interactions more efficiently, compute fewer in full, or change how information is mixed. Here’s a visual guide. 👇 Thanks to Nunchux AI and collaborators for VC-Attention, bringing training-free low-bit acceleration to MiniMax-H3, with better fidelity than SageAttention2 in the B200 evaluation. The approach balances speed and fidelity: V-Smooth reduces value quantization error, while ExpCast-FP8 makes softmax faster through approximation. Excited to see the community keep building on H3. Could combining low-bit computation with sparse methods like Sol-Attn push efficiency further? We’re looking forward to seeing that explored.

@NunchuxAI

Introducing VC-Attention: fast and accurate low-bit attention without retraining. > On MiniMax-H3, VC-Attention speeds up attention by 1.6× on B200 and 1.5× on B300 over FlashAttention-4, with better fidelity than SageAttention2. It also works with existing sparse attention

过于明目张胆了吧！？ 狗头.JPG

@MaxForAI

朴实无华的商战🤣 > MiniMax Code的Agent 会话迁移功能，当前仅支持ZCode🫡

MiniMax Code CLI is open now: > @MiniMaxAgent: >

• github.com
• x.com

You can run locally Minimax H3-NS/FW Uncensored model. - Ref-to-video NSFW for H3 - se/x/ytime v1.2 for se/?-scene motion - time scenes + coherent motion first, then detail. - Not a checkpoint. - A late-night adapter. - softer sharper detail, slightly more surreal - audio that doesn’t fall apart if you use the right sampler. -

@0x0SojalSec

Uncensored MiniMax-H3-encoder run locally. > - Qwen3-VL encoder, INT8 + ConvRot, built for ComfyUI. - Smaller file than BF16. - Loader: CLIPLoader to minimax - Same node path. - Uncensored label, > -

• huggingface.co
• huggingface.co

Minimax H3 by Hao AI Lab “FastVideo FastH3 V2” is now available on HuggingFace

@haoailab

(1/8) Open weight FastVideo FastH3 V2 is ready! Up to 9x speedup on @NVIDIA Blackwell with lossless quality! > Lossless quality compared to @MiniMax_AI 50-step H3! Judge for yourself below! > - Collab with @nuvalab , NVIDIA FastGen, NVIDIA Enterprise Products - Day 0 @ComfyUI template available! - Day 0 API serving available on @reactorworld ! - Omni-ref is currently training 👨‍🍳 > More comparisons and info below!

• huggingface.co

Minimax 🐻 BUNNY H3 Semantic Bridge (V1) -custom node for Comfy users -works on H3 conditioning, help H3 better understand who is doing what, who is interacting with whom, which object belongs to whom, and how the scene state should continue . 👇

• huggingface.co

视频生成加速框架 FastVideo 最新释出 FastH3 8-Step V2，并全面打通 Mac 本地 MLX 推理链。 它不是单纯的模型搬运库，而是一套覆盖分布式微调与端到端优化的完整工作流（目前 4.4k Stars）。对于想要在本地完成高品质视频生成的开发者，这次更新直接命中了算力和硬件门槛的痛点。 核心工程进展： • 算力开销大幅压缩：新发布的 FastH3 8-Step V2 基于 MiniMax-H3 进行 DMD2 步进蒸馏，引入高达 80% 的视频稀疏注意力（Video Sparse Attention），极大降低了推理成本。 • Apple Silicon 原生支持：告别云端依赖。借助 MLX 框架与 FastMetal-QAD，Mac 用户现在可以原生运行从 1.3B 到 14B 参数的视频生成模型。 • 多端适配与实时编辑：除了主流 NVIDIA 显卡，现已支持 DGX Spark 环境（注：ARM64 架构目前暂无预编译 wheel，需从源码编译 CUDA kernel）。其内置的 Dreamverse 模块可实现本地视频流的实时“Vibe Directing”控制。 如果你习惯在 macOS 桌面上（例如 32GB 统一内存的 Mac 环境）进行模型部署与自动化测试，FastVideo 提供的清晰 CLI 与 Python API 能帮你快速跑通从代码到视频的最后一公里。 项目文档与源码：

• github.com

Thank you, Japan, for the incredible support! 🇯🇵 The "IP × AI" Summit, co-hosted by KAGAMI AI and MiniMax, has officially concluded. More than 150 companies from Japan and the U.S. joined us, alongside distinguished guests, including AKB48 producer Yasushi Akimoto, KAGAMI AI Chairman Takami Kondo, and KADOKAWA editor and producer Motoi Chujo. Global AI leaders @runwayml, @higgsfield, @krea_ai, and @HeyGen came together to explore the future of IP and generative AI. At the summit, MiniMax unveiled H3 IP Edition, bringing the power of MiniMax H3 together with officially licensed Japanese IP for a new generation of AI-powered storytelling. The summit received extensive coverage from major Japanese media, including a dedicated segment on TV Tokyo's WBS, @wbs_tvtokyo. Even more excitingly, over 60 companies have already expressed interest in adopting MiniMax H3 IP Edition. This marks a pivotal step toward bringing officially licensed Japanese IP into the generative AI era. Japanese IP × Global AI. A new

MiniMax Design、実はMidjourney8.2の画像生成ができてしまう。 動画生成はそのままH3 Max Turbo #MiniMaxH3 #MiniMaxDesign @Hailuo_AI

MiniMax Code 2.0 just launched - rebuilt from the core on the Pi Agent framework with remote control baked in. Connect your phone to a live desktop session, follow agent output, send instructions, and handle approvals from anywhere. Your coding agent now fits in your pocket.

Open weights. Shared progress. MiniMax H3 is moving fast. We built MiniMax H3 for video generation with native stereo audio and multimodal reference control. The open-source community is making that capability faster, more accessible, and easier to build on. Recent highlights: • FastH3 — FastVideo, Nuva Lab and NVIDIA: 4-step distillation, now running on DGX Spark and Apple Silicon. • Sol-H3 — NVIDIA’s SANA team: now on DGX Spark with a two-stage H3 + LTX-2.5 pipeline. On 8×B300, the team reports 15 seconds of 768p video + audio in 6.6 seconds of warm inference.* • VDN — Haocheng Xi and the OpenVDN team: rethinking attention for faster H3 inference, with weights, training and inference code released. • PDD — NVIDIA’s distillation method, brought to H3 by Alibaba PAI as 8-step Acc-LoRAs, now supported in ComfyUI. • LightX2V — 4- and 8-step Turbo LoRAs, with workflows for text, image and reference-conditioned video + audio. Behind every release are people training, optimizing, quantizing, testing and sh

Open Minimax-h3_Singularity the H3 fine-tune built for motion, - Stronger melee, VFX, camera moves - HDR that stays sharp in fast action - Distant faces that don’t collapse - Less oily / plastic skin - Native ComfyUI + 4-step turbo LoRA - T2V / I2V / Ref2V / V2V in ComfyUI. If you’re already on H3, this is the quality swap not another random merge. -

• huggingface.co