Skip to content

Companies and Organizations

MiniMax / Hailuo Latest

MiniMax updates: the Hailuo video generation model, speech and music capabilities, and open-source and commercialization progress.

71 picksLast 30 days 43 itemsCollected 316 items

updated

Pick archive · page 4

Oct 8Thu61–71
  1. Stuubzzz · 视频工作流实测71

    MiniMax H3 Four-Step Output: PDMD LoRA vs. NVIDIA SoL Refiner Benchmarked on RTX 5090

    The author benchmarked MiniMax H3's PDMD 4-step LoRA and NVIDIA SoL Refiner on an RTX 5090: PDMD averaged 101 seconds to output a 6.6-second clip at 2520×1440, while the 20-step Quality path took 753 seconds.

    Why it matters: The author benchmarked PDMD 4-step LoRA and NVIDIA SoL Refiner on an RTX 5090, providing itemized timing and image quality differences that can serve as a selection reference.

Oct 7Wed
Oct 6Tue
  1. FastVideo 视频模型 · 模型发布62

    FastVideo Releases FastH3 Trim Lightweight Model and ComfyUI Repackaged Files

    FastVideo released the experimental lightweight model FastH3 Trim, reducing transformer blocks from 50 to 42, using rank-16 timestep conditioning, 8-step sampling, with a 4.2x smaller size than the base H3 and faster than FastH3 V2.

    Why it matters: FastH3 Trim offers a lighter video generation solution with 42 layers, 8-step sampling, and 4.2x size compression, along with ComfyUI files and parameters for various precisions.

  2. 阡陌的小站 · 本地创作实战70

    Tutorial: deploying Qwen-Image-2.1 and MiniMax H3 locally in ComfyUI with 8GB VRAM

    Qianmo published a tutorial on generating images and videos locally via ComfyUI on an RTX 4070 Laptop with about 8GB VRAM, using the Qwen-Image-2.1 jailbreak GGUF workflow for images and MiniMax H3's 04_i2v_fl2v_4step_sla_lowvram workflow for videos.

    Why it matters: A complete path to running Qwen-Image-2.1 and MiniMax H3 under 8GB VRAM, including model files, directory structure and parameter configuration, reproducible by comparison.

Oct 3Sat
  1. FastVideo 视频模型 · 模型发布62

    FastVideo releases the FastH3 Trim 8-step audio-video generation model

    FastVideo released the experimental model FastH3 Trim, which removes 8 of the least impactful blocks from FastH3 V2's 50 MiniMax H3 transformer blocks and replaces each block's AdaLN timestep projection with a shared rank-16 basis, trained with 8-step DMD2 (checkpoint 300), enabling 8-step text-to-video generation with synchronized audio.

    Why it matters: FastH3 Trim uses pruning plus shared AdaLN to bring 8-step audio-video generation to consumer-grade hardware, and can be compared against V2 to weigh image quality against size.

  2. note · wise_azalea8300 · 图像视频创作分享65

    ComfyUI Video Perspective Change Workflow: Running Locally with MiniMax H3 and Angle Change LoRA

    The author released a ComfyUI workflow that uses the video generation model MiniMax H3 combined with an angle change LoRA to alter the camera perspective of existing videos, enabling wide-angle, side, low-angle, and 360° Orbit effects around a target. The workflow runs entirely locally and is free, with various techniques allowing execution in an 8GB low-VRAM environment. It supports both SFW and NSFW content, and the article includes download links and import instructions.

    Why it matters: The author published a ComfyUI workflow for changing perspectives using MiniMax H3 and an angle change LoRA, along with a local running solution for 8GB low-VRAM environments.

Oct 1Thu
  1. note · eeeemo_japan · 图像视频创作分享66

    MiniMax open-sources video generation model H3: runs locally, supports Omni Reference

    MiniMax freely released the open-source video generation model MiniMax H3 in August 2026; it can be installed and run on your own computer, reducing the cost of repeated regeneration to electricity. It supports Omni Reference, allowing up to 9 images, 3 videos, and 3 music tracks as references to maintain character and scene consistency; according to ASCII.jp, its cloud version's performance is considered comparable to Seedance 2.0.

    Why it matters: The open-source video model pushes the cost of a single generation down to electricity, and the article gives VRAM thresholds and licensing regional restrictions to help assess the feasibility of local production.

Sep 30Wed
  1. note · synth_brain · 图像视频创作分享65

    Three ways to make MiniMax H3 high-speed and high-resolution: Fused-Turbo-Int8, Spectrum, and Latent Upscaler

    The author shared a high-speed, high-resolution all-in-one ComfyUI workflow for MiniMax H3, the core of which is replacing the base model with the minimax-h3-fused-turbo-int8-convrot model, which integrates FL2V and Ref2VA, has a built-in 4STEP TurboLoRA and Mystic V2 LoRA that suppresses jitter and ghosting, and is about 21GB in size.

    Why it matters: The author tested MiniMax H3's model replacement and speed-up nodes, providing a comparison showing an 800×800 eight-second video dropping from 7 minutes 13 seconds to 3 minutes 37 seconds, which can be transferred to existing video production workflows.

Sep 25Fri
  1. Hedra 官方博客62

    Hedra Compares API Pricing Across 11 Video Models Including Seedance, Kling, Veo, and MiniMax: 10-Second 1080p with Audio Ranges from $0.60 to $7.91

    Hedra compared API pricing across 11 video models for the same task (10 seconds, 1080p, native audio, one starting frame plus a text prompt), with the lowest at $0.60 for LTX-2.3 Fast and the highest at $7.91 for Seedance 2.5—a 13x difference. Nine models meet the spec; Veo 3.1 maxes out at 8 seconds and MiniMax H3 has no 1080p setting, so both are listed separately as close substitutes.

    Why it matters: For the same 10-second 1080p with audio task, API unit prices across 11 video models differ by 13x, along with a reusable method for model selection and cost verification.

Sep 18Fri
  1. fal 官方博客62

    fal Details H3 Max: Built on the fal Inference and Training Platform

    fal's official blog explains how H3 Max completes the journey from post-training to distribution across its three-layer platform: Compute, Serverless, and Model API. H3 Max generates a 5-second video in under 3 seconds and ranked first in quality, prompt understanding, and aesthetics in fal's human preference evaluation; independent benchmarks from Artificial Analysis and Design Arena also ranked it first.

    Why it matters: fal officially breaks down the three-layer platform that takes H3 Max from training to launch, offering a look at the inference and scaling mechanisms behind low-latency video generation.