{"rewrite":{"id":"r_7b236a547e49cdd4b0d149d6","clusterId":"c_82b9a9e8143a9bbcb2340263","slug":"minimax-h3-video-generation-cut-to-31-6-seconds-on-a-single-rtx-5090","model":"deepseek-v4-1-flash","headline":"MiniMax H3 Video Generation Cut To 31.6 Seconds On A Single RTX 5090","summary":"A first-person report on GameBusiness.jp timed an RTX 5090 running MiniMax H3 locally. One 5-second i2v clip went from 193.5 seconds to 31.6 seconds in 41 days, a 6.12x gain, using a 20-step baseline, the official Turbo 8-step LoRA, Hao AI Lab's FastH3 4-step, a community-converted TaoMate 3-step LoRA, INT8 attention, and sparse attention. H3 Max weights have not been released 28 days after fal said on X that it would publish them. Cloud H3 Max Turbo ran at 1.49 seconds on different hardware.","whyItMatters":"If a single user can reach 6.12x on a 19.5GB open model at home, the argument for waiting on the unreleased H3 Max weights tilts toward prompt adherence rather than speed.","webCardHtml":"\u003cp\u003eThe baseline is 1280x704, 5.2 seconds, 124 frames, same first image, same prompt, same three seeds, i2v, four runs per condition with the first discarded and the median of three kept. Each LoRA arrived on a different date, which is why the gains landed in bursts rather than a curve.\u003c/p\u003e\u003cp\u003eThree step-plus-attention combinations carried most of it: the official Turbo 8-step LoRA reached 90.7 seconds, FastH3 4-step reached 55.3 seconds, and a community ComfyUI conversion of a TaoMate 3-step LoRA reached 44.1 seconds. Dropping INT8 attention from the last of those fell back to 59.0 seconds, so the quantization alone was worth 1.34x. Sparse attention at 0.3 then 0.1 brought the clip to 35.0 and 31.6 seconds.\u003c/p\u003e\u003cp\u003eThe next bottleneck is VAE decoding, at 39 percent of total time. ComfyUI 0.36 shipped a new VAE; tested in the production setup, it came out even. A 30-clip music video moved from 1 hour 37 minutes to 16 minutes.\u003c/p\u003e","blueskyPost":"MiniMax H3's local speedup comes from stacking a Turbo LoRA, a FastH3 4-step, a community TaoMate 3-step, INT8 and sparse attention. H3 Max weights still have not appeared 28 days after fal said they would.","twitterPost":"MiniMax H3 hit 31.6 seconds on one RTX 5090, but H3 Max weights remain unpublished 28 days after fal said they were coming.","threadsPost":"MiniMax H3 went from 193.5 to 31.6 seconds on a single RTX 5090 in 41 days, and the gains come from stacking community tools like a TaoMate 3-step LoRA and INT8 attention. H3 Max weights have not been released 28 days after fal said they would publish them.","newsletterBlurb":"A single RTX 5090 cut MiniMax H3 generation from 193.5 seconds to 31.6 seconds for a 5-second clip, a 6.12x gain over 41 days, built from the official Turbo LoRA, Hao AI Lab's FastH3, a community ComfyUI conversion of a TaoMate 3-step LoRA, INT8 attention, and sparse attention. VAE decoding is now the largest remaining cost at 39 percent, and a new ComfyUI 0.36 VAE came out even in the production setup. H3 Max weights remain unreleased 28 days after fal said it would publish them.","attributionJson":"[{\"source\":\"GameBusiness.jp\",\"url\":\"https://www.gamebusiness.jp/article/2026/09/24/28128.html\",\"title\":\"MiniMax H3 Video Generation Optimized To 40 Days, Is H3 Max Weight Release Unnecessary (CloseBox)\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":6354,"outputTokens":916,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1791214632,"createdAt":"2026-10-05T15:32:35.000Z","publishedAt":"2026-10-05T15:36:06.000Z","updatedAt":"2026-10-05T15:36:06.000Z"},"cluster":{"id":"c_82b9a9e8143a9bbcb2340263","canonicalTitle":"最高水準の動画生成AI「MiniMax H3」高速化を40日で極めた。本命「H3 Max」のウェイト公開はもう不要なのか？（CloseBox）","representativeArticleId":"a_213c2dd3ec12424e780ac55e","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"MiniMax H3\",\"H3 Max\",\"Seedance 2.5\"],\"studios\":[\"MiniMax\"],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-09-24T07:27:56.000Z","lastSeenAt":"2026-09-24T07:27:56.000Z","updatedAt":"2026-10-05T15:36:06.000Z"},"attribution":[{"source":"GameBusiness.jp","url":"https://www.gamebusiness.jp/article/2026/09/24/28128.html","title":"最高水準の動画生成AI「MiniMax H3」高速化を40日で極めた。本命「H3 Max」のウェイト公開はもう不要なのか？（CloseBox）"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["MiniMax H3","H3 Max","Seedance 2.5"],"studios":["MiniMax"],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":null}