從真實影片案例學習 MiniMax H3 提示詞

伊藤大河 · @itotaiga_

A 3060 Ti 8GB completes a 1280×704, 15-second H3 clip in about 40 minutes, with 16GB host RAM disrupting other work.

BittensorMan@AIエンジニア · @xb_bittensor

A company reports RTX 3090 H3 production cuts assembled from two eight-second segments for a product promo pipeline.

えーてぃーえふ · @ATFzwei

A Hack Studio user reports roughly 3.5 minutes for five seconds and 7.5 minutes for ten seconds with ComfyUI in StabilityMatrix.

Risa · @kantakanta1233

A creator generates a Halloween character video with H3 through the Topview plugin connected to ChatGPT.

Hao AI Lab · @haoailab

The FastH3 team shows local H3-derived clips and reports eight-step V2/Trim timings across RTX GPUs and DGX Spark.

Kamimoto(かみもと) · @sep_is_heim

A three-way RTX 4070 test reports 61, 78 and 140 seconds while changing SLA/Veda, step count and low-resolution X2 VAE setup.

麥客不停 · @smardio

A creator credits Dot with following a local H3 music-video task through completion while generation runs slowly.

Joe Sparks · @joesparks

An RTX 5090 user shares a ComfyUI FastH3 first/last-frame example and finds its quality weaker than regular H3.

Jennie X · @zennie17x

A Korean history video uses Claude Code automation, Qwen images, H3 animation, Typecast narration, Remotion and FFmpeg.

Toshi🎬AI動画💫MV制作🎧 · @Toshi_nyaruo_AI

A matched RTX 3080 test reduces peak VRAM from 9.6 to 7.3GB with Trim, but total runtime rises from 138 to 162 seconds.

Nami｜EC運営×生成AI｜商品ページ自動化の実験ログ · @NamiDesignWorks

An RTX 3090 creator finds actual 20-step runs behind a saved four-step LoRA node and compares standard versus reference timing.

リド · @ohshie1973

An author reports 480p, 15-second, four-step local runs under three minutes with a custom GUI and explicit settings.

あぼん · @chujoakira

A creator reports roughly two-minute H3 videos in twenty minutes, with prompt tuning and audio limitations.

烁皓 · @eternityspring

Matched 640×640 eight-second runs report 58.23 seconds for FastH3 V2 and 255.92 seconds for official H3 INT8.

keys 🧪 · @u1tra_instinct

A TensorFold Apple Silicon update reports a 1312×736, 243-frame H3/FastH3 V2 run in 7.9 minutes.

Matt · @MrCollison

A game prototype reports 117 animations from GPT Image first frames, H3 Max video and GHVMR motion conversion.

Chloe Kao · @ChloeXChaCha

A ten-episode anime build log explains last-frame chaining and human review after fourteen shots failed continuity.

A. Sparks · @asparks1010100

A user reports using a RunPod MCP integration to provision an H3 server, execute prompts and download outputs.

Ankit Kumar · @AnkitKumar65837

An RTX 5070 Ti user with 24GB system RAM reports H3 runtimes of 7, 19 and 38 minutes for 5, 10 and 15 seconds.

Shoji Kai · @Shoji_ai

A creator uses four GPT Image stills as consecutive start/end pairs for five-second H3 stages, then joins the clips.

ITOK · @ITOKtw

A local DMAD four-step test reports BSA tau, LoRA strength and visible motion and color tradeoffs.

Alexey Fateev · @superalesha

An overnight local batch chains Qwen Image 2.1, fifteen-second H3 audio clips and H3 upscaling on one RTX 3090.

ɹǝʍoqɐuı · @public_inabower

A technical tutorial combines OpenFOAM and ParaView simulation visuals with H3 video and VOICEBOX narration.

Kamimoto(かみもと) · @sep_is_heim

An author shares a modified DMAD recipe targeting about one-minute local generation on an RTX 4070.

鎌原淳三(Junzo KAMAHARA) · @jun3k

An app creator introduces H3cApp for running local H3 video generation on a Mac with 24 GB memory.

Pet Riot · @TwokgoodTangwei

A FreeVideo user reports generating a 25-second local H3 clip in twenty minutes with 8 GB VRAM.

AI Bard Guild · @IsekaiBardGuild

A ComfyUI node extracts viewpoints from an H3 orbit video, estimates poses and depth, then corrects coordinates for Qwen pose control.

87Boss AI 工作流實驗室 · @87BossAI

An H3 workflow combines a Selflift second pass with a 2× VAE to produce standard and enlarged versions in one pipeline.

maharshi · @maharshii

A creator reports using H3 Max to make fifteen-second explainer videos for over 300 results in a mathematical catalogue.

gokaygokay · @gokayfem

H3 Max fifteen-second repository explainers sit in a navigable 3D map of 1,000 GitHub projects clustered by shared topics.