從真實影片案例學習 MiniMax H3 提示詞

Oddar · @don56palmer

An original comic page becomes a voiced Japanese joke about monsters buying an energy drink.

CALMDUST aka ちゃんこあ · @core_tan

A WebUI example stacks character images into one video reference and combines character and audio RefMod with FlashGen.

Mike Pevzner · @Mike_Pevzner

A creator shares a family-involved Recast demonstration, linked to fal’s H3 Max character replacement announcement.

M7［mi7］AI · @mi7_crypto

A five-part game promo reconstructs a ruined comic city into CG, flowers and a cyber forest through scanlines and spatial slices.

kuripon · @malonpon

H3 Max uses an AI-generated 3D hand-dance previs and character sheet to test face fidelity at awkward camera angles.

NOBU · @nbykos

An H3 Max versus H3 example reports unexpected cuts despite identical references and a single-shot instruction.

🟢DeepForest_Studio_AI🟢 · @DeepForest_Inc

A side-by-side test compares INT8 ConvRot, Standard and LynnReal Light VAE processing after the same H3 sampler conditions.

shi3z · @shi3z

A local H3 demonstration on DGX Spark reports roughly twice the speed using training-free Layer Adaptive SLA.

Kamimoto(かみもと) · @sep_is_heim

An RTX 4070 I2V example reports 124-frame TaoMate H3 output in 2:26, with a follow-up publishing its ComfyUI optimizations.

葦沢かもめ · @AshizawaKamome

A three-panel test compares H3 dance footage, motion applied in Blender and an overlay of both results.

Glenn Williams · @GlennHasABeard

A locked bottle-height camera holds on a bubble growing to watermelon size before it detaches and floats upward.

ぽむ · @pompompokopomch

A dark phone-loop music video explores clay and gummy textures with chant, Jersey Club and heavy 808 drone styling.

Alvaro Somoza · @OzzyGT

Two short game-clip examples use MiniMax H3 to generate sound effects from visual references.

Greg Horvay · @greg_horvay

An Intel B70 H3 workflow reports a 1344×768, 107-frame Ref2VA clip dropping from 42 minutes to about nine.

Sparkful · @SparkfulArt

A vampire scene compares a six-second H3 Max clip with a seven-second Seedance 2 clip for visuals and spoken dialogue.

The Rogue AI · @TheRogueA1

A creator uses MiniMax H3 inpainting to insert themself into a recognizable Ripley scene from Alien.

さいり🥚AI Creator · @sairin_hamumu

A semi-automated TapNow workflow aims to remake H3 lyric music videos by replacing the lyrics without planning every cut.

Shami · @ShamiWeb3

A wolf splashes a rabbit’s tiny convertible, setting up a larger and wetter revenge gag made with H3 on TapNow.

Bilal · @deepwhitman

A piano-led dance piece explores movement and emotional expression using H3 Max and Max Turbo with Opus direction.

Kent Dhani · @kentdhani

A local Ref2V tracking shot keeps one airport arrival in focus amid bodyguards, fans, paparazzi and flashing cameras.

Soran · @Soranlan

A 15-second POV sword-flight prompt tests continuous physical motion; the author reports poor transfer from Seedance-style prompting.

Xanderwow · @Xanderwow_A

A widely viewed H3 counterattack short from Xanderwow demonstrates a reversal-based action premise.

T · @tkvtk

A Japanese H3 study tests dolly zoom, rack focus, golden-hour lighting and swirling ink across realistic and anime footage.

Ima Studio · @ImaStudio_ai

Ima Studio compares H3 and Seedance 2.0 Mini on one 15-second aircraft story with shared characters and prompts.

Jonathan Fischoff · @jfischoff

An H3 Max Insert Video demonstration presents the author’s video-editing result in a longer source clip.

혼나고랩 · @honnago

A local H3 Max recreation adapts a published Korean campus DV-vlog prompt and includes Korean speech on dual RTX 3090s.

Inner-Reflections · @InnerRefle11312

A local H3 demonstration explores temporal upscaling as a way to improve generated video results.

ゆずき · @uzuki425

A Japanese H3 experiment shares a memory-optimization table and contrasts optimized execution with slower low-VRAM offloading.

Kamimoto(かみもと) · @sep_is_heim

A 12GB RTX 4070 benchmark library compares Ref2VA and FL2VA outputs and provides downloadable generation workflows.

ρŁ𝐀𝔰Ｍʘ · @plasm0

A wordless synesthesia music video combines local H3, Seedance 2.5 and Suno v6 to explore sound as color.