從真實影片案例學習 MiniMax H3 提示詞

WTR · @wtry1102

A director plan feeds 23 reference-conditioned H3 shots on Colab L4, then tracked lyrics are composited over the edit.

DeCat · @DeCat2025

A black-and-white anime fight allows saturated color only at the instant of a physical strike.

Alex Shev · @AlexshevPm

A ComfyUI workflow passes ten seconds of video and audio context into subsequent sections before latent upscaling.

Kamimoto(かみもと) · @sep_is_heim

A reference-conditioned H3 test uses a 16-bit pixel-art LoRA to explore game-like motion and style.

中村剛 · @cavinaka

H3 footage for a TOWA TEI music video is finished in a custom editor that maps graphic movements to individual sounds.

kays · @pabloadaw

A creator-trained orbit LoRA uses first and last frames for consistent full-circle camera moves and loops.

Pravin Uttarwar · @uttarwarpravin

A 77-second cartoon advertisement pairs open-weight H3 video generation with Claude Code orchestration.

Leo Kane · @aiworkflowprolk

A 15-second H3 clip with dialogue and sound is generated on Colab and then upscaled from 768p to 1080p.

akatz · @akatz_ai

The adapter creator tests a 1,000-step character-swap LoRA, preserving source motion and audio while documenting drift.

Alperen · @AlperenKonukbay

A guessing game asks viewers to identify where real footage ends and H3 Max continuation begins.

Wildminder · @wildmindai

An H3 style LoRA explores retro science fiction with cyborgs, laser guns and synthesizer sound.

汤光头 · @tangpanqing

A modern Shanghai family-drama prompt specifies three characters, spoken Mandarin dialogue and no subtitles.

CALMDUST aka ちゃんこあ · @core_tan

A WebUI test combines RefMod with three-step Ref2VA generation at 768p on an A100 80 GB.

SmallKino · @SmallKino_

A comparison changes reference-video size to explore its influence on VFX transfer and character output.

Stable Diffusion Tutorials · @SD_Tutorial

A work-in-progress reference adapter compresses image or video references into small latents without training.

鈴木憂一 | Highdrama · @yu_ichi_suzuki

An experiment applies TaoMate by block group and resumes a three-step preview to an eight-step render.

Wildminder · @wildmindai

An H3 LoRA produces full-sphere equirectangular video while retaining native audio.

いのり · @lnkiai

A lighthouse witch short uses H3 for visuals and sound effects, with MiniMax Music 3.0 for the score.

Rhino Quant · @RhinoQuant

A five-second magic-and-rubber-duck clip tests local H3 video and audio through h3-apple on an M5 Max.

Refiner ✦ · @Refiner_Studio

A portrait pizza commercial combines macro food shots, floating ingredients, kinetic typography and a cheese-pull payoff.

antic · @antic

A creator describes a Suno-to-H3 music-video pipeline with mood boards, human review and targeted final-segment rerendering.

Solutions Joe · @SolutionsJoeG

A 40-second restaurant promo turns food and terrace photos into H3 footage, then adds tracked graphics and a sound mix.

石ちゃん🎤LLMでアプリ開発 · @Isichan_Hitori

A local ComfyUI experiment removes token_refiner to test higher-resolution H3 generation within a fixed VRAM budget.

Michael Guo · @Michaelzsguo

An engraved newsprint world transforms into color in a music-video pipeline using local H3 and beat-aligned editing.

山原 慎也（しんやん）｜AIプラットフォーム運営 · @AI_RESKILL

A stacked comparison reuses one motion-graphics prompt across H3 Max and Claude, with audio from H3.

Stan_Katayama · @Dokko57932483

A widely shared Rogue fan-character video, attributed by its creator to MiniMax H3.

近添真琴（ライター・AIクリエイター） · @chikazoemakoto

A Japanese pool-cleaning gag combines a sudden splash with AI-generated video and audio.

にも｜AIによって爆誕 · @ai_ai_ailover

A mixed Claude, HyperFrames and Floyo H3 workflow explores changing angles in a motion-graphics video.

SmallKino · @SmallKino_

A side-by-side character-swap test compares LoRA strength 1.0 on the left against native H3 on the right.

とすくん · @tokyo_Valentine

A four-shot anime comedy turns an energy beam into a rainbow selfie, with timed Japanese speech and sound cues.