Tech2Wild · @Tech2Wild
Four Spark LLM and ComfyUI Memory Sharing
A creator cohosts a TP4 DeepSeek model and four Qwen Image/H3 ComfyUI studios across four DGX Sparks.
經查核的提示詞與原始出處
對照創作者公開的提示詞與完成的 MiniMax H3 影片,了解場景安排、動作、運鏡、連貫性與聲音如何搭配。
每個案例都附有原始出處連結。你可以直接複製公開的提示詞,也能保留其結構,改寫成自己的場景。
753 個經查核案例
依運鏡技巧、分鏡、動畫、參考控制、音樂、對白或動態圖像瀏覽案例。已公開提示詞與僅介紹工作流程的案例皆有清楚標示。
找到 753 個案例 · 顯示第 721–750 筆
Tech2Wild · @Tech2Wild
A creator cohosts a TP4 DeepSeek model and four Qwen Image/H3 ComfyUI studios across four DGX Sparks.
久田誠良(Seira Hisada) · @SeiraHisada
A three-minute-plus memorial film combines native Ref2VA at eight steps with Suno music and Python/FFmpeg editing.
VARIEN · @varien
An anime dialogue sample separates FastH3 OmniRef video, SeedVR2 finishing, Fish Audio Drama 3 and Whisper subtitles.
Donnie · @dkfiander
Two fifteen-second Mars flybys compare Wan from a Flux still with FastH3 from a new still on an RTX PRO 6000.
FoldingEnd A.I. · @EndFolding79421
A matched-seed fifteen-second I2V test reports BF16 taking 39m08s versus Singularity INT8 at 27m49s, without LoRAs.
PixelDojo · @PixelDojoAI
A portrait Turbo scene animates a paper boat, rainfall, neon reflections and a raincoat-clad observer with one slow push-in.
Tony | World Model Lab · @tony_worldmodel
An I2V red-ball test asks for rise, apex and fall, while reserving quantitative physics claims until image tracking is complete.
Phantom|錬金学総論 · @Phantom_202601
A creator uses frame feedback, scoped prompt edits and same-seed H3 reruns to correct composition, flags, extra balls and foreground trees.
Zhanke Wang · @zkwang07
An agent reconstructs a scene, extracts FlowHMR motion, retargets and renders it, then uses H3 for photorealistic refinement.
BennyDaBall · @BennyDaBall_OG
A five-second diner study compares stock H3 and acceleration stacks using warm-run medians, with separate motion and scoring caveats.
Kujirachan/くじら · @balaena01
A Halloween dance composites a VRM character onto six H3 street shots, using road lines for camera recovery and Blender shadows.
EchoxStudio · @EchoxStudio
A music-video workflow rebuilds shot images and H3 clips for widescreen while retaining the vertical version’s song edit timing.
VIDEO REBIRTH · @video_rebirth
HyperFlow Infinity generates a shot, reviews it, updates scene memory and directs the next, demonstrated with an hour-long Alice film.
あいり🤖AI実用 · @airi_ai_navi
A local FastH3 pixel-art comparison restores the grid, limits the palette to 32 colors and reduces playback from 24 to 12 fps.
さいり🥚AI Creator · @sairin_hamumu
An anime maker prepares Irodori-TTS dialogue for H3 lip sync, then uses ducking and J-cuts to carry speech across visual inserts.
Rehan Sheikh · @rehan_shei
Anymove uses H3 Max front-and-side motion video and a fine-tuned DINOv2 tracker to animate nonhuman rigged assets.
BennyDaBall · @BennyDaBall_OG
A frozen-latent H3 study compares decoder-only warm medians, visual differences and memory costs across stock, Sol and LightVAE.
Gabi C. Duncombe · @gcduncombe
A looping film regenerates each played shot from the same prompt and character references while preserving a visible history of old takes.
aicreataro · @aicreataro
An H3 and Python experiment turns black letterbox bars into surfaces that a character crosses, lands on, grinds along and breaks.
Phantom|錬金学総論 · @Phantom_202601
A three-second Tessera test completes after moving the text encoder to a second GPU, with smoother texture but altered face and hands.
鈴木憂一 | Highdrama · @yu_ichi_suzuki
A four-decoder H3 anime comparison weighs flat cel shading and cleaner lines against texture retention and decode time.
Denoche · @liangkun43
A 768p-to-2K H3 test uses original-frame AddGuide conditioning and an explicit three-step sigma schedule.
二五六月ユウ · @256tsukiyuu
A conversational character app feeds live game captures into H3 references to display gameplay on a television inside its animated scene.
絵空事テレビ / AIカワウソ · @boke_ai
A map-based Blender street and shopfront images guide H3 live-action conversion while testing Japanese sign retention.
cocktail peanut · @cocktailpeanut
A crude GPT Image GIF supplies camera and action timing for a 480p four-step H3 Singularity reference-video finish.
DEXTUNE · @dextune
CrossView pairs the source clip with a depth-warp guide, using prompt text to fill unseen regions at a new camera angle.
BennyDaBall · @BennyDaBall_OG
Four matched H3 drafts compare 2K regeneration, IC guidance, LTX Restore and ReCaVSR by whole-job time, identity and sharpness.
KireiStudios · @KireiStudios
Two USB4-linked Strix Halo nodes run H3 with reported bit-identical output while reducing a 5.2-second test from 6.8 to 4.9 minutes.
aratama 璞 · @aratamadao
A labeled 3DGS comparison uses two 13.44-second H3 clips and local retiming to restore beat-aligned camera events.
ピーター · @_petert
A before-and-after H3 example uses Meta SAM3 Audio to remove unintended background music during postproduction.
使用合集
將每個案例視為可重複運用的結構。保留控制動作與連貫性的部分,再替換原始場景中的創意細節。
瀏覽分類,或依主體、運鏡技巧、創作者與模型搜尋,選擇動作及結構符合預期影片的案例。
若有公開文字,請先確認是完整提示詞還是片段再複製。沒有提示詞的案例,可參考原始出處與改寫建議。
換成自己的主體與場景,提供所需的參考素材,並在生成前檢查所有限制條件。
複製經查核的案例、調整創意細節,使用 MiniMax H3 生成自己的影片。