gordonlu
- 40 skills
- 0 followers
- 11 hours ago last updated
- ▌ H3 Promo Film · gordonlu bundleWrite runnable MiniMax H3 promo & trailer prompts with no reference assets at all — food promos, product campaigns, brand teasers, cinematic trailers. Use when the user wants a publicity or marketing short generated purely from text. Produces timecoded beats per base-en.txt; requests with reference images or videos are redirected. Produces integrated_multimodal_description with timecoded beats, overall_soundscape, and non_diegetic_music following the official base-en.txt guide. Requests that bring reference images or videos are redirected to the image-to-video or motion-transfer skills.
- ▌ Suv Commercial · gordonlu bundleGenerate a top-tier, ultra-premium future luxury SUV commercial — elegant, restrained, real, like an international luxury car brand spot — as one continuous 15s 16:9 24fps film from a single vehicle reference image. Use when the user wants a luxury car / SUV brand film, a showroom or Brutalist design-studio ad, a cinematic car-launch commercial, or a "car ad" that must feel expensive and calm rather than like a tech demo or racing game. Locks the vehicle to the reference image (no redesign, no second interior), enforces LEFT-HAND DRIVE consistency in every interior shot, restricts the palette to a luxury night showroom look, allows only five typography lines with soft-fade motion, and uses NO BGM — pure cinematic SFX only. Original prompt by Flkrstudio (source below); this Skill packages that prompt with reference-locking rules and a verified 15s output.
- ▌ H3 Keyframe Film · gordonlu bundleWrite runnable MiniMax H3 keyframe-film prompts — multi-keyframe montages, first-last frame transitions (FL2VA), final-frame landings (L2VA), and I2VA first-frame animation. Use when the user has reference images (keyframes, character art, product shots, final poses) and wants the image identity locked while text drives the motion. Follows base-en.txt exactly, including the keyframe-alignment instruction lines. Pure-text requests are redirected to h3-promo-film; reference-video motion goes to reference-motion-transfer.
- ▌ Clay Cafe Stop Motion · gordonlu bundleGenerate a 15s cozy claymation café short in handcrafted stop-motion style. Use when the user wants a warm, cute, tactile paper-clay animation with soft natural lighting and delicate macro cinematography. Includes a verified 15s output.
- ▌ Jet Dogfight Generator · gordonlu bundleCreate cinematic 7-second MiniMax H3 text-to-video clips of high-speed jet dogfights above and through cloud decks. Full real-time speed, contrail arcs and a clean high-speed merge under a locked camera; no whip or orbit moves, no vapor cones. No slow motion, no external footage feel — one continuous aerial shot with a fixed lens.
- ▌ Jk Shanghai Frozen Run · gordonlu bundleGenerate a 15s 16:9 mixed-media one-shot: a 2D anime JK schoolgirl runs through photorealistic Shanghai, trips, and time freezes as the camera orbits her in bullet-time. Use when the user wants a mixed-media (anime character × real environment) continuous-shot short with a frozen-time orbit. I2VA with one character reference (image only, not first frame). Includes a verified 15s output.
- ▌
- ▌ H3 Performance Director · gordonlu bundleCompanion skill for MiniMax H3 5–15 second video generation. Directs the subject itself rather than the camera: performance intent, visible behavior, movement quality, body mechanics, weight, timing, force, anticipation, follow-through, recovery, silhouette readability, and subject-type-specific motion grammar. Designed to work with h3-micro-cinematic-director and h3-shot-pattern-library. Supports both text-directed performance and reference-motion transfer workflows.
- ▌ H3 Shot Pattern Library · gordonlu bundleCompanion skill for h3-micro-cinematic-director. Provides a reusable shot-pattern library for MiniMax H3 5–15 second clips, including pattern selection, H3 reliability tiers, duration compatibility, T2V/I2V/FL2V guidance, composition arcs, spatial choreography, and short prompt examples. Use this skill together with the base h3-micro-cinematic-director skill when a clip needs stronger camera language, more deliberate spatial design, or a proven shot skeleton instead of inventing camera movement from scratch.
- ▌ Cavalry Charge Generator · gordonlu bundleCreate cinematic 7-second MiniMax H3 text-to-video clips of full-speed cavalry charges across dust-scoured plains. Every beat runs at natural real-time speed or faster — no slow motion; the camera presses low through the dust wall alongside the charge. Dust, armor, sand-spray and low golden backlight carry the epic texture. T2VA only, no reference material needed.
- ▌ H3 Reality Physics Guard · gordonlu bundlePre-director gate for MiniMax H3 5–15 second generation. Runs BEFORE any camera, blocking or performance design: classifies Reality Mode, verifies known facts about real subjects, runs an eight-category reality check, resolves contradictions against a fixed priority ladder, and hands the director a Reality Sheet of allowed / forbidden / must-verify constraints. Catches physics, anatomy, medium, light and temporal bugs — belly-first landings, floating cavalry, sliding aircraft, morphing geometry, teleporting clouds — before credits are spent. Does not write prompts and does not direct; it only guards.
- ▌ Reference Motion Transfer · gordonlu bundleTransfer the motion from a user-supplied reference video onto a new target subject with MiniMax H3. Use when the user wants a character, mascot, product, or creature to perform the dance, gesture, action, locomotion, or camera move they saw in another video. Covers Ref2VA motion transfer (primary route), plus I2VA / FL2VA / L2VA fallbacks when no usable reference video exists. Pure-text requests have nothing to transfer from and are redirected to the base `h3-prompt-writing` skill. Not for style-only transfer with no motion, direct editing of the source video, real-person impersonation without consent, or sequences longer than 15 seconds without segmentation.
- ▌ Cinematic Key Art Animator · gordonlu bundleTurn one high-quality character illustration, concept art image, fantasy portrait, game key art, anime-style character image, or original mecha artwork into a polished 7–8 second MiniMax H3 moving key-art clip. Automatically route between atmospheric Living Key Art and high-density Action Burst. Preserve identity and art direction while allowing meaningful pose changes, multi-joint action, fast short-form pacing, and multiple escalating peaks when the subject is action-oriented.
- ▌ Link Skills With Base Loop · gordonlu bundleGenerate a magazine-grade character showcase sheet from a single character reference image — a hero key visual plus detail grid, three-view turnaround, palette board, and nameplate — held as a near-static 4s 16:9 shot with only the slightest living motion (fabric, hair, ink edges). Use when the user wants a character setting / showcase sheet, a "design this character's reference board", a character-introduction card for a game or series, or a prompt-built "skills with a base loop" still showcase. Locks the character's face, body, proportions, hairstyle, skin tone and outfit to the reference; no second character, no camera movement, no large motion.
- ▌ H3 Micro Cinematic Director · gordonlu bundleA short-form cinematic direction skill designed specifically for MiniMax H3's 5–15 second generation window. It converts a rough video idea or reference image into one continuous cinematic event with deliberate blocking, spatial geometry, composition arc, subject choreography, camera behavior, depth design, visual peak, and a final H3-ready prompt. Use when H3 generations feel static, lack camera language, have weak action choreography, meaningless camera movement, poor spatial readability, or no clear visual climax.
- ▌ Redirect Mass With Leverage · gordonlu bundleGenerate a grounded, hyper-realistic dark-fantasy tavern brawl in a single take — an elf ranger uses speed, footwork and leverage to take down a massive half-orc bruiser in a crowded seaside tavern, from a two-character reference image. Use when the user wants a physical fight scene, a bar/tavern brawl, a David-vs-Goliath melee, or any cinematic hand-to-hand sequence that must feel real (believable momentum, real weight, grounded handheld camera) — never floaty, never anime-exaggerated, never superhuman. Locks both characters' look to the two-character reference image; keeps real physical momentum; no flying, no slow-mo, no cartoon.
- ▌ Stage Deadpan Chain Failure · gordonlu bundleGenerate a 10s 16:9 hyper-real cinematic absurdist comedy short in a luxury circus big-top, starring ONE male ostrich as the only character — a dominant runway strut that escalates into controlled chaos (fanning a spotlight sideways, knocking a mannequin off stage) and ends in a deadpan "shed-the-cape mic-drop" that breaks the audience. Use when the user wants a single-joke runway-comedy scene, a one-character absurdist stage bit, a "confident model who wrecks the show without reacting" comedy, or a circus/ostrich gag — T2VA (pure text generation, no reference image). Locks the ostrich's proportions, style and features consistent throughout; only one ostrich, one tent, one main joke.
- ▌ Create Cross Medium Reaction · gordonlu bundleGenerate a cross-medium comedy short — live-action night-market stir-fry stall footage and one perfectly flat 2D sticker character (the sticker Shiba Inu 旺财) sharing the frame and reacting to each other — as a 10s 16:9 24fps single continuous POV shot from the vendor's slightly-downward wok-tossing eye level. Use when the user wants a flat 2D sticker character interacting with real footage, a "sticker steals the chili / gets bonked / gets spicy-fired" gag, a cross-medium mixed-media clip, or a first-person wok-cooking comedy. T2VA (text-to-video) only, no reference image. Hard rules: only one live-action hand may appear at a time and never two hands in the same frame; the sticker dog stays perfectly flat 2D with no 3D volume, no realistic hair, shadow or clay texture; left hand never touches the spatula, right hand never holds the jar; no dialogue.
- ▌ Kinetic Title Card Generator · gordonlu bundleCreate a 7–8 second MiniMax H3 text-to-video clip where a short title, brand name, or slogans are the hero: a title card / text animation built on text-driven motion. Use for brand-name opens, film/series title cards, song-title cards, event posters, or any request whose main subject is displayed words. The text must be short (1–6 characters per line), placed once with explicit position, timing, font flavor, and entrance behavior; never let it morph, relocate, or regrow. Different from music-video-subtitle-generator (lyric karaoke) and promotional films with incidental text (h3-promo-film): here the words ARE the video. Follows base-en.txt exactly. Reference images or videos route to h3-keyframe-film / reference-motion-transfer.
- ▌ Mythic Cloud Whale Generator · gordonlu bundleCreate cinematic 7-second MiniMax H3 text-to-video clips of colossal mythological beings crossing an endless sea of clouds. One legendary subject (cloud whale, sky leviathan, cloud dragon…) is the primary motion owner; the clip is built on a fast-slow contrast rhythm — explosive break-through actions and real-time speed beats, never an all-slow-motion piece — with volumetric light, drifting dust and layered mist for an epic, high-texture look. T2VA only, no reference material needed.
- ▌ Stage Solo Posture Comedy Arc · gordonlu bundleGenerate a 12s 16:9 single-person talking-head comedy short: a young man alone on a sofa facing camera, casually riffing about fitness and telling 2–3 self-deprecating jokes in a deadpan rambling tone. Use when the user wants a "talk about X, but make jokes" solo comedy spot, a deadpan self-deprecating monologue, or a low-key talking-head bit that reads as natural casual chat — T2VA (pure text generation, no reference image). Hard rules: homey lighting, natural performance, stable camera with slight push-ins at each punchline, only this one character, no subtitles, no BGM, no canned laughter, no reality-show exaggeration.
- ▌ Haute Couture Atelier Ad Style · gordonlu bundleGenerate a top-tier Western haute-couture and high-atelier ad in a dark luxury space — a large Chinese high-embroidery piece (ruyi cloud, twining-lotus, koi and scroll motifs on cinnabar-red / ink-blue / jet-black silk) repeatedly close-up in the lens language of Dior Haute Couture, Chanel Métiers d'Art, Valentino Haute Couture and Hermès craft films — as a 15s 16:9 24fps short: five-color thread sheen, ordered fine stitching, gold couching, layered embroidery, warm-gold key light with soft rim light and macro highlights, dramatic chiaroscuro on dark grounds, slow pushes and lateral moves with soft dissolves, no narration. Use when the user wants a luxury fashion / craft atelier brand film, an embroidery or heritage-craft hero spot, a "haute couture" product showcase, or an expensive, restrained, elegant and mysterious craftsmanship ad. T2VA (text-to-video) only, no reference image. Hard rules: keep the same single embroidery piece throughout (never swap motifs), one restrained bilingual title only at the end
- ▌ Structure Product Proof Launch · gordonlu bundleGenerate a flagship product-launch ad with an escalating 9-hard-cut structure — deep navy-black, electric-cyan, incandescent-white text and glowing particles on every frame, branded with h3skills.com / H3SKILLS — from a single style/composition reference image. Use when the user wants a high-energy product / platform launch commercial, a "launching now" brand film, a tech-flagship style promo, or any launch film that must feel expensive, fast and futuristic (never cheap, messy or retro). Locks the visual language (navy-black base, electric-cyan, white text, glowing particles) to the reference; allows only the listed brand strings; no blurry/semi-transparent/scaling text, no watermarks, no lens flares.
- ▌ Use Material As Progress Clock · gordonlu bundleGenerate a heritage-craft documentary ad in which a material's appearance itself is the clock of the film — a 15s 9:16 vertical short following a Miao (Hmong) silversmith forging an ornament through eight steps, where the "silver-white" in the frame goes from zero to everywhere as time passes. Use when the user wants an intangible-heritage craft ad, a slow elegant process documentary for a handmade product, a "from nothing to finished object" brand film, or a craft piece where the material's transformation drives the narrative. Locks the final silver piece to a single reference image (no added dragons/phoenixes/buddhas/text), keeps real-metal lighting (no glow VFX), and uses subtitle-card-driven storytelling with on-site sound only (no music, no narration).
- ▌ Alternate Obstacles Toward Goal · gordonlu bundleGenerate a one-take realistic disaster-action film — a female freight-ship engineer sprinting alone through a narrow reactor corridor toward an emergency seal console while crimson fuel-fire waves erupt from floor grates and severed pipes burst into sparks — as a continuous 15s 16:9 24fps shot with fast, clear, non-blurry action. Use when the user wants a cinematic disaster / escape scene, an industrial sci-fi corridor run, a "one goal, one storyline" single-take chase, or a fire-wall escape sequence. T2VA (text-to-video) only, no reference image. Hard rules: one continuous shot with no cuts, a single main character, no slow motion, no facial close-ups, no text/logo, no cartoon look — and the engineer must always stay inside the blast-door safe zone after the door seals (never outside the door, never re-entering from outside).
- ▌
- ▌ Verify Recurring Identity Board · gordonlu bundleGenerate a 15s 16:9 24fps cinematic anime fashion short that keeps ONE character locked to a reference image across the entire clip: a blonde female pilot beside an emerald vintage sports car on a California coast highway at sunset. Use when the user wants a character-consistent showcase — same face, body, outfit and proportions throughout, no second person, no costume/face change — a high-end anime PV + retro travel-fashion + road-movie look. I2VA with one character reference (image only, not first frame). Includes a verified 15.08s output.
- ▌ Direct Imperfect Memory Farewell · gordonlu bundleGenerate a hyper-realistic documentary-style short of a forgotten 1998 MiniDV home video — a 20-year-old Japanese woman waiting at a quiet seaside Kamakura station, watching the sea, and waving goodbye before boarding — as a 15s 16:9 24fps clip wearing full 90s Sony MiniDV camcorder artifacts: violent handheld shake, off-frame compositions, auto-focus hunting between the train and her face, exposure riding on the daylight, faded color, soft contrast, slight motion blur, MiniDV compression, no stabilization. Use when the user wants a "found footage / old camcorder memory" film, a nostalgic farewell at a train station, an imperfect real-family-video look, or a memory flashback with an authentic period camera texture. T2VA (text-to-video) only, no reference image. Hard rules: no music and no narration — only ambient sound; no stabilized footage; no text on screen; identity, clothing and hairstyle stay consistent throughout.
- ▌ Natural Ambient Living Generator · gordonlu bundleCreate a 7–8 second MiniMax H3 text-to-video clip of a full-scale natural environment living and breathing at real-world size — forest, ocean, sky, river, desert, snowfield, window rain — where an environmental system (wind, light, water, cloud) is the declared primary motion owner. The scene must show continuous visible change every 1–2 seconds and reach a soft peak beat; no camera-only clips, no static desktop-wallpaper output. Follows base-en.txt exactly. Miniature/diorama scenes belong to miniature-world-landscape-generator; ink-paper subjects to living-ink-painting-video-generator; construction to construction-timelapse-video-generator.
- ▌ Follow One Encounter And Departure · gordonlu bundleGenerate a 1970s Greenwich Village street cinéma-vérité piece — a 16mm film camera tailing a twenty-something completely make-up-free young woman through streets and squares, brushing past a group of children, and dissolving into the moving street — as a 15s 4:3 24fps short with raw 16mm film grain, organic texture, pure natural daylight, heavy shoulder-held shake, no stabilization, occasional focus loss, slight light leaks, and an unsettled frame until the very end. Use when the user wants a street-follow / documentary-style long take, a 1970s analog film look, a cinéma-vérité chase after one person, or a "one encounter then departure" observational short. T2VA (text-to-video) only, no reference image. Hard rules: no subtitles and no logo, no music — only on-location sound; no modern digital quality; no on-screen text; keep 4:3 framing and 16mm grain.
- ▌ Stage Earnest Upgrade Displacement · gordonlu bundleGenerate a 15s 1:1 (1440×1440) 24fps 1950s three-strip Technicolor TV commercial — "AMBER MIND INC. / The Future Arrives Thursday" — in a bright warm appliance showroom where a family admires the old TRITON thinking machine, a larger new TITAN is wheeled in and the TRITON is wheeled away, dragging the boy who refuses to let go across the polished floor. Use when the user wants a retro 1950s real-shot-feel commercial, a single-camera locked tripod + one slow dolly spot with exactly four hard cuts, a warm saturated Technicolor ad, or an earnest upgrade/displacement gag — T2VA (pure text generation, no reference image). Hard rules: no text anywhere except the final end card, max 4 cuts, no modern objects or humanoid robots, full 1950s wardrobe/hair.
- ▌ Stage Progressive Character Reveal · gordonlu bundleGenerate a 15s 16:9 cinematic character-entrance short from a single character reference image — a progressive four-stage reveal (defining detail → partial body → face close-up → full-body hero frame) that builds presence, silhouette, temperament and restrained action, e.g. a fog-harbor lantern-bearer slowly ascending the steps. Use when the user wants a "progressive reveal / slow unveiling" character entrance for a game, anime or film intro, a character-presence cinematic shot, or a "detail-first, hero-frame-last" opening — I2VA with @图片1 as the character reference only, never used as the first frame. Locks the face, body proportions, hairstyle, outfit, colors, materials and overall silhouette to the reference; no redesign, no simplification, no replacement of any defining visual feature.
- ▌ Stage Threshold Inspection Passage · gordonlu bundleGenerate a 15s 16:9 hyper-real first-person scientific-expedition documentary short: two volcanologists in anti-slip boots, helmets and headlamps squeeze into a newly formed lava tube on a misty volcanic wasteland at dawn (5:30–5:45) to record it before its structure changes further — entry, wall reveal, steam anomaly discovery, cautious check, marking and retreat toward the mouth-light. Use when the user wants a first-person expedition / cave-inspection documentary, a "threshold crossing into an unstable structure" scene, or a realistic scientific survey short — T2VA (pure text generation, no reference image). Hard rules: rock walls must read as real solid volcanic rock (not props), natural light diffusion, no fantasy, no supernatural, no impossible cave geometry, calm professional demeanor, breathing and footsteps always audible.
- ▌ Living Ink Painting Video Generator · gordonlu bundleCreate a 7–8 second MiniMax H3 text-to-video shot in which a traditional ink-wash subject painted on rice paper comes alive, moves briskly within the flat painting, then accelerates and breaks free from the paper as a living ink-and-water form while preserving ink material continuity.
- ▌ Miniature World Landscape Generator · gordonlu bundleCreate cinematic 7–8 second MiniMax H3 text-to-video clips featuring richly detailed miniature landscapes across natural, urban, historical, fantasy, geological, terrarium, tabletop, and sci-fi themes. Each generation combines one landscape family, one presentation mode, and one clearly visible hero effect with appropriate tempo and strong scale cues.
- ▌ Pencil Transmutes Reality Into Sketch · gordonlu bundleGenerate a POV magic-pencil film — a hand holding an orange pencil points at moving subjects in real Hong Kong and completely transmutes each into animated 2D hand-drawn illustration while the subject keeps its natural motion — as a 10s 16:9 24fps short converting four subjects in sequence: an MTR train, a street cat on a tong-lau windowsill, a red taxi and a woman on a park bench, against a bright realistic Hong Kong environment. Use when the user wants a "magic pencil" effect, a live-action-to-animation transmutation, a POV doodle-wand video, or a "whatever the pencil points at turns into a sketch" clip. T2VA (text-to-video) only, no reference image. Hard rules: the transmutation must be total (no live-action color, material or light left; no half-real-half-sketch states) and synchronous with the subject entering frame (the pencil starts drawing the moment the subject appears — never after a delay); conversion applies only to the pointed subject, never spreads to the environment; converted subjects keep the
- ▌ Construction Timelapse Video Generator · gordonlu bundleCreate a 7–8 second MiniMax H3 accelerated construction timelapse showing a coherent build from an empty or incomplete state to a finished structure, environment, machine, interior, infrastructure project, or miniature world. Automatically route between FL2VA, L2VA, T2VA, and I2VA according to available start/end images. Enforce persistent construction state so completed work never disappears, resets, relocates, or changes design.
- ▌ Record Memory Through Fixed Viewfinder · gordonlu bundleGenerate a 15s 4:3 24fps nostalgic CCD-camera memory short: a young East Asian woman films herself with an early-2000s Sony Cyber-shot point-and-shoot, the whole frame styled like old memory-card footage pulled out years later. Use when the user wants a handheld, imperfect, deeply nostalgic look — persistent camera UI overlay (REC, focus box, battery, timestamp "8.1.2009 3:15 AM", "1/30 F3.4 ISO100"), direct-flash hard light, soft grain, digital compression, LCD pixel texture, slight wrong white balance, tiny autofocus search. I2VA with one character reference (image only, not first frame). Includes a verified 15.08s output.
- ▌ Prove Product Through First Person Test · gordonlu bundleGenerate an authentic first-person UGC food-tasting review (iPhone selfie-vlog style, handheld, fast jump cuts, TikTok/Reels aesthetic) from a single food-product reference image — a young Chinese girl reviews a roujiamo (Chinese hamburger) in Mandarin across 8 shots. Use when the user wants a product-tasting review, a first-person "prove the product" short, a UGC-style food review, or an e-commerce KOC-style food clip. Locks the product (bun shape, filling thickness, juice state, toppings, proportions) to the reference in every shot; no commercial-ad staging, no cinematic grade, no CGI food.
- ▌ Miniature Creature Awakening Video Generator · gordonlu bundleTurn one image of a small creature, mascot, fantasy pet, figurine-like character, or tiny animal into a stable 7–8 second MiniMax H3 image-to-video clip. The creature must visibly wake, perform at least one clear body action, interact with one nearby object or effect, and finish in a readable pose while preserving identity.