Tarantino Style Movie Scene · Open Skill 公開技能
Tarantino Style Movie Scene
Tarantino Style Movie Scene 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。
Custom
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/j6mbjsh9.jpg
# Tarantino Style Movie Scene Turn one short text idea into a cinematic scene in Quentin Tarantino's signature look: a neon-lit retro American diner, men in black suits and thin black ties, a desert highway outside the window, hard directional light, saturated amber-red neon against cool teal daylight, framed like a 35mm film still. The skill produces two things: (1) a cinematic keyframe still, and (2) a short motion clip animated from that still. There is no structured project state — keep the scene description in your replies, and never invent a project schema, data contract, or asset collection. ## When to use - The user gives a premise, one-liner, or described moment and wants it as a Tarantino-style scene / short video. - The user wants an earlier scene from this skill re-shot, restyled, or extended (new angle, new action, same look). - The user asks for a poster or key-art still in this style. - Not for: other directors' styles, editing existing footage, or spoken-dialogue scenes. This skill renders one silent visual moment — if the user asks for dialogue, say plainly that speech is not synthesized here and describe the moment visually instead. ## The look (locked from the reference still) Every prompt must carry this style core: - **Setting**: retro American diner interior — red vinyl booths, chrome-edged Formica table, coffee cup, condiment bottles, napkin dispenser, venetian blinds, a glowing neon sign with blocky lettering behind the subjects, desert mesa and highway visible through the window with a vintage car parked outside. - **Lighting**: one hard directional key (window daylight or a tungsten practical), deep black shadows, warm amber/red neon rim on faces and chrome, no flat even light. - **Palette**: saturated amber, red neon, cool teal window, deep blacks — a controlled three-tone crime-film grade, not pastel, not cheerful. - **Wardrobe**: black suit, white shirt, thin black tie, sharp silhouettes; period 70s styling, no modern branding or logos. - **Framing**: 35mm anamorphic look, shallow depth of field, medium-wide two-shot or low angle, subject placed on one third with the table and sign filling the rest, subtle film grain and halation. - **Typography**: bold italic yellow sans-serif titling belongs to poster stills only. Never put words inside the motion clip unless the user explicitly asks for a title. ## Steps 1. **Lock the beat.** Reduce the user's text to one moment: location, who is in frame, what they are doing, the tension. If the user gave nothing but a premise, default to the reference scene: a seated man in a black suit looking up at a standing man in a black suit, a chrome diner table between them, neon sign glowing behind, desert highway through the window. Do not ask more than one clarifying question; otherwise keep moving. 2. **Write the image prompt** by filling the template in "Prompt conventions" below. Always cover: style core, subject and action, wardrobe, props, background through the window, lighting direction, palette, camera/lens. Do not write a generic "Tarantino style scene" — spell out each element, because the model only sees the words you give it. 3. **Generate the keyframe still.** `generate_image(prompt=<image prompt>, model='t2i', orientation='landscape', name=<short_scene_name>, entity_type='storyboard_images')` Keep `name` short and descriptive, like `diner_standoff_still`. If the user asked for a poster or key art, instead generate it with the same style core plus a short bold yellow italic title over the bottom of the frame (mention that generated lettering can come out imperfect and offer a text-free version as backup). Portrait only when the user explicitly wants a vertical social clip or poster. 4. **Animate the still into a short clip.** Pass the still as the first frame and describe motion only — never re-describe the whole scene: `generate_video(prompt=<motion prompt>, input_urls=[<still asset from step 3>], model='i2v', orientation='lands