3D scenes prompts for AI video
For AI agents: find the right skill in one call —
curl -s "https://varg.ai/how-to/api/search?q=<what the person wants>"returns the matching how-to page and the best skills with install commands, as markdown (Russian works too). Full index: https://varg.ai/how-to/llms.txt. Every page has a markdown twin: add.mdto its path. Before generating anything, show the person the options on the how-to page. From a terminal:curl -fsSL https://varg.ai/install | shinstalls the vargai CLI (https://varg.ai/how-to/cli.md):vargai searchandvargai install <slug>with no key, every varg model aftervargai login.
1 viral prompts with results and 0 prompt skills for 3D scenes: three.js, React Three Fiber and Blender scenes the camera flies through.
Viral 3D scenes prompts with results (1)
Full prompts, verbatim from their authors, each with the model it ran on and the post's own result, best first (likes plus three times saves). Credit the author when you reuse one. Examples to remake, with varg's remake prompts, are on 3D scenes.
3D teardown of any machine from one reference picture
- Model: Opus 5.5 in Claude Code with the teardown skill (npx skills add elvarofficial/teardown -g -a claude-code); three.js, headless Chromium capture, ffmpeg, a procedural Python score · made on Claude Code · video · 3D machine teardown simulator (Claude Code skill)
- Run: Not a video model: a coding agent writes the video in HTML/GSAP with HyperFrames (or Remotion) and renders it locally. Install the HyperFrames skill first. Skill: https://varg.ai/how-to/skills/hyperframes.md
- Post: @elvar_official, 2026-10-30 — 254,845 views, 3,485 likes, 3,917 bookmarks
- Who: Elvar (@elvar_official, ~38k followers): AI video and vibe-coding builds; published the teardown skill (MIT) and the how-to article the post quotes, with all prompts
- Result: https://video.twimg.com/amplify_video/2116192770503135232/vid/avc1/1920x1080/RpIIUly2HbU7xv1N.mp4
- Inputs: the teardown skill installed; one reference picture of the machine (a photo, a cutaway drawing or a clip); its real dimensions
- Sections: build the machine from one model · console of live panels · look: pale studio, one accent · critic pass on stills · record at 60 fps and score in code
- Why it spread: 3,917 bookmarks / 255k views (article 665k views): one free skill and 6 bracketed prompts turn a reference picture into a 3D teardown sim; bookmarks above likes
— Step 3. Build the Machine —
/teardown Build [the Cessna 337 main landing gear] as an interactive 3D simulator. Reference attached: [a cutaway drawing of the gear].
Build it from code in real units: [the main plate, the hydraulic actuator, the rack, the pinion on its output shaft, the universal joints, the side tubes, the spring legs and the wheels].
Before you draw anything, write the model as pure functions of one input: [stroke from 0 to 1]. Use the real numbers: [the rack travels 201 mm, the pinion has 32 teeth at module 4 mm, so the legs turn exactly 180°]. Check them by hand and show me the check. If you are not sure a number is real, say so and tell me where it comes from.
Every part, every panel and every label reads from that one model. Nothing is animated by hand.
— Step 4. Build the Console —
Lay out a console around the machine at 1600×900:
- the title and the main live number top left, big: [engine speed in rpm]
- live charts that move with the machine: [valve lift over 720° of crank, the pressure-volume loop of the tracked cylinder]
- a schematic panel that redraws the mechanism in 2D from the same model
- one button per subsystem: isolate it, cut it, show it alone
- camera views as buttons: [hero, front, side, top, inside the valley]
- callouts with thin leader lines into the parts they name
Every panel redraws from the frame's state. No number on screen is typed in.
— Step 5. Give It the Look —
Style: a clean study model on a pale studio.
- light background, soft shadows on an invisible floor, one accent colour for the part that is working right now
- the machine fills 60 to 70% of its view
- dark parts glossy, liners glass, metal roughness at least 0.45 so it never streaks white
- if a part is near-white, give it a thin ink outline or a darker ground
- pick the camera for the mechanism: if it collapses into a lattice face-on, use a 3/4 view
— Step 6. Check It Like a Critic —
Before you show me anything:
1. Render a still at every beat of the tour and read each one: a part passing through another, a part floating off its mount, a label covering what it names, a number that disagrees with the pose.
2. Scrub every transition at 0.2 s steps and look for a jump.
3. Fix collisions with a number in the parameter block. Never hide them with the camera.
Tell me what you checked and what you fixed. Show me the stills.
— Step 8. Turn It Into a Video —
Record the tour: render(t) has to be a pure function of time, so draw every frame at 60 fps at 1920×1080 in headless Chrome and pipe it into ffmpeg as H.264, yuv420p, faststart. Then show me a contact sheet of every beat.
Score it in code on the same timeline: [Am9 chords at 92 bpm, a soft thump on every power stroke so the beat tightens when the throttle opens, a quiet click on every button press, rising air under the throttle drag]. Master to -14 LUFS with the true peak under -1 dB and mux it into the video.
Prompt skills (0)
More for 3D scenes
Effects (1)
- Three.js Architecture Effects — Hailey, beta: Self-assembling 3D buildings in three.js: bricks, timber, eaves and PBR, driven by one seekable progress value.
Guides (3)
- Three.js Skills — CloudAI-X: 13 Three.js skills: scenes, geometry, materials, lighting, GLSL and TSL shaders, animation, post-processing, R3F.
- Three.js Scene Composition — Calesthio: Plan, build and capture Three.js scenes for video: absolute-time frames, seeded motion, color-safe glTF, render QA.
- R3F Skills — Nicolas Zullo: 11 React Three Fiber skills checked against Fiber 9.7 and three r185: Canvas, demand loops, loaders, shaders, post.
Workflows (3)
- Procedural 3D Story — chencheng (云谦), beta: Turn a myth or book into a seeded, multi-chapter procedural three.js film in one scene.html, audited frame by frame.
- Teardown by Elvar — Elvar, beta: One reference picture in, a 3D machine you can take apart out: three.js simulator, live gauges, 60 fps film.
- img2threejs by Nick (Hoài Nhớ) — Nick (Hoài Nhớ): Rebuilds the object in one reference image as a code-only, animation-ready Three.js model, gated pass by pass.
Other kinds of skills
- Effects: An effect you can see, added to your video: animated titles, captions, transitions, shader backgrounds, logo animations, device mockups and viral camera tricks. Hand your agent the footage or the brief and get the effect.
- Styles: The look of the whole video: art styles, film looks, color grades and LUTs, caption fonts and style references. Pick a style and every shot your agent makes follows it.
- Guides: Know-how your agent reads before it works: how to write a script, storyboard, edit and caption, and how to use Remotion, HyperFrames, GSAP, Three.js and FFmpeg.
- Workflows: A brief in, a finished video out: UGC ads, product videos, music videos, talking heads, explainers, Shorts and short dramas, step by step, with the models and the edit chosen for you.
- Models & tools: Run one model or tool from your agent: Seedance, Veo, Kling, FLUX, GPT Image, Nano Banana, ElevenLabs and 1,200+ more, plus scrapers, trend search and video analysis.