GoodCase.aiGoodCase.aiAI Case Evidence Library
CasesDailySkillsCreatorsFavoritesSubmit
中文EN
Agent Access →
GoodCase.ai · AI Case Evidence LibraryLoading the case library.
ChangelogFeedback
GoodCase.aiGoodCase.aiAI Case Evidence Library
CasesDailySkillsCreatorsFavoritesSubmit
中文EN
Agent Access →
AI VideoX2026-07-26Evidence L1#youmind#Seedance 2.0#human-reviewedSeedance 2.0

Glitch Art Multi-Dimensional Face Transformation

Ushizaru / うしざる avatarCreatorUshizaru / うしざる →

An experimental glitch-style prompt that uses rhythmic jump cuts to switch between 3D CG, live action, animation, and pixel art.

View original source ↗
Source ♥ 80

Prompt

After a rhythmic approach like stop-motion, a glitch video that aggressively moves back and forth between completely different dimensions (2D⇔3D), such as 3D CG, live action, 2000s anime, and pixel art, without minding if the facial structure changes. It ends with a browser-crasher-like screen collapse. Video concept: Completely eliminate smooth camera movement. The camera approaches the face in stages (jump cuts) to the rhythm of “Pa! Pa! Pa!” After the approach, the girl's face does not need to preserve its “original features.” The entire screen is pulled in and keeps being violently swapped (moving back and forth) through hard cuts into “completely different art styles and dimensions (2D⇔3D),” such as dimensional 3D CG, raw live action, a 2000s-style anime beautiful girl, and flat retro pixel art. Camera and editing policy: [Extremely important: Smooth changes and zoom-ins are absolutely not allowed. The sequence must be composed of stop-motion-like jump-cut approaches and “complete switching” into art styles from different dimensions.] 0.0–3.0 seconds: [Stop-motion approach] Start from the first shot. Instead of the camera moving smoothly closer, it approaches the face in stages with jump cuts at a steady rhythm every 0.5 seconds: “Pa! Pa! Pa! Pa!” At the moment each cut switches, strong chromatic aberration (RGB shift) noise runs across the image. 3.0–8.0 seconds: [Dimensional collapse; complete art-style replacement] Reach an extreme close-up of the face. From here, the rendering style of the video itself keeps crackling and switching (moving back and forth) every 0.5 to 1 second, without concern for changes in the facial structure. It is forcibly swapped with flashes into completely different art styles that cross between 3D and 2D dimensions, such as “dimensional polygonal 3D CG” → “ultra-high-definition live-action human” → “2000s-style cel anime beautiful girl character (2D)” → “completely flat retro pixel art (2D).” 8.0–10.0 seconds: [Browser-crasher-like collapse ending] The violent back-and-forth between art styles exceeds the processing limit, and the video completely collapses. Using the face footage as an underlay, “massive error windows,” “video panels with different visual expressions,” and “vividly colored block noise” keep popping up chaotically in multilayered layers like a browser crasher. The entire screen is completely filled with a buggy video collage, becoming a chaotic collapse screen that retains none of the original video's form. Constraints: During the midsection art-style changes, “smooth changes that try to preserve the original face silhouette” are absolutely unnecessary. The structure may change, so violently repeat “complete hard cuts (swaps)” many times into rendering styles from different dimensions: “dimensional (3D)” and “flat (2D),” as well as “live action” and “2000s anime.”

Reusable method

Key decisions

Check whether camera continuity, transition logic, and action resolution all hold in the final result.

Adapt it to your subject

Extract “target result → the subject, scene, camera movement, and final action → constraints” as a reusable template for similar work.

Where it breaks

Run single-variable comparisons around shot length, transitions, and motion amplitude. Record the model, cost, and failures; turn it into a Skill only after it repeats reliably.

Recommended models

Seedance 2.0
Stability
Cost
Medium
This case has not yet been distilled into a Skill.Request it in feedback →
Ushizaru / うしざる avatar

Creator

Ushizaru / うしざる

Ushizaru / うしざる focuses on AI video production and is active on X. “Glitch Art Multi-Dimensional Face Transformation” is the current representative case, built with Seedance 2.0 with its stability retest open to votes.

Editor’s pickAI Video
View creator →

Keep exploring

Explore related cases.

View all AI Video Case →
Retro 16mm Aerobics Fitness Vlog
AI VideoX07/25/2026𝗦𝗮𝗻𝗶𝗮 →

Retro 16mm Aerobics Fitness Vlog

A cinematic prompt for creating a nostalgic early-2000s-style fitness vlog starring a Korean idol, emphasizing authentic handheld camerawork and a lo-fi videotape texture.

Skill · POV and vlog presence

Prompt

DV 16mm handheld tape camcorder POV by CHASE, who occasionally props the camera on a treadmill console or nearby bench. Natural hand shake, crooked framing, delayed autofocus, clum

Source heat
74
Stability
View case →
2000s MiniDV Han River Bike Ride
AI VideoX07/26/2026Zephyra Leigh →

2000s MiniDV Han River Bike Ride

A comprehensive multi-scene prompt designed for Seedance 2.0, intended to mimic the nostalgic handheld MiniDV home-video style of 2004.

Prompt

Main Subject: Young Korean woman, 22, natural appearance, shoulder-length dark brown hair tied in a loose ponytail with wispy bangs. Wearing a light blue oversized hoodie, black cy

Source heat
80
Stability
View case →
2D Sticker Capybara Kitchen Prank
AI VideoX07/26/2026Soran →

2D Sticker Capybara Kitchen Prank

A mixed-media video prompt for Seedance 2.0 combining a realistic 1st-person kitchen POV with a 2D animated capybara sticker character performing a salt-pouring prank.

Skill · Animation style and character consistency

Prompt

Live-action footage + 2D anime sticker composite comedy short video 【Style】 Live-action footage + Flat 2D Sticker Composite comedy short video. First-person cooking perspective, with a strong contrast between the realistic kitchen texture

Source heat
33
Stability
View case →
GoodCase.ai · AI Case Evidence LibraryEach detail page keeps the work, method, creator, and retest evidence in one reading path.
ChangelogFeedback