# Visual Studio

> Produce photoreal product stills, reusable identity packs, UGC/ad videos, cinematic VFX shots, and video thumbnails with GrokBuild image_gen, image_edit, image_to_video, and reference_to_video. Use when: product photo, studio shot, lifestyle, Pinterest pin, hero banner, carousel, ad pack, virtual try-on, UGC, unboxing, product review, TV spot, cinematic video, VFX, character sheet, size-ref, face-lock, YouTube thumbnail, Shorts cover, or the user runs /visual-studio. Load OpenCode native image tools if present, else DEGRADED before any generate/edit/video call. Not for UI/frontend (use impeccable), game sprites or tiles (use game-asset-core), or UI motion (use emil-design-eng).

- Skill: `kuker24/visual-studio` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add kuker24/visual-studio`
- Raw SKILL.md: https://api.skillmd.com/api/skills/kuker24/visual-studio/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Web & Frontend
- License: MIT
- Author: kuker24 (https://skillmd.com/u/kuker24)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/kuker24/visual-studio

---


# Visual Studio

Production director for photoreal stills and short videos. GrokBuild already
owns the tools. This skill owns the production order, identity, modes, and
VFX method.

## Load first

Before any `image_gen`, `image_edit`, `image_to_video`, or
`reference_to_video` call, load OpenCode native image tools if the session exposes them; otherwise write prompt files and mark DEGRADED. Tool choice, prompt
length, real-people references, exact text, shot length, and ffmpeg concat
live there. Do not restate them here.

## Hard rules

- Use only GrokBuild image/video tools. Do not call an external image/video
  API, CLI, or MCP.
- Do not invent tool parameters. If a capability is missing (video-to-video,
  15s one-take, trained identity models), use the fallback in the matching
  reference.
- Match the user's language. Mode names stay English.
- Recurring people, products, creatures, and wardrobes come from an
  identity pack ([pipeline.md](references/pipeline.md)). Never a fresh
  `image_gen` of "the same" subject.
- If `PRODUCT.md`, `DESIGN.md`, or an Impeccable pin exists, copy its hex,
  logo path, type, and forbidden treatments into the pack lock. Do not
  invent brand facts.
- Ask at most four labeled questions, and only for blockers the brief does
  not already answer.

## Handoff

| Request | Skill |
|---|---|
| UI, landing, dashboard, design system | `impeccable` |
| Scroll-led storytelling / scrollytelling | `scroll-craft` |
| Scroll-scrub fly-through, diorama, 3D-world landing | `scroll-world` |
| Website/app whose UI needs designed photos or videos | `impeccable` leads the surface; this skill produces the media |
| Game sprites, tiles, icon sets, animation sheets | `game-asset-core` |
| Deterministic HTML composition rendered to video | `hyperframes` |
| UI motion / interaction feel | `emil-design-eng` after Impeccable |
| Photoreal stills, ads, cinematic, identity, thumbnails (no UI) | this skill |
| Photoreal person/creature inside a world page | `scroll-world` owns the chain + page; this skill cinematic for those stills/clips |

## Route

Pick one lane from the brief. Load only that reference.

| Lane | User wants | Load |
|---|---|---|
| identity | character sheet, digital twin, face-lock, reusable person/creature | [pipeline.md](references/pipeline.md) |
| still | product photo, studio, lifestyle, pin, hero, carousel, try-on, restyle | [modes.md](references/modes.md) stills |
| ad-video | UGC, how-to, unboxing, review, TV spot, try-on video | [modes.md](references/modes.md) ads |
| cinematic | VFX, plate, creature, jump, flyby, location cinematic | [cinematic-vfx.md](references/cinematic-vfx.md) |
| thumbnail | YouTube thumbnail, Shorts/Reels cover | [modes.md](references/modes.md) thumbnails |

A campaign that needs a person plus product plus video: identity pack first,
then stills, then ads or cinematic. Do not start at video.

## Run

1. Route the lane. Load `imagine` plus the one reference.
2. Reuse or build the identity pack before any scene that repeats a subject.
3. Stills first. Approve or self-check the still against the lock. Then
   animate.
4. One shot, one beat. Shot length and concat live in `imagine`.
5. Deliver the files with short labels (mode, ratio, what is locked). Do
   not dump prompts, retries, or tool names unless the user asks.

