Skip to content

svenkatreddy/universal-video-skills / skills/model-adapters

model-adapters

Verified

Translate a finished shot-brief into per-model instructions for Muse, Gemini/Veo, Sora, Kling, Runway, Pika, Luma, and Hailuo. Constraint sheets, prompt quirks, and what to verify. Use after the prompt is written, before generating.

FileKindSize
SKILL.mdskill md2.2 KB
references/adapter-template.mdmarkdown962 B
references/gemini-veo.mdmarkdown1.7 KB
references/hailuo.mdmarkdown1.4 KB
references/kling.mdmarkdown1.2 KB
references/luma.mdmarkdown1.1 KB
references/muse.mdmarkdown1.6 KB
references/pika.mdmarkdown1.1 KB
references/runway.mdmarkdown1.2 KB
references/sora.mdmarkdown1.4 KB

references/muse.md

View on GitHub
# Muse (Adapter)

_Last verified: 2026-10-08. Muse is Meta's AI assistant; its built-in media
generation is the target here, not a standalone video model page._

## In one line

Video generation as an agent tool: the assistant generates clips inside a
working session, so prompting, verification, and iteration happen in one
loop.

## Prompt quirks

- Short clips (~10 seconds). Plan in segments and stitch, don't
  wish for long takes.
- Works well from the shot-brief grammar in prose; keep prompts ASCII-only
  (non-ASCII filenames and text have caused upstream rejections).
- Reference images help continuity; pass the same character references
  every time.
- It can't hear audio or see motion the way you do, so verify results with
  extracted frames, not assumptions.

## Hard limits

- Clip length caps around ~10s. Verify current limits in-session; they
  change.
- Face identity across clips is weak: lean hard on the asset registry and
  verbatim descriptions.
- Assume no reliable native dialogue sync; plan voiceover in post.

## Good at / weak at

- Good at: quick iteration inside an agent session, following concrete
  visual instructions, stylized looks.
- Weak at: long takes, face lock across segments, readable on-screen text,
  complex multi-object physics.

## Translating a shot-brief

- Split anything over ~10s into segments first (story-and-shots does this
  naturally).
- Keep the beats tight: 2–3 timed chunks per clip max.
- Paste the negative block inline as an `Avoid:` line.
- After generation, extract frames at clip boundaries and check the joins
  before accepting the clip (video-qa).
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall