Scenario Creative Skills
Give AI coding assistants workflows for creating images, video, audio, and 3D assets through Scenario and local creative software.
- Source repo
- scenario-labs/skills
- Stars
- ★ 3.9k
- Last updated
- 1d ago
- License
- MIT
- Primary language
- Python
- FA score
- 61/100 · Some gaps
At a glance
- How it runs
- Works with
- Platform-specificCodex · Claude Code
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A game art team uses Scenario to make style-consistent sprites, tilesets, textures, skyboxes, and 3D models.
- Not a fit if
- Teams that need fully offline generation without the Scenario MCP
- Developers seeking a general coding assistant rather than creative asset workflows
- Website
- skills.sh/scenario-labs/skills ↗ set on GitHub
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-20latest release skills-v0.62.0 (2026-11-11)+27.0 ★/day over 7 days
- Source review
- 61/100 · Some gaps
What does this agent do, and when should you use it?
This repository provides Agent Skills for AI coding assistants, including Claude Code, Cursor, Codex, and Copilot. Its main skills use the Scenario MCP to discover tools and parameters, run generation jobs, and create images, video, audio, textures, skyboxes, 3D assets, and custom models. The workflows cover model selection, pre-run pricing, character and brand consistency, review, and asset organization. A separate set of expert skills drives local applications such as Blender, Maya, ZBrush, Unreal Engine, Unity, Godot, and Roblox Studio. Scenario workflows require the Scenario MCP; expert tools require the relevant installed application, and the README describes different verification levels across tool families.
After installation, an assistant follows the skills to connect to Scenario MCP, discover available operations and parameters, run generation jobs, wait for results, then display or download assets. Image workflows include text-to-image, editing, multi-reference composition, and masked inpainting; other skills cover video, audio, 3D, consistency, model training, asset analysis, quality review, format adaptation, and workflows. Users can estimate costs before generation and assess results through quality gates, model comparisons, and refinement loops. Expert tool skills route work to installed DCC applications or game engines and document their local execution and verification paths.
- A game art team uses Scenario to make style-consistent sprites, tilesets, textures, skyboxes, and 3D models.
- A video producer builds ads, music videos, captions, or social clips from storyboards or product images.
- A brand designer creates reusable character or product identities and keeps them consistent across image sets.
- A creator generates music, sound effects, speech, or video scoring, then assembles the audio and video output.
- A Blender, Maya, or game-engine user wants an AI assistant to drive local modeling, animation, rendering, or asset export.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npx skills add scenario-labs/skills --skill scenarionpx skills
How do you install or deploy this agent?
Install all Scenario skills with npx skills, or select an individual skill. Scenario workflows require a Scenario MCP connection and a Scenario account for OAuth sign-in or an API key. Expert tools install as a family and require the corresponding application on the machine.
npx skills add scenario-labs/skills --skill scenarioFor Claude Code, add the Scenario MCP server:
claude mcp add --transport http scenario https://mcp.scenario.com/mcpAlternatively, add https://mcp.scenario.com/mcp to an MCP client and sign in with a Scenario account via OAuth or use an API key.
How do you use this agent?
Once installed and connected, describe the asset or task to the assistant and use the matching skill. The core scenario skill follows a discover, inspect parameters, run, wait, display, and download loop.
npx skills add scenario-labs/skills --skill scenarioExpert tools require the relevant DCC application or game engine installed locally; install each family as a set using the command in its README.
What are this agent's strengths and limitations?
- One skill repository spans image, video, audio, texture, skybox, and 3D asset workflows.
- Model-specific skills give concrete guidance on model choice, parameters, costs, and constraints.
- Consistency, quality review, model comparison, and asset organization support the broader production workflow.
- Expert tool families extend assistant workflows into local Blender, Maya, ZBrush, and several game engines.
- Scenario creative workflows depend on Scenario MCP and require a Scenario account or API key.
- Expert tools require the corresponding local application; verification varies by software family.
- Use requires a terminal, network access, and configuration in a client that supports Agent Skills and MCP.
- The README does not state an overall price; generation costs and account permissions may affect use.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Scenario Creative Skills This agent | 61 · Some gaps | Agent plugin / skill | ★ 3.9k | 1d ago | Python | Codex · Claude Code |
| Generative Media Skills | 79 · Good | CLIFree + model costs | ★ 7.1k | 3d ago | Python | Claude Code · Claude.ai |
| PAI-Pro Filmmaking Studio | 64 · Some gaps | Self-hosted serviceFreemium | ★ 358 | 1mo ago | JavaScript | Codex · Claude Code |
| ArcReel Video Creation Workspace | 77 · Good | Self-hosted serviceFree + model costs | ★ 6.7k | 1d ago | Python | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
The skills drive Scenario MCP generation, uploading assets and spending credits, so external effects are real. Positive evidence: SKILL.md files repeatedly require dry_run pricing, jobs_wait instead of re-running model_run, and no resubmission on timeout. least_privilege scores 1 because no allowed-tools are declared (AGENTS.md explicitly leaves the field unused), so permission scope rests on runtime defaults. data_flow_transparency scores 1: asset upload to mcp.scenario.com is only described in the scenario skill, not in the sampled files. sensitive_data_handling scores 1: SECURITY.md notes signed URLs are credentials, but skill bodies do not systematically constrain secret handling. rollback scores 1: ad-variants keeps per-variant state and retry rules, but there is no general undo/recovery path. source_attribution scores 1: model and tool references point at public docs but are not individually cited. dependency_security scores 2: CI pins action SHAs and uses pip --require-hashes, though devDependencies use ranges.
self_consistency scores 2: the AGENTS.md authoring contract (frontmatter limited to name/description/license, word budget, model-naming rules) broadly matches the sampled SKILL.md files, but README lists many vendor-named skills while AGENTS.md forbids naming generative models, leaving a tension the reader must resolve. dependency_availability scores 1: the product depends on the Scenario MCP service, OAuth/API keys, and Blender 5.2.1, none verifiable from the packet; package.json requires node>=22.18 and pnpm 11. failure_messages scores 2: ad-variants and scenario-3d give concrete failure semantics (403 ModelAccessRestrictedError, timeout is not failure, reconcile transport failures before resubmission), but coverage is partial.
audience_and_scenarios scores 3: README groups skills by role and goal, INSTALL.md gives per-role commands, and coverage spans 2D artists to producers and game art, with clear evidence. capability_boundaries scores 2: several SKILL.md files state explicit non-goals (ad-variants excludes new commercials and Cartesian batches), but other boundaries rely on sibling cross-references. trigger_precision scores 2: descriptions start with 'Use when' and pack keywords as required, though keyword density risks over-triggering. environment_fit scores 2: the repo separates MCP skills from expert tools needing local DCC software and pins a Blender release, but host-environment differences are not fully covered.
information_architecture scores 3: layout, name-equals-directory rule, one-level supporting-file linking, and tests/ placement are all specified and script-checked. install_notes scores 3: README and docs/listing-and-install.md detail npx skills add, commit-pinned installs, MCP connection, and picker behavior. naming_stability scores 2: naming rules are explicit and uniqueness is checked, but plugin ids carry two-digit position prefixes that renumber on regrouping. examples_and_faq scores 2: each skill is required to have a worked example and Common mistakes, but no FAQ appears in the sample. known_limitations scores 2: 'authoring time' caveats, regional restrictions, and timeout semantics appear, but there is no consolidated limitations list. license scores 3: MIT LICENSE file matches license: MIT in frontmatter. versioning_changelog scores 2: README describes release-please and CHANGELOG flow, but no CHANGELOG file is in the packet. maintenance_responsibility scores 1: the publisher is unverified, SECURITY.md gives a reporting channel, but no explicit maintainer or support commitment is stated; no brand credit is given.
output_usability scores 2: outputs target production formats (GLB/FBX, splats, finished video, captions) with download and preview paths, but actual output quality cannot be verified statically. marginal_value scores 2: skills add model selection, pricing, and consistency knowledge over raw MCP calls, yet much content defers to runtime schema discovery, so the increment needs real use to confirm. cost_benefit scores 2: dry_run pricing, Turbo-then-Quality iteration, and numOutputs batching show cost awareness, but the large skill count means loading every description at startup, which README itself flags for --skill "*".
claim_traceability scores 2: AGENTS.md requires every tool and parameter claim to be grounded in mcp.scenario.com/docs/tools, and SKILL.md files cite public docs and siblings, but specific figures (1300 CU, about an hour) are only marked 'authoring time' without per-claim links. cross_source_corroboration scores 1: README, AGENTS.md, and SKILL.md corroborate each other only partially, and the sampled tests cover admin-analytics and Blender scripts, not the generation skills. fact_inference_separation scores 2: skills distinguish 'authoring time' observations from schema contracts and defer to model_schema_get, though some statements still blend inference with fact.
- No allowed-tools are declared, so permission scope rests on runtime defaults; confirm least-privilege MCP client configuration before adoption.
- Generation and uploads spend credits and cause external effects; always dry_run first, and treat quoted prices and timings as authoring-time data that may be stale.
- The publisher is unverified by the FollowAgents registry, so maintenance responsibility and update path are unclear; pin a commit and review changes yourself.
- README's --skill "*" installs every skill and loads all descriptions at startup, adding context cost; install by role or goal instead.
- The packet samples only some skills and tests; behavior of unsampled generation skills is unknown and should not be inferred from the sample.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
skills-v0.62.0 skills: v0.62.0
Release notes excerpt
[0.62.0](https://github.com/scenario-labs/skills/compare/skills-v0.61.0...skills-v0.62.0) (2026-11-11) Features * scenario-model-training: teach MiniMax H3 video LoRA training ([#414](https://github.com/scenario-labs/skills/issues/414)) ([2e1f272](https://github.com/scenario-labs/skills/commit/2e1f272919778b3c67dd60772bf8bf18bea50823)) * scenario-playable-ads: start from an idea or a genre name ([#410](https://github.com/scenario-labs/skills/issues/410)) ([0afcbaa](https://github.com/scenario-labs/skills/commit/0afcbaa2bca4dd9d5b294609726ce50c89588f07)) …View on GitHub ↗ -
skills-v0.61.0 skills: v0.61.0
Release notes excerpt
[0.61.0](https://github.com/scenario-labs/skills/compare/skills-v0.60.1...skills-v0.61.0) (2026-11-08) Features * scenario-blender-blockout: add a bicycle kind and a rider recipe ([#315](https://github.com/scenario-labs/skills/issues/315)) ([d12e237](https://github.com/scenario-labs/skills/commit/d12e237c7fbdb5ac15024252af0b4b327997f770)) * scenario-comic-creator: add comic creation skill ([#353](https://github.com/scenario-labs/skills/issues/353)) ([7cde6f7](https://github.com/scenario-labs/skills/commit/7cde6f7c8434f6468c3b3c304065a2defce97273)) …View on GitHub ↗ -
skills-v0.60.1 skills: v0.60.1
Release notes excerpt
[0.60.1](https://github.com/scenario-labs/skills/compare/skills-v0.60.0...skills-v0.60.1) (2026-10-29) Bug Fixes * skills: drop unversioned installs and pin the kinetic setup ([#340](https://github.com/scenario-labs/skills/issues/340)) ([b8544b9](https://github.com/scenario-labs/skills/commit/b8544b9c86e0685844cc6effbb27f3885f744d3a)) Documentation * agents: point skill-development at its new upstream home ([b8544b9](https://github.com/scenario-labs/skills/commit/b8544b9c86e0685844cc6effbb27f3885f744d3a)) …View on GitHub ↗
FAQ
What do I need to use the Scenario skills?
Can I install just one skill?
npx skills add scenario-labs/skills --skill scenario installs the core skill; other individual skill names are listed in the repository.Do the expert tools require Scenario MCP?
How much does generation cost?
Is Scenario Creative Skills open source?
Is Scenario Creative Skills safe to use?
Is Scenario Creative Skills still maintained?
Who makes Scenario Creative Skills?
What is Scenario Creative Skills's official website?
Thanks — we use this to improve the page.