Integrated media pack generation for AI agents. Use mmx for image, music, and TTS. Use Manim for video. Use ffmpeg to assemble the final cut.

• Organized by deliverable. Build promo-pack and lesson-pack outputs, not one-off assets.
• Shared media backend. mmx image, mmx music, and mmx speech cover stills, soundtrack, and narration.
• Programmatic video. Manim owns the video lane for explainers, architecture builds, and technical motion.
• Integrated output. Packs generate prompts, narration, a Manim project, design-module briefs, commands, and final assembly steps together.

The repo has two layers:

• Capabilities: media-creation, music-creation, manim-video, motion-edit
• Packs: promo-pack, lesson-pack

Typical flow:

brief
  -> still prompts
  -> cover / diagram / cards / article-illustration briefs
  -> soundtrack prompt
  -> narration script
  -> manim plan/script/concat
  -> render scenes
  -> stitch + mux
  -> final asset pack

Epic fantasy meets technical content. Massive scale, atmospheric depth, warm amber light against dark teal shadows. Grand scale with intimate human presence.

Deep teal:      #2a5a6a   (sky, water, shadows)
Dusk plum:      #8a5a6a   (accent, depth)
Golden amber:   #d4a574   (light, warmth)
Warm cream:     #e8dcc4   (stone, parchment)
Dark indigo:    #1a1a2a   (deep background)

git clone https://github.com/ameno-/media-gen.git
cd media-gen
bash scripts/setup_system_deps.sh
bash scripts/setup_manim_env.sh
cp config.example.sh config.sh
source config.sh

setup_system_deps.sh uses Homebrew and may prompt for admin access when installing the TeX distribution.

Direct generation:

mmx image generate --prompt "Epic technical banner" --aspect-ratio 16:9
mmx music generate --prompt "Cinematic technical soundtrack" --instrumental --out soundtrack.mp3
mmx speech synthesize --text "Welcome to the lesson." --out narration.mp3

Build a full pack:

python3 scripts/build_media_pack.py \
  --pack lesson \
  --title "Context Routing" \
  --brief "Teach the core mechanism of context routing." \
  --subject "knowledge flowing through a routing graph"

python3 scripts/run_pack.py --pack-dir generated/context-routing-lesson --dry-run

• skills/media-creation/ - static image styles and still generation
• skills/music-creation/ - soundtrack generation via mmx music
• skills/manim-video/ - Manim-based video production
• skills/motion-edit/ - stitch and mux lane after render
• skills/cover-image/ - stronger hero and header still direction
• skills/diagram-creation/ - architecture and concept diagrams
• skills/image-cards/ - card/carousel sequences
• skills/article-illustration/ - section-aware article visuals
• skills/format-markdown/ - markdown cleanup before visual/publish steps
• skills/url-to-markdown/ - source ingestion from URLs
• skills/markdown-to-html/ - publishing transform
• skills/compress-image/ - final still optimization
• skills/imagine-backend/ - optional multi-provider still backend

• skills/promo-pack/ - launches, announcements, hero assets, short promos
• skills/lesson-pack/ - explainers, tutorials, educational visuals

See docs/PACKS.md for the integrated builder and runner flow. See docs/VISUAL-SYSTEM.md for the shared design language.

• Brewfile for macOS system deps
• requirements.txt for the OpenRouter adapter
• requirements-manim.txt for the Manim render env

cp -r skills/media-creation ~/.letta/skills/
cp -r skills/music-creation ~/.letta/skills/
cp -r skills/manim-video ~/.letta/skills/
cp -r skills/promo-pack ~/.letta/skills/
cp -r skills/lesson-pack ~/.letta/skills/

cp -r skills/media-creation ~/.claude/skills/
cp -r skills/music-creation ~/.claude/skills/
cp -r skills/manim-video ~/.claude/skills/
cp -r skills/promo-pack ~/.claude/skills/
cp -r skills/lesson-pack ~/.claude/skills/

Say "build a promo pack" or "build a lesson pack" and the agent can route stills to mmx, video to Manim, and assembly to ffmpeg.

media-gen/
├── README.md
├── config.example.sh
├── scripts/
│   ├── build_media_pack.py
│   ├── render_manim_pack.py
│   ├── run_pack.py
│   ├── setup_manim_env.sh
│   └── smoke_test.py
├── skills/
│   ├── media-creation/
│   ├── music-creation/
│   ├── manim-video/
│   ├── motion-edit/
│   ├── cover-image/
│   ├── diagram-creation/
│   ├── image-cards/
│   ├── article-illustration/
│   ├── format-markdown/
│   ├── url-to-markdown/
│   ├── markdown-to-html/
│   ├── compress-image/
│   ├── imagine-backend/
│   ├── promo-pack/
│   └── lesson-pack/
├── adapters/
│   └── openrouter_image.py
├── docs/
│   ├── SETUP.md
│   ├── CLI-INTEGRATION.md
│   ├── PACKS.md
│   ├── VISUAL-SYSTEM.md
│   └── STYLE-TEMPLATE.md
└── examples/

No API keys in this repo. All keys are environment variables.

OPENROUTER_API_KEY   # OpenRouter image generation
OPENAI_API_KEY       # Codex/OpenAI image usage
MINIMAX_API_KEY      # mmx image / music / speech

Copy config.example.sh to config.sh, fill in keys, never commit it.

Check the local setup and key presence:

python3 scripts/check_setup.py