Stability Studio

Local image and video generation for Stability Matrix + ComfyUI, exposed as an MCP server for Cursor, Open Interpreter, and other agents.

Talk in plain language — "anime portrait", "juggernaut cinematic" — the server picks checkpoints, builds workflows, and queues generation on your GPU.

Features

• Image generation — SD 1.5, SDXL, Pony, Flux.2 Klein via style presets and model_families
• Image editing — unified edit_image, four art food groups (anime / fantasy / cyberpunk / photoreal) — IMAGE-EDITING.md
• Video generation — Wan T2V/I2V from saved Stability Matrix workflow JSON (default I2V: i2v_5b)
• Style catalog — aliases, LoRAs, prompt prefixes in catalog.yaml
• Workflow converter — UI-format ComfyUI workflows → API prompts with Wan model/T5 remapping
• Agent-ready — MCP tools + Cursor rules + Open Interpreter skill examples

Requirements

• Stability Matrix with ComfyUI package
• Python 3.11+
• Models per workflow (SDXL checkpoints for images; Wan + umt5 for video — see docs)

Quick start

Not a one-click installer — an MCP workflow that lets your AI assistant drive ComfyUI for you.

git clone https://github.com/Jamescd1980/stability-studio.git
cd stability-studio
.\install.ps1

Copy workflow JSON into Stability Matrix before video/MOSS: see stability-studio-mcp/workflows/README.md (or bundled-workflows/).

1. Copy .cursor/mcp.json.example → .cursor/mcp.json if Cursor does not auto-detect Python
2. Open this folder in Cursor (or configure Open Interpreter — see docs below)
3. Tell the agent: "Help me set up Stability Studio" — it reads the onboarding pack
4. When ready: launch ComfyUI from Stability Matrix; complete Tier 1 (images) before video

Audience	Start here
Less technical + AI assistant	onboarding/README.md
Developers / power users	CURSOR-INTEGRATION.md
Storyboard example	STORYBOARD-QUICKSTART.md

MCP tools

Tool	Description
get_generation_context	model_families, style_readiness, styles, GPU limits, Wan assets
check_style_assets / download_style_assets	Flux2 / image model manifest (companion downloads)
list_styles / list_checkpoints / list_loras	Library scan
list_video_workflows	t2v, i2v_5b, i2v, …
check_backends	ComfyUI / InvokeAI reachability
check_wan_assets / download_wan_assets	Wan model manifest check and Hugging Face download
check_comfyui_dependencies	Missing custom nodes for video
install_comfyui_dependencies	Clone known node packs
edit_image	Unified natural-language edit (food_group=anime|fantasy|cyberpunk|photoreal)
setup_image_editing	One-shot edit stack (IP-Adapter + ControlNet SDXL/SD1.5 + segmentation)
generate_image	Style-aware T2I
list_art_food_groups	Four art food groups + default styles
generate_video	Wan T2V/I2V (ComfyUI draft); image_path required for I2V
generate_video_hero	Wan2GP Lightning v2 hero I2V (headless MCP)
check_gpu_backend	ComfyUI vs Wan2GP policy (required before GPU tools offline)
get_onboarding_context	Start here — tiers, questions, VRAM rules, install checklist
plan_storyboard_scene	Hero I2V + MOSS + splice plan from a short script (no GPU)
check_storyboard_readiness	MOSS + Wan2GP + GPU + project layout for storyboards

Storyboard (Rin example): STORYBOARD-QUICKSTART.md — Wan2GP hero + MOSS + generate_storyboard.py (v1.0.0-beta)

Documentation

| STORYBOARD-QUICKSTART.md | Linked hero clips + MOSS + splice (Rin example) |

Doc	Audience
CURSOR-INTEGRATION.md	Cursor setup, local vs cloud agents
OPEN-INTERPRETER-INTEGRATION.md	OI + LM Studio, troubleshooting, lessons learned
AGENTS.md	AI agent instructions (all platforms)
HARDWARE.md	GPU tiers and generation limits
MODEL-FAMILIES.md	SD 1.5 / SDXL / Pony / Flux2 / Wan — samplers, files, agent checklist
IMAGE-EDITING.md	Edit tools, decision tree, lessons learned, roadmap
IP-ADAPTER-SETUP.md	IP-Adapter + ControlNet automated setup
WAN-ASSETS.md	Wan model manifest and downloads
GITHUB.md	Publish / zip checklist
stability-studio-mcp/README.md	Package-level detail

Project structure

studio-agent/
  .cursor/mcp.json.example      # Copy → mcp.json (gitignored)
  .cursor/rules/                # Agent rules for generation
  config-examples/              # OI TOML, Cursor JSON, OI skill
  bundled-workflows/            # Wan/MOSS JSON → copy to SM Data/Workflows/
  stability-studio-mcp/
    server.py                   # MCP entry
    catalog.yaml                # Styles + video workflow ids
    config.yaml.example         # Path template (copy → config.yaml)
    workflows/                  # Same workflow JSON + README
    studio/                     # Engine, ComfyUI client, converter

Wan2GP + storyboard (Rin reference)

Linked hero sequences: walk → bow → lunge with MOSS dialogue and ffmpeg splice.

# Plan manifest (no GPU)
python stability-studio-mcp/scripts/storyboard/generate_storyboard.py plan --title rin --script-file beats.txt

# MCP: check_storyboard_readiness → generate_video_hero (×3) → generate_audio → splice
python stability-studio-mcp/scripts/storyboard/generate_storyboard.py splice

Full walkthrough: STORYBOARD-QUICKSTART.md · Release notes: RELEASE.md

Status

Feature	Status
Image (generate_image)	✅ Working
Video t2v (Wan 2.1)	✅ 81 frames max @ 16 fps on 16 GB
Video i2v_5b	✅ Default draft I2V — 65 frames max on 16 GB
Hero I2V (Wan2GP)	✅ generate_video_hero — 49f @ 832×480 (~125 s on 16 GB)
Storyboard CLI	✅ studio/storyboard_cli.py + generate_storyboard.py
Video i2v (14B)	✅ Legacy; explicit workflow_id=i2v
InvokeAI image fallback	Optional

License

MIT — see LICENSE.