Backlot — local AI render engine: ComfyUI workflow registry, MCP server, web studio, Blender bridge, and QC suite. Engine half of the ai-director two-session pattern.
View on GitHub ↗A local AI production studio that drives ComfyUI and Blender — controllable by humans (web UI) and by AI agents (MCP), on your own GPU, at $0/generation.
Backlot wraps a running ComfyUI server with a typed engine, a no-build web studio, and an MCP server, so the same capabilities are drivable three ways:
1. Web studio — http://127.0.0.1:8765: prompt → image/video/music, gallery, editing tools, storyboards, an autonomous Director. 2. HTTP API — every feature is an endpoint with a wait flag (POST /api/generate?wait=true returns the finished asset in one call), so other apps and scripts can drive it. 3. MCP server — AI agents (e.g. Claude Code) get listworkflows / runworkflow / directvideo / render3danimation tools and can produce finished videos autonomously.
Backlot is an independent project, not affiliated with or endorsed by Comfy Org (ComfyUI) or the Blender Foundation. Names are used descriptively.
Lane Capabilities ------ Image FLUX.1-dev · SDXL · FLUX Kontext instruction editing · SDXL inpaint (mask brush) · Real-ESRGAN upscale · Kontext presets (replace background, camera angle, relight, multi-view turnaround) Video Wan 2.2 (5B + 14B + Lightning 4-step) · LTX-Video · SVD image-to-video · motion-sync (reference clip → depth → restyled subject) Talking heads InfiniteTalk (Wan 2.1 14B audio-driven portrait — photoreal lip-sync) · multi-person two-shot (per-speaker audio routed to faces) · Sonic (legacy) Audio ACE-Step text-to-music · Kokoro TTS voice-over (8 voices) 3D / Blender Headless Blender bridge (bpy runner, GPU Cycles) · depth-sequence export → Wan VACE coher
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/lnguyen503--backlot.html)
A Centralized MCP Server for Fetching Data from Many Countries' Govs
axisrobo/mneme-openOpen-source client SDKs (Python, TypeScript, Go), CLI, API docs, and prebuilt binaries for the Mneme agent cognition runtime (memory, context, ontology, knowledge). Apache-2.0.
ahadkhan9/workspace-mirror-skill Lulu-The-Narwhal/dali-mcpCreative intelligence MCP — score your prompt before you spend the credit
KernelLord/pickysteveSkill router and context picker for Claude Code, Cursor, and Aider. Auto-picks the right skill for every prompt, with prompt-injection filtering for MCP.
billy12151/memory-arbiter-mcpMCP memory server with dual-timeline conflict arbitration. Not just store & search — resolve contradictions.