Badgids

(Alan Guice)

Technologist · Computer Scientist & Software Engineer · Project Manager
Musician · Maker · Community Advocate

I build practical systems where technology, creative work, and real-world problem solving meet.

“Keep the neurons firing!”

Active repository with recent development work.

Maintainer-review-driven TDD repair loop for the Pi coding agent.

• Exact per-finding HEAD contracts. A finding starts from the ledger's exact current HEAD and completion verifies ancestry, cleanliness, changed-file scope, and commit count.
• Configurable test profiles. Define focused, regression, smoke, and final command groups. prrepairtests executes them and records evidence.
• Fresh-context independent review. A separate read-only pi -p process, with no extensions, skills, prompt templates, context files, or write tools, tries to disprove each repair against the original maintainer reproduction.
• CI ingestion. Reads GitHub combined commit status and check-runs for the repair HEAD.
• Maintainer filtering. By default only OWNER, MEMBER, and COLLABORATOR review material is ingested; explicit include/exclude lists are configurable.
• Review-head comparison. Tracks the PR head, inferred reviewed head, and local HEAD so a repair does not silently start against the wrong revision.

JavaScript

ComfyUI custom nodes for deterministic multi-speaker timed audio mixing and exact MiniMax H3 target-audio latent locking.

Author/Developer: Alan Guice (Badgids) Copyright: © 2026 Alan Guice (Badgids). License: MIT License

• Locks an exact user-controlled waveform into MiniMax H3's target audio latent.
• Keeps H3 video denoisable while the target audio stream remains fixed.
• Accepts an effectively unlimited number of timed audio events through ComfyUI's native Autogrow inputs.
• Places every source on H3's 24 fps target-video timeline using deterministic integer frame-to-sample conversion.
• Mixes multiple speakers or sound sources sample-accurately into one H3 target waveform.
• Supports overlapping dialogue and layered audio.

Python

Story-Film Skills is a local-first Agent Skills suite for story writing, book development, screenwriting, image/audio/video generation, directing, feature-film production, postproduction, and release delivery. It uses durable project files, stable IDs, deterministic validators, and recoverable workflows.

Author/Developer: Alan Guice (Badgids) License: Apache License 2.0 Copyright © 2026 Alan Guice (Badgids).

Python · ai · ai-agent · ai-coding · ai-tools · comfyui · film · film-editing

A ComfyUI set of custom nodes to set up threejs interactive scenes to be used as video reference

An interactive 3D scene staging, actor acting, and camera directing suite of custom nodes for ComfyUI.

• Multi-Actor Chaining: Sequence and record multiple independent actors (humanoids and vehicles) on the same stage with synchronized playback.
• Camera Spring Arm (Anti-Collision): Real-time obstacle avoidance prevents camera clipping against walls and geometry during dynamic camera moves.
• Actor Color Customization: Dynamic mesh coloring to visually distinguish different actors across the 3D viewport and node timeline.
• Universal Stage Builder: Universal 3D geometric engine with 5 core primitives and new presets (courthousesquare, gasstation).
• Vue.js Architecture: Modernized modular reactive UI components (StagingWidget, ActingWidget, DirectingWidget).
• Animated 3D Humanoid Model: Custom 3D human model with bone structure and armature animations.

JavaScript

A standalone application for installing, launching, updating, repairing, inspecting, exporting, importing, and sharing ComfyUI installations, models, and workflows.

Author and creator: Alan Guice (Badgids) License: Apache License 2.0 Version: 0.8.8

• a compact Textual terminal interface for people;
• a complete non-TUI command-line interface for scripts and automation;
• portable YAML-based setup profiles;
• shared external model and workflow libraries;
• update snapshots and rollback;
• a YAML-backed runtime cache for fast startup and explicit rescans;

Python · comfyui · comfyui-api · comfyui-custom-nodes · comfyui-manager · comfyui-manager-2026 · comfyui-nodes · comfyui-plugin

Clip chaining for MiniMax H3 in ComfyUI - motion and audio genuinely continue across joins

Build a multi-scene MiniMax H3 video with one reusable sampling graph. Review each scene, compare takes, resume from checkpoints, and assemble the result later—without keeping the whole production in memory.

• Bring nightly's fractional video/audio denoise-mask correction to main,
• Correct only missing streams; leave native fixes and source latents intact.
• Conflicting VIDEO PNG exports automatically use 2, 3, etc. folders,
• Copy verified earlier scenes safely when a rerender changes mid-sequence;
• Capture a still from a saved Review Gate preview into the Project Asset
• Resize the optional prompt editor using its new drag handle; height is

Python

Embed AI assistants in Godot with the ability to read and write code in Godot's Code Editor.

Godot AI Assistant Hub ========================== Latest version: 2.1.0 (What's new?) (Upgrading to a newer version)

• Godot context - When tools are enabled, the assistants can know what scene and script you are editing, so you can simply ask things like "What does this code do?".
• Multiple assistants - Have multiple chat sessions with different types of assistants simultaneously.
• Conversation editor - Edit the conversation history in case your assistant gets confused by some of your prompts, or you want to delete something that is using too much context.
• IP configuration - Even for local LLMs, you can connect to a local computer and call remotely. This can be useful, for example, if you are a team but have only 1 powerful computer to run LLMs.
• 6.1. Fill up the data requested in the window. Icon is optional.
• 6.2. Click "Set tools access" if present. Hide whatever tools are not required. Using all tools can use too much context for some computers.

GDScript · ai · assistant · gemini · godot · jan · llm · ollama

Neural latent upscaler for Minimax H3 (24ch). Bypasses costly 5B-param VAE decode/encode. Upscale low-res latents directly, then refine. Accelerates high-res video gen, outperforms naive interp.

Neural Latent Upscaler for Minimax H3 Video Generation Learned · High-fidelity · 2D & 3D Variants

• ✅ Learned latent upscaling — neural network trained for Minimax H3 latents, far sharper
• ✅ Two backbones — pick the fast 2D variant or the temporally-coherent 3D variant.
• ✅ Three ways to set output size on the 3D node — scale by multiplier, target dimensions,
• ✅ Tiled hires-fix upscaler (MMH3 Split Upscale combo) — re-sample the H3 AV latent at higher
• ✅ 24-channel Minimax H3 latent — uses the exact per-channel mean/std normalization from
• ✅ Auto architecture detection — reads inchannels, block counts, temporal config and

Python

I am a Chattanooga-area technologist, project manager, musician, digital creator, woodworker, outdoorsman, and community advocate. Online, I work as Badgids. My path has never fit neatly into one lane: I move between structured systems and creative chaos, especially where the rules are still being written.

Across open-source AI, emerging industries, community organizing, music, digital media, and physical craft, I tend to serve as a builder, organizer, communicator, and translator between worlds. I am more interested in how a system behaves in real use than in polished hype around it.

• 🧠 Technology & AI: Python, automation, local inference, LLM experimentation, ComfyUI, open-source model engineering, agent tooling, and practical software.
• 🧭 Project leadership: Organizing cross-functional work in uncertain, fast-changing environments and translating goals, constraints, and execution.
• 🎵 Music & media: Writing, producing, and releasing original music as Dead Man's Mandolin, alongside years of YouTube experimentation and documentation.
• 🪵 Craftsmanship: Building tangible work through RockCreek ShopWorks, with a focus on woodworking and respect for the material.
• ⚖️ Community advocacy: Public communication and organizing around bail reform, pretrial justice, incarceration, equity, and community accountability.
• 🏕️ Outdoors: An Eagle Scout and lifelong camper, hiker, bushcrafter, and curious student of history, archaeology, and the natural world.

My technical work emphasizes practical experimentation: build it, run it locally, find where it breaks, and make it useful. That has included automation and bots, real-time Whisper transcription, local generative-AI workflows, autonomous CAD experiments with CadQuery and CQparts, and open-source model work on Hugging Face.

I develop and merge GGUF models such as Gonzo-Chat-7B-GGUF, built for local inference with tools including llama.cpp, Ollama, and LM Studio. I also work with ComfyUI, Flux GGUF, PuLID, local CUDA systems, and agent-compatible engineering skills.

Before my work at Snapdragon Hemp, I created the Hamilton County Community Bail Fund in partnership with CALEB—Chattanoogans in Action for Love, Equality and Benevolence. During this period, I also served as a Director of CALEB through a highly active era of bail-reform and pretrial-justice organizing.

I was interviewed about bail funds and pretrial detention in at least five local newspaper articles, including coverage by the Chattanooga Times Free Press and the Cleveland Daily Banner. I also co-hosted a local talk-radio program covering bail reform, criminal justice policy, community accountability, and equity.

After my bail-fund and CALEB work, I served as a Project Manager at Snapdragon Hemp during the hemp and CBD industry's early expansion. I coordinated work across business goals, operational realities, and a rapidly changing regulatory environment around the Farm Bill era. The job demanded comfort with uncertainty—the kind of environment where I do some of my best work.

Music is not a side note in my life; it is another way I design systems, pacing, structure, atmosphere, and meaning. As Dead Man's Mandolin, I write and produce original work that pulls together roots instrumentation, gothic atmosphere, indie-rock energy, and heavier influences.

My YouTube channel began with prolific bushcraft, camping, gear, and DIY experimentation and grew to include technology, commentary, creative work, and ongoing project documentation. I use video as a public workshop and an archive of ideas in motion.

The discipline and outdoor foundation I developed as an Eagle Scout still shapes how I work. Through RockCreek ShopWorks, I bring that mindset into woodworking—turning raw material into useful, carefully made pieces while staying connected to traditional building, the outdoors, and hands-on problem solving.

I am especially interested in collaborating on machine learning, applied AI, local inference, automation, agent systems, and open-source tools. Ask me about computer science, programming, ML/AI, ComfyUI, or turning an ambitious idea into a working system.

I've never followed a particularly straight career path. I've spent most of my life learning how things work, building things, getting involved when I thought something needed changing, and following interesting problems wherever they led.

I earned the rank of Eagle Scout when I was younger. Scouting gave me a solid foundation in leadership, service, outdoor skills, self-reliance, environmental stewardship, and figuring things out with whatever tools were available.

A lot of that stuck with me. I'm still a camper, hiker, bushcrafter, maker, and the kind of person who would rather understand how something works than just accept that it does.

Badgids started as my online identity long before I was doing anything with AI.

I became a prolific bushcraft creator on YouTube, making videos about camping, outdoor skills, gear, DIY projects, experiments, and whatever I happened to be learning at the time.

YouTube became a way to document what I was doing and share it with other people. Over the years the subject matter changed, but I never really stopped doing that.

The same channel and online identity eventually expanded into music, technology, software, AI, commentary, and all the other things I work on now.

At one point my attention shifted heavily toward criminal justice and community organizing.

I created the Hamilton County Community Bail Fund in partnership with CALEB, Chattanoogans in Action for Love, Equality and Benevolence, and served as a CALEB director during a very active period of work around bail reform, pretrial detention, incarceration, equity, and community accountability.

That work put me in a much more public role than I had been used to. I became a spokesperson on bail and pretrial justice issues and was interviewed several times by local newspapers, including the Chattanooga Times Free Press and the Cleveland-area press.

I also co-hosted a local talk radio program, where we discussed bail reform, criminal justice policy, community issues, and politics affecting people in the Chattanooga area.

It taught me a lot about organizing people, explaining complicated issues in plain language, dealing with disagreement, working with the public, and turning an idea into something that actually operates in the real world.

After my work with the bail fund and CALEB, I became a Project Manager at Snapdragon Hemp.

This was during the period when the Farm Bill had opened up a rapidly changing hemp and cannabinoid industry. The laws, regulations, products, operations, and business landscape were all changing at the same time.

Project management in that environment meant dealing with a lot of moving parts and solving problems without having a nice established rulebook to follow.

That suited me pretty well.

Music has been part of my life alongside everything else.

As Badgids and through Dead Man's Mandolin, I write, produce, record, and release original music. My work pulls from bluegrass, Appalachian music, rock, metal, blues, dark country, and whatever else fits what I'm trying to make.

I also create videos, characters, artwork, and other media around the music.

For me, music and software aren't as far apart as they might look. Both involve building something piece by piece, trying things that don't work, changing them, and eventually getting everything to fit together.

My longtime interest in computers eventually led me into machine learning and local AI.

I started working with Python, automation, bots, Whisper transcription, local inference, LLMs, model merging, GGUF, llama.cpp, Ollama, Hugging Face, CUDA, and generative AI tools.

On Hugging Face I released the Gonzo Chat and Gonzo Code model families, including GGUF versions for local inference.

On GitHub, that work grew into software, utilities, custom nodes, integrations, workflows, experiments, and agent tooling.

I've always been especially interested in local and open systems. I like software that I can run on my own hardware, inspect, modify, repair, and adapt instead of depending entirely on somebody else's service.

A lot of my current development work centers around ComfyUI and agent-driven creative tools.

I started out building workflows and eventually found myself building the tools around the workflows too.

That now includes custom ComfyUI nodes, workflow management, model management, film-production tooling, continuity systems, reference management, audio and video tools, local LLM infrastructure, CAD and 3D experiments, and AI agents that can work with real project files and applications.

Projects such as ComfyUI Setup Manager, Story-Film-Skills, ComfyUI-Pi-Agent, Pi-ComfyUI, and my other ComfyUI projects have grown out of that work.

A big part of what I'm interested in now is making AI agents actually useful for real work. Not just answering questions, but understanding a project, using tools, working with files, keeping track of state, and helping carry something from an idea to a finished result.

I also still like making things that don't require a GPU.

I do woodworking under RockCreek ShopWorks, building with both new material and ethically sourced reclaimed or recycled wood.

Woodworking scratches a different version of the same itch as programming. You start with raw material, work within its limitations, solve problems as they come up, and hopefully end up with something useful that will last.

It also gives me an excuse to get away from a computer once in a while.

These days I split my time across open-source software, local AI, ComfyUI, agent systems, media production, music, YouTube, woodworking, and whatever project has managed to get my attention next.

I've been an Eagle Scout, bushcraft YouTuber, community organizer, bail fund founder, CALEB director, radio co-host, project manager, musician, software developer, AI tinkerer, and woodworker.

It probably looks like a strange collection of things from the outside.

From where I'm standing, it's mostly been the same habit repeated for a long time:

Find something interesting. Learn how it works. Build something. Share what I learned. Move on to the next problem.