Practical AI Guides & a Free Claude Code Course

This is a LoRA — a small add-on file that changes how an existing AI model behaves. It is not a new model on its own. It comes from Pika, a video-generation company known for its Pika app. The file fuses Pika's own motion style with speed-up techniques built for MiniMax-H3. MiniMax-H3 is an open video-generation model from the Chinese company MiniMax. The result: a five-second clip renders in 2, 4, or 8 steps, instead of the 6 steps Pika's official version needs. The video comes out with sound already matched to the picture. ComfyUI is free, open-source software for building AI image and video workflows. Someone running MiniMax-H3 there downloads one of the LoRA files from Hugging Face, a site that hosts AI models. The file goes into the loras folder of that setup. Then they add a "LoraLoaderModelOnly" step to their workflow, and pick the 2-, 4-, or 8-step version depending on whether they want a quick draft or a finished render.

Phonon is a free text to speech model developed by Gradium. You can play around with it here on the Hugging Face website which is a platform dedicated to artificial intelligence models and demos. No installation required, go to the Phonon space, the model will load automatically (less than 75 MB), select one of the 100 available pre-trained voices and start typing! There are 3 models available, the first one is a small model of 31M parameters and supports only English, while the other 2 are of 40M parameters each and support English, French, German, Spanish and Portuguese. After loading the model in the browser, it will be stored in the cache memory and the audio will be played immediately, without much delay.

zeekk0/OpenAmika

OpenAmika is an open source AI friend developed by a small team called MekaHime. She is a Live2D rendered anime girl which means she’s a 2D image that’s animated rather than being a 3D model. She can hold conversations with you via speech and text in real time, she remembers things you say and she is designed to live in your city on your desktop (Windows only). All of her senses (listening, speaking, remembering, seeing) are handled locally on your machine. The only thing that isn’t local is the language model which is either hosted by yourself or via a third party service called OpenRouter which acts as a middleman for multiple language models using one key. According to the developers this should cost around $1.50 USD/month if you use her 30 times per day or $5 USD/month if you use her 100 times per day. If you want to host the language model locally yourself you can do so via another application called Ollama however you will need a model with at least 27 billion parameters (a way of measuring the complexity and power of a language model). As the developers state very clearly “anything less than 70 billion parameters can be really bad” because she won’t have any personality, she won’t remember things you say and her responses won’t be formatted correctly. The developers consider anything above 70 billion parameters to be reliable. The source code is made available under the MIT license which is an open source license meaning you are free to use it however you wish. However her voice, Live2D model and art assets are proprietary and licensed only for private use meaning you must accept the license agreement when setting her up. There is also a version that doesn’t require installation which can be accessed via your web browser at mekahime.com. In order to run OpenAmika you need Windows 10 or 11 and an NVIDIA graphics card with the RTX 20 series or higher with at least 8GB of VRAM (16GB if you want to use her eyes). You then need to follow the installation guide which can be found on the GitHub repository (a website that hosts developer code) which uses approximately 45GB of storage space when complete.

voyager-so/vork

Vork is a rendering engine for 3D scenes and visual effects designed to be used by a code agent. There is no user interface to click through; instead, the agent programs a scene (in JSON or via a Python/TypeScript toolkit), gets a fast preview render, inspects the resultant image, tweaks the code, rinses, repeats…until the shot is looking good. The final image is generated on the GPU via Vulkan (on Windows/Linux) or Metal (on Apple silicon), combining rapid approximate occlusion with ray-traced lighting based on real hardware. Check out some of our example scenes, like a million birds flocking on the GPU, approximated black holes warping space-time, and postprocessing effects overlaid on regular phone videos—all truly rendered on the GPU, not AI-generated! If you want to play around with Vork yourself, just ask your code agent to git clone github.com/voyager-so/vork, follow the instructions in the README, check out the built-in help (vork help), and then compile and render a basic scene. This is currently available for free and for research purposes only.

shanselman/gputaskman

GPUTaskMan is a tool that allows you to monitor activities within your GPU. Instead of just seeing how much video RAM you’re using, this tool provides a breakdown of what processes are consuming video RAM. Specifically, it breaks down the weight and buffers used by machine learning models. For example, in one test, a local machine learning model with 4 billion parameters consumed 3.95 GB of video RAM. GPUTaskMan determined that 2.49 GB was used for weights and 1.21 GB was used for buffers. These figures are consistent with the GPU's own readings. GPUTaskMan supports Windows (NVIDIA/AMD GPUs), Linux, and MacOS (Apple Silicon). It also runs within WSL (Windows Subsystem for Linux). It can identify models being executed via Ollama or LM Studio. These are tools that allow you to execute AI models locally. You can get started by installing the package using either pipx install gputaskman or uv tool install gputaskman, followed by running the gputaskman command in your terminal. You can find more information about this project here: https://github.com/shanselman/gputaskman.

majorminorlabs/hermes-voice

Hermes Voice is a free Home Assistant Add-On which enables you to interact with a Hermes Agent Profile (an open source AI agent framework created by research group NousResearch) through the voice user interface of Home Assistant (an open source home automation software). The add-on achieves this by installing an alternative open source software called echo-dot-assist onto an older Amazon Echo Dot/Show device, replacing the existing Amazon Alexa software. By doing so, we can keep all of the hardware components (microphones, speaker, audio processing) of the Echo device but disable the Alexa software. The voice pipeline of Home Assistant is used to convert our speech into text and send it to the Hermes Agent Profile where it will be processed and returned to us through the speakers of the Echo device. A custom wake word model has been developed which enables us to say “Hey Hermes” instead of “Alexa” to activate the agent. To run Hermes Voice, you will need to have Home Assistant running and a working Hermes Agent Profile with its own AI Access. Due to a recent security patch blocking the installation of non-verified add-ons, our one click installation option is currently unavailable. Follow the instructions on our Github page (github.com/majorminorlabs/hermes-voice) for manual installation.