One 7B model both draws and edits, does native 2K and real transparent backgrounds, and fits the card already in your desktop. What it takes to run, the settings that matter, the licence terms — and why older Qwen-Image LoRAs will not fit it.
The open YuE2 3B model turns your lyrics into a melody-and-chord score you can edit, and only then sings the whole song from it. It runs on your own GPU and PocketComfy has it from version 3.48.
An old phone on a USB cable shows your computer temperatures and load, free memory, ComfyUI generation progress and the remaining limits of your AI coding agents. One HTML page with no libraries, a few PowerShell scripts and an APK a few dozen kilobytes in size.
Generative models give you two usable takes out of five and leave the picking to you. I built a Claude Code skill wired into my own ComfyUI that does the picking itself — and I am releasing it for free.
TRELLIS.2 and Pixal3D run natively in ComfyUI 0.34.0: one photo in, a 3D model with PBR materials out, on an 8 GB card, for free and with no licence stopping you from selling the result.
I built a web interface for Higgs Audio v3: Czech text to speech, voice cloning from a half-minute sample, dialogues for six characters and voice-to-voice conversion. Three-step install, running entirely on your own GPU.
The Photoshoot node for ComfyUI describes a person across 44 fields, then shoots a full series on its own — framing, pose, expression and aspect ratio vary while the person stays the same. No LoRA training, no extra dependencies.
The open-source ALL in ONE MiniMaxH3 node packs eight modes of the MiniMax H3 video model — from text-to-video with native audio to latent-space clip chaining — into a single node. Runs locally, no API key.
Boson AI released Higgs Audio v3, an open-source voice AI that sounds like a real person — it clones voices from a few seconds of audio, conveys real emotion and handles 100+ languages. In blind tests it even beats paid leaders. Here is why it is one of the most exciting models of the year.
A single ComfyUI node that unifies everything around LTX 2.3 generation – timeline, prompt relay, audio and keyframes.
Anthropic has launched Claude Design, a new AI feature for creating designs, prototypes, and presentations via chat. In this article, you'll find out what it can do, where the limits are, how to integrate it into your existing workflow, and which resources and videos to watch to get started quickly.
An open-source multi-agent orchestrator that coordinates a team of specialized AI models — Claude, GPT, Gemini, and local models — and selects the best one for each task.