ComfyUI
11 storiesComfyUI is a modular, graph-and-node interface, API and backend for running diffusion models locally, aimed at generative image work. Recent coverage includes the v0.40.0 release, Kandinsky 6.0 video generation and upscaling adding ComfyUI support, and tools such as Berth that swap ComfyUI and other engines on a single GPU. NVIDIA's RTX Spark compact desktops also list ComfyUI among supported software.
Related topics
ASUS, Dell and MSI launched RTX Spark compact desktops with a 20-core Grace CPU, Blackwell RTX GPU and up to 128GB unified memory, delivering up to 1 petaflop for running local models and always-on agents, backed by Adobe, Blender and ComfyUI software support.
A Redditor couldn't sell an old AM4 system at €750, but after adding two R9700 GPUs and demoing Qwen3.6 and ComfyUI, it sold for €5,400; buying replacement cards cheaper left him effectively earning €1,300 for the original system. The post warns buyers to check used-parts prices before buying a local-AI PC.
Modular diffusion model GUI, API and backend with a graph/nodes interface.
A modular graph/node diffusion model GUI and fast local inference engine for generative image users.
A Reddit user released Berth, an open-source tool that monitors and automatically swaps AI engines such as llama-swap, strata, ComfyUI and OpenWebUI on a single 24GB GPU, avoiding VRAM conflicts. It suits multi-task GPU sharing and can prevent processes from grabbing the card during gaming.
Modular diffusion model GUI, API and backend with fast local inference for creators
The most powerful and modular diffusion model GUI, API and backend with a graph/nodes interface.
ComfyUI, the AI image generation workflow tool, released version v0.40.0, though the community post did not detail the specific changes.
Kandinsky 6.0 has been released in Pro (29B) and Lite (3B) versions for video generation and upscaling, with support for ComfyUI and Diffusers.
A Reddit user reported that on a dual-GPU setup (R9700, RTX 5060 Ti) with 64GB DDR5 RAM, running a quantized Qwen3.8-Flash-Next via Strata boosted inference from about 21 to 60 tokens per second, but memory usage hit 96%, preventing simultaneous ComfyUI image/video workloads, highlighting high system RAM demands of local LLMs.
A developer got a 1980s Tandy 1000 to do chat and text-to-image generation by linking over WiFi to a PC server running Qwen and Krea 2, producing a thumbnail in about 9 seconds; the code is GPLv3.
That is everything