MOSS Image, a 5B diffusion transformer from OpenMOSS, launched under an Apache 2.0 license with weights on Hugging Face. Its distinguishing feature is JSON-driven layout control for precise text placement in generated images. A YouTube review by Sam Witteveen demonstrates text rendering, editing, inpainting, and LoRA workflows.
A YouTube video by Sam Witteveen reviews BreezeTTS2, a 3B-parameter open-weights text-to-speech model from BreezeBlue that runs 100% locally in real time. The video walks through the accompanying blog post and demonstrates voice design and voice direction features. For practitioners, it highlights a fully local real-time TTS option with weights available on Hugging Face.
A YouTube video by Sam Witteveen reviews BreezeTTS2, a 3B-parameter open-weights text-to-speech model from BreezeBlue that runs 100% locally in real time. The video walks through the accompanying blog post and demonstrates voice design and voice direction features. For practitioners, it highlights a fully local real-time TTS option with weights available on Hugging Face.
A YouTube walkthrough by Sam Witteveen covers two open image decision models, ImaJev 4B and Jev-Omni, aimed at RPA decision tasks like checking screenshots, photos, and forms. The video explains why prior custom deep learning approaches fell short and demonstrates a Form Decision Inspector with confidence thresholds per question. Both models are available on Hugging Face, making them practical options for developers adding decision logic to automation pipelines.
A YouTube walkthrough by Sam Witteveen covers two open image decision models, ImaJev 4B and Jev-Omni, aimed at RPA decision tasks like checking screenshots, photos, and forms. The video explains why prior custom deep learning approaches fell short and demonstrates a Form Decision Inspector with confidence thresholds per question. Both models are available on Hugging Face, making them practical options for developers adding decision logic to automation pipelines.
A YouTube video by Sam Witteveen reviews Holo4, H Company's open-weight model family that operates a GUI, writes and runs code, and calls MCP tools within one agent loop. The video links H Company's blog and Hugging Face weights, showcasing the release to developers.
A YouTube video by Sam Witteveen reviews Holo4, H Company's open-weight model family that operates a GUI, writes and runs code, and calls MCP tools within one agent loop. The video links H Company's blog and Hugging Face weights, showcasing the release to developers.
A new arXiv paper introduces Autoregressive Mosaics (AM-Bench), a benchmark testing whether text-only LLMs that generate image-drawing code possess internal 2D spatial representations or merely translate verbal descriptions into code. It separates these factors via translation and layout tasks, giving practitioners a tool to measure genuine spatial reasoning.
A new arXiv paper introduces Autoregressive Mosaics (AM-Bench), a benchmark testing whether text-only LLMs that generate image-drawing code possess internal 2D spatial representations or merely translate verbal descriptions into code. It separates these factors via translation and layout tasks, giving practitioners a tool to measure genuine spatial reasoning.
A new method addresses the lack of composition control in text-to-image models like Stable Diffusion, allowing users to specify pose, camera angle, and composition instead of the model choosing them arbitrarily.
A new method addresses the lack of composition control in text-to-image models like Stable Diffusion, allowing users to specify pose, camera angle, and composition instead of the model choosing them arbitrarily.