Sam Witteveen · 2026-10-26 · notable
Sam Witteveen — '100% Open Image Model Lets You Place Text With JSON'
Sam Witteveen tests MOSS-Image, OpenMOSS's 5B Apache-2.0 diffusion transformer that lays out images from JSON, covering text rendering, style control, editing, inpainting and LoRA fine-tuning.
Sam Witteveen tries MOSS-Image, an open image model where a JSON layout decides where text and objects go.
What is it?
'100% Open Image Model Lets You Place Text With JSON' went up on Sam Witteveen's channel on 26 October 2026. The video covers MOSS-Image from the OpenMOSS team, a 5-billion-parameter diffusion transformer released under the Apache 2.0 license.
How does it work?
Chapters walk through where the model runs (Hugging Face, ComfyUI, SGLang and vLLM-Omni), overall image quality, lighting and style control, and long-text rendering benchmarks. The core demo uses JSON to place each element on the canvas, then edits layouts and styles, inpaints, and shows an 8-step LoRA plus custom LoRA training.
Why does it matter?
Placing exact text in a generated image is still a weak spot for most image models. A permissive open model with JSON layout control lets designers and app builders script posters, slides and UI mockups locally, and the video also covers its limits.
Who is it for?
developers building image-generation pipelines