Community image model Kromoss ships refined updates trained from GLM-caption + bbox layout SFT
mikemend · reddit · 2026-11-08
New Kromoss models by lodestones have been released on Hugging Face, described as further refined and optimized versions of the previous iteration.
The update includes an OPD-priming intermediate checkpoint (update 22000), continuing from v2's update 11000 on top of the v3 GLM-caption + bbox layout SFT, with ComfyUI-ready bf16 and int8 convrot weights. Details are limited to the author's release notes.
More from Multimodal
- Builder Uses Agents to QA Every Shot, Reworking Composition and Lighting With Measurable Metrics — DavidmComfort · 2026-11-08
- Surreal AI Art Showcase: Holographic Figures and a Rainbow-Haloed Spirit Bird — OVolosin82152 · 2026-11-08
- AI video drops Malfoy into a Friends-style sitcom in viral meme recreation — Riddy_93 · 2026-11-08
- AI-powered cutscene remaster revives Indiana Jones and the Fate of Atlantis (1992) — Megatronagaming · 2026-11-08
- AI-Generated Rap Video 'Chrome Chainz' Shows Off Stunning Robot Baller Visuals — BotLordBotflix · 2026-11-08
- Qwen Image Edit users report weird criss-cross skin artifacts in one-image edit workflows — AkaToraX · 2026-11-08