Dev trains RAW Upscale MOSS detail LoRA with LoRA Dataset Studio, full recipe shared
Ill-Ant-9489 · reddit · 2026-10-22
The author trained RAW Upscale MOSS, a detail/texture LoRA for MOSS-Image, using the open-source LoRA Dataset Studio. Key settings: base mossimagesftbf16, 4-step acceleration adapter at strength 1.0, 6 sampling steps, CFG 1, Euler/Simple, denoise 1.0; training ran 2,000 steps at rank 16 / alpha 16, 1024 resolution, exported as FP16 SafeTensors. Seeds and resolutions are shared for reproducibility. The LoRA is in a 9-day early-access window (10,000 Buzz) and becomes free on Oct 30, 2026.
More from Multimodal
- wavefnx tests the Moov editor: 60s training film remixed with brainrot music — wavefnx · 2026-10-22
- SMD few-step diffusion distillation: no score model or GAN, validated up to 33B, fully released — ArashVahdat · 2026-10-22
- Gandalf short film generated in ~16 minutes on a local RTX 3080 with H3 and Pinokio — cocktailpeanut · 2026-10-22
- HeyGen launches Video 1: 2K video with sound at $0.01/s in under a minute — HeyGen · 2026-10-22
- Fibo 2.0 quality degradation and missing 2K output reportedly fixed — sktksm · 2026-10-22
- Fibo 2.0 native 2K stretching fixed: Redditor shares RoPE settings for direct hi-res generation — sktksm · 2026-10-22