Surprising result: character LoRA trained on the SFT model gives the best likeness
1or4s · reddit · 2026-10-26
A Reddit user shares a counterintuitive finding: they've been training a character LoRA on the SFT (instruction-tuned) version of Moss rather than the base model, and the character likeness is the best of any model they've tried. Noting that others training on base struggle with likeness, they ask whether training on SFT might actually be the better approach for character models.
More from Multimodal
- User runs Google Gemini AI video generation on Windows XP via Supermium browser — MussleGeeYem · 2026-10-26
- Kimi's 3D Make turns a sketch into an editable, 3D-printable model — sven_ai · 2026-10-26
- SoundBit: open-source chiptune workstation with a built-in MCP server for live AI composing — RenasDev · 2026-10-26
- Matthew Berman: Image, video and voice gen are now good enough to run locally — MatthewBerman · 2026-10-26
- FOSS local-first OmniImage organizes 40k+ ComfyUI outputs with offline CLIP search — MatheusWMac · 2026-10-26
- Malfoid, a Harry Potter AI film, hailed as the best AI movie by traditional filmmaking standards — granawkins · 2026-10-26