Fizgig 6.5 adds Qwen Image 2.1 LoRA training and a free high-res training adapter
shootthesound · reddit · 2026-09-28
Fizgig 6.5.0 is out with Qwen Image 2.1 LoRA and LoKR training, turbo previews, and full workbench tooling, running on 10GB+ VRAM. The author retrained a Qwen Image 2.1 training adapter at higher resolution on a larger dataset, substantially improving sharpness, and released it free for any trainer on Hugging Face. The release also introduces a per-model driver system to speed up integration, with a guide coming so the community can PR models from SDXL to LTX; Qwen edit training is coming soon. Battle-tested "hit and run" training defaults are included.
More from Multimodal
- Creator shares GPT Image + Seedance 2.5 prompt combo for realistic selfie-style AI videos — techhalla · 2026-09-28
- GPT-6 Astra turns a room into Studio Ghibli in real time on Apple Vision Pro — Scobleizer · 2026-09-28
- Creator hooks Claude to MCP and lets it autonomously edit motion graphics — aziz4ai · 2026-09-28
- Atmospheric AI video demo: "Casting shadows, casting spells" — piotrbinkowski · 2026-09-28
- Cinematic techniques guide for AI video: camera, lighting, and prompts with examples — zetarediculi · 2026-09-28
- Redditor uses Claude Opus 5.5 Motion Graphics skill to build a site hype reel — VIV-AF-D · 2026-09-28