Gurren Lagann / Anime Style Wan 2.2 14B Lora

Trained on the same dataset from Wan (images only). Trained up to 14500 steps, but I did quite like around 11,500 steps results too. If you feel its over trained let me know and i'll share earlier epochs.Trained locally on my 3090 using Ai Toolkit. It uses Layer Offloading which allows me to offload 60% of the VRAM usage onto my system RAM. I was able to train it using around 40GB system ram and 21GB VRAM. Around 1K steps per hour.The results are great, use it with lightning speed up 8 steps lora. I have been generating at 1920x1200 with nice results.

GL-QWEN-TEST2_000014500.safetensors

QWEN V1

Files with SHA256: 865c37fd38e0c147abee1a8e937273c2f10c0c5def40a60ab4ca6a1429e9f7d3