Benchmarking Krea2 LoRA Training on RTX 5070 Ti: Done in Under 3 Hours
y3kdhmbdb2ch2fc6vpm2 · reddit · 2026-08-03
A developer shared their hands-on experience training a Krea2 LoRA using an RTX 5070 Ti GPU on Reddit.
- Setup: RTX 5070 Ti, 32GB RAM, NVMe SSD, utilizing the OneTrainer framework.
- Training Details: Trained on a dataset of 63 images at 1024 resolution (1 MP). With an offload set to 0.5, the speed was roughly 2.5 seconds per iteration, completing full training in about 2.5 to 3 hours.
- Parameters: Following kohya's documentation, the timestep shift was set to 2.5 for 1024 resolution, which yielded solid results. All samples were generated at 2 MP.
The author noted that training Krea2 is highly enjoyable and provided a link to the CivitAI model page.
Related event: RTX 5070 Ti Tested for Efficient LoRA Training(2 posts)→
More from Multimodal
- ForcingBooster halves computation of autoregressive video diffusion with patch-wise denoising — VAST-AI · 2026-10-16
- Tokyo's Gemini AI Photo Booth turns visitors into plush toys and dioramas — bradneuberg · 2026-10-16
- Orchestrating a 100-image portrait of humanity for aliens with ChatGPT, Argon and p5.brush — dx8xb · 2026-10-16
- Indie Creator Builds an AI-Native Animated Show Solo — ComprehensiveLime654 · 2026-10-16
- Ideogram teases 4.5 arrival and open weights: 'Not yet, but soon' — East_Shoe8814 · 2026-10-16
- Opus 5.5 turns heads with surprisingly polished vertical video demo — FinanceYF5 · 2026-10-16