Krea2 LoRA training experiment boosts detail with high-res data and low-noise steps
Jolly-Rip5973 · reddit · 2026-07-27
The author experiments with training Krea2 LoRAs using a resolution-heavy setup: 1536, 1280, 1024, and 768 training sizes only. They train the first 750 steps with Adam and weighted high-noise, then switch to weighted low-noise for the rest of the run.
The goal is to push fine detail by emphasizing higher-resolution data and low-noise training, which is supposed to preserve more detail. The result at 2,000 steps is described as having "crazy detail," and the author says they will continue to 3,250 steps while saving checkpoints.
More from Multimodal
- LTX 2.3 close-up video looks sharper at 1080p, but a 6-second clip takes 12–15 minutes on an RTX 3060 — iiTzMYUNG · 2026-07-27
- Opus 5 one-shot a Paper Mario–style game prototype in a single run — Acid_God_ · 2026-07-27
- Runway demo turns a snowfall dance into a cinematic magical-realist video — LudovicCreator · 2026-07-27
- A Midjourney origami prompt turns any subject into a silhouette test — tisch_eins · 2026-07-27
- Midjourney 8.2 is said to outperform Leonardo AI’s Pro Upscaler — aziz4ai · 2026-07-27
- Prototype video model uses Wan 2.1 1.3B with 16x spatial and 8x temporal compression — ostrisai · 2026-07-27