Qwen-Image-2512 LoRA training issues: higher weights ignore prompts
Kind-Illustrator6341 · reddit · 2026-08-16
A user faces issues training character LoRAs on Qwen-Image-2512. Low step counts lack likeness, while higher steps cause the model to stop responding to prompts above specific weights (0.3 or 0.5). At weight 1.0, likeness is good but prompts are completely ignored. Solutions are requested.
More from Multimodal
- Grok Imagine generates incredibly smooth video transitions with tweaked prompts — umesh_ai · 2026-08-16
- ByteDance Releases Trace Anything for 4D Video Representation via Trajectory Fields — tom_doerr · 2026-08-16
- 30-Second French Comedy Sketch Generated with Seedance 2.5 — azed_ai · 2026-08-16
- AI Generated Resident Evil 1 Style Gameplay Demo — tmk_lmsd · 2026-08-16
- Wan2gp Anime Video Generation Demo — tmk_lmsd · 2026-08-16
- 60-Year-Old Animator Creates 20-Minute Cyberpunk Anime Using AI — SimplyAnnisa · 2026-08-16