Exploring Workflows for Iterative AI-Generated LoRA Training Datasets

Sudden-Complaint7037 · reddit · 2026-08-04

With image generation models becoming more polished, the author wants to upgrade their LoRA training datasets without the hassle of a photo studio.

They explore using a single high-quality photo iteratively via AI to create a multi-angle dataset. Avoiding cloud services like Nano Banana for privacy reasons, they found local tools like Flux Klein 9b ineffective, and Qwen-Image-Edit with BFS LoRA could only copy-paste heads without generating new angles. They are asking the community for better local models or workflows.

Related event: Generate High-Quality LoRA Datasets from a Single Photo with AI(2 posts)→

Original post →

More from Multimodal

Multimodal channel →