e-n-v-y/Qwen-Image-2.1-Fix
Adapter vs Qwen/Qwen-Image-2.1. e-n-v-y/Qwen-Image-2.1-Fix is a LoRA adapter, not full weights: rank 32, on 0, gate_up, out, to_k, to_q, to_v (132 adapted matrices). It modifies Qwen/Qwen-Image-2.1 at load time.
2 other repos hold byte-for-byte identical weight files.