This repository contains selected LoRA adapters from the MinorPerception R2I diffusion experiments.
The base model weights are not included.
-
sd35_v3_qwenvl_hybrid_grpo_lora/
- Selected Stable Diffusion 3.5 LoRA from the Qwen-VL hybrid GRPO reward experiment.
- Intended base model:
stabilityai/stable-diffusion-3.5-medium.
-
flux_v3_clip_grpo_stage3init_memfix_lora/
- Selected FLUX LoRA from the stage3 warm-start, memory-fixed CLIP-GRPO experiment.
- Intended base model:
black-forest-labs/FLUX.1-dev.
These adapters were trained for R2I-style prompt-to-resolved-caption alignment experiments.
They are experimental research artifacts and should be evaluated against the exact inference scripts
and prompts used in the repository.
