Loading…
Compact .safetensors deltas for Wan 2.2 image-to-video and Flux Klein image bases. Filter by family, weight and safety level.
This catalog tracks open-source LoRAs for two uncensored bases: Flux Klein (image) and Wan 2.2 (image-to-video). Each entry lists the family and trigger prompt so you can shortlist before generating.
A LoRA (Low-Rank Adaptation) is a compact .safetensors file that bolts onto a base checkpoint to inject a style, character, pose, or capability. LoRAs are much smaller than full checkpoints because they only encode the delta from the base, and you can stack several at different scales to compose effects without retraining the model.
No — LoRAs are only compatible with the base they were trained on. A Flux Klein LoRA will not load on Wan 2.2 (and vice versa), and neither works on stock SDXL or FLUX.1 [dev]. Each card lists the family in its badge so you can shortlist compatible weights before downloading.
Wan 2.2 LoRAs ship as a paired high-noise and low-noise file. The high file shapes coarse motion and composition during the early denoising steps; the low file refines detail and texture during late steps. Most workflows load both and let the sampler swap between them, but you can ship only the high file if you want a smaller download.
The LoRAs in this catalog are open-source weights published by their authors. Hosting and running them locally is your responsibility — verify the upstream license and your local regulations before generating, sharing or selling output. NSFW-tagged LoRAs are intended for adult use only.
Locally via ComfyUI, A1111, Forge or any inference UI that supports LoRA loading. Wan 2.2 video LoRAs typically need 12–24 GB VRAM depending on resolution and frame count; Flux Klein image LoRAs run anywhere a Flux Klein checkpoint runs (16–24 GB VRAM at FP16, 8–12 GB at FP8). Public hosted endpoints usually reject uncensored LoRAs — local generation is the standard path.
