dreamshaper8-lcm-gpu

DreamShaper 8 with the LCM-LoRA fused into its weights, for real-time on-device generation in Latent Studio. Sample with the LCM scheduler, 4 steps, guidance 1.

Converted to MNN for mobile GPUs (OpenCL/Vulkan), fp16, 512x512. The UNet is split into three graphs (unet_a/b/c.mnn) to stay inside mobile GPU buffer limits. Four steps per image; runs on Adreno and Mali GPUs.

License

Derived from DreamShaper 8 (Lykon) and Stable Diffusion 1.5 under CreativeML OpenRAIL-M; the Attachment A use-restrictions apply. LCM-LoRA (latent-consistency/lcm-lora-sdv1-5) is OpenRAIL++.