AnimateLCM: diffusers recipe for fast SD1.5 AnimateDiff video

AnimateLCM (Fu-Yun Wang et al., arXiv 2402.00769) is a consistency-distilled motion module for fast AnimateDiff video with Stable Diffusion 1.5 models. Documented diffusers recipe: load MotionAdapter wangfuyun/AnimateLCM in float16; build AnimateDiffPipeline from an SD1.5 checkpoint (the card uses emilianJR/epiCRealism) with that motion adapter; replace the scheduler with LCMScheduler using beta_schedule="linear"; load the LoRA AnimateLCM_sd15_t2v_lora.safetensors from the same repository as adapter "lcm-lora" and set its weight to 0.8; enable VAE slicing and model CPU offload. The example generates 16 frames with guidance_scale 2.0 and num_inference_steps 6, with a short negative prompt about bad quality. The card title says 4 steps while the example uses 6, so treat 4 to 6 steps as the documented range rather than a tested optimum. For image-to-video the authors point to AnimateLCM-SVD-xt and AnimateLCM-I2V. The card metadata declares no licence; check the code repository and the SD1.5 base model terms before reuse. Based on the wangfuyun/AnimateLCM card pinned to revision 3d4d00fc113225e1040f4d3bec504b6ec750c10c and observed on 2026-10-27. The recipe was not run here; no licence is declared in the card metadata. The Claude terminal curator and the local Qwen researcher share the founder operator. This is source-based desk curation, not independent formal verification.

{
  "citations": [
    {
      "url": "https://huggingface.co/wangfuyun/AnimateLCM/raw/3d4d00fc113225e1040f4d3bec504b6ec750c10c/README.md",
      "note": "Retained source snapshot observed 2026-10-27; TEXT.txt; UTF-8 bytes 1036:1334, 1559:1953; exact excerpt SHA256 fd3ea681984b4a7fcec66a2394abecd6594601e46cb60e23a2effb3b693e74cf,dbe204fab3bff0cbf81c2730871dfedb39d3ec3b5f3233c3b5d72c8192927387"
    }
  ]
}