ModelScope text-to-video 1.7B in diffusers: settings, long videos under 16 GB, licence

ModelScope's text-to-video model (about 1.7B parameters) has three parts: a text feature extractor, a UNet3D diffusion model in video latent space, and a decoder from latent to pixels. It accepts English prompts only and is meant for research. diffusers: pip install diffusers transformers accelerate torch; pipe = DiffusionPipeline.from_pretrained("damo-vilab/text-to-video-ms-1.7b", torch_dtype=torch.float16, variant="fp16"); pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config); pipe.enable_model_cpu_offload(); video_frames = pipe(prompt, num_inference_steps=25).frames; export_to_video(video_frames). For long videos add pipe.enable_vae_slicing() and pass num_frames=200; the card says this, with Torch 2.0 (and diffusers from source at the time), allows up to 25 seconds in under 16 GB VRAM. The card notes the MP4 plays in VLC but may not in some other players. Limitations: biases from training data (WebVid, LAION-5B, ImageNet and others, filtered by aesthetic and watermark scores and deduplicated), no film-quality output, no clear text, English only, weak compositional prompts, and no realistic depiction of people or events. The card prohibits demeaning, pornographic, violent or false content. Licence: non-commercial, but inconsistent: the metadata says CC-BY-NC-4.0 while the Model Details section links CC-BY-NC-ND 4.0, which also forbids distributing modified versions. Assume the stricter terms until clarified. Based on the ali-vilab/text-to-video-ms-1.7b card pinned to revision 8227dddca75a8561bf858d604cc5dae52b954d01 and observed on 2026-10-27; the code loads damo-vilab/text-to-video-ms-1.7b. Nothing was run here; the licence conflict is reported as found, not resolved. 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/ali-vilab/text-to-video-ms-1.7b/raw/8227dddca75a8561bf858d604cc5dae52b954d01/README.md",
      "note": "Retained source snapshot observed 2026-10-27; TEXT.txt; UTF-8 bytes 981:1230, 2333:2564, 3477:3565, 4194:4268, 4:100, 1621:1906; exact excerpt SHA256 d1d101dcfeff1780bb2514f3f3f7a05f9f88753c91bfda1e62acfd56fc1ca379,e358f6fee717f07978eadb97408f63b0cbcf01c0a6d3492b9a15bcc37a7cdfad,bdbbb8cdc6573903622833763d8e90a478024febc3957fd01f3f4a2d50e61586,5"
    }
  ]
}