Questions

Docs · Questions

StreamDiffusionTDUsage

How do I download models for StreamDiffusionTD? Should I use the built-in HuggingFace downloader?

GPU and suggested starting settings - use this guide for current configuration advice.

StreamDiffusionTD Model Download Methods

Use the model ID directly in the “Model Id” parameter:

  • stabilityai/sd-turbo (default)
  • stabilityai/sdxl-turbo (v0.3.0+ only)
  • prompthero/openjourney-v4
  • Any HuggingFace model ID

The operator downloads and caches the model automatically on first use.

Method 2: Direct .safetensors File

  1. Download .safetensors file from HuggingFace or CivitAI
  2. Place anywhere on your system
  3. Use full file path in “Model Id” parameter
    • Example: C:\Models\sd-turbo.safetensors

Method 3: Diffusers Format Folder

  1. Download model in diffusers format (folder with unet, vae, tokenizer subfolders)
  2. Use full folder path in “Model Id”
    • Example: F:\StreamDiffusion\models\Model\stable-diffusion-v1-5

⚠️ DEPRECATED: Built-in HuggingFace Downloader

DO NOT USE the old “Hugging Face Model Downloader” option (v0.1.5-v0.2.x).

  • No longer recommended
  • Can cause incomplete downloads
  • Removed in v0.3.0

Model Compatibility - Important!

Not all models work with StreamDiffusionTD!

Model Type Works? Notes
sd-turbo ✅ Yes Best choice
sdxl-turbo ✅ Yes v0.3.0+, memory use depends on configuration; see the GPU guide
SD 1.5 + LCM-LoRA ✅ Yes Need matching LCM-LoRA
Flux ❌ No Not realtime-capable
Video models ❌ No Use ComfyTD instead

See FAQ “Model Compatibility” for full guide.

LoRA Compatibility Warning

sd-turbo is SD 2.1 based - SD 1.5 LoRAs won’t work with it!

Model LoRA Version Needed
sd-turbo SD 2.1 LoRAs (rare)
SD 1.5 models SD 1.5 LoRAs
SDXL models SDXL LoRAs

Git LFS Requirement

If you manually clone models from HuggingFace:

  1. Install Git LFS: https://git-lfs.github.com/
  2. Run: git lfs install
  3. Then clone the model repo

Error if missing: no file named diffusion_pytorch_model.bin found

Working Models List (Auto-Populated)

After successful streaming (200+ frames), models are saved to:

StreamDiffusion/streamdiffusionTD/working_models.json

These appear in the “My Models” dropdown for quick access.

Offline Usage

To run offline:

  1. Run the model once while online (downloads configs)
  2. Models are cached at: ~/.cache/huggingface/
  3. After first successful run, works offline

First run requires internet - even local .safetensors files need config downloads.


Updated: November 2025 - Added model compatibility info and LoRA warning