New! Added quantized versions!

There is:

Q5_1, 5-bit round-to-nearest quantization. Small but still ok(ish) quality.

Q2_K, 2-bit quantization. Very small. Quality...not that good, but useable?

https://github.com/leejet/stable-diffusion.cpp/blob/master/docs/quantization_and_gguf.md

This is a small test of the new functionality in RuinedFooocus that let you create .merge-files and use them as models. Instead of keeping dozens of models you can have a small file with the recipe for the merge and it will merge the model for you on-the-fly when it is needed.

{
  "comment": "Top 5 models merge 2024-07-06",
  "base": {
    "name": "cyberrealisticXL_v21.safetensors",
    "weight": 2742
  },
  "models": [
    {
      "name": "halcyonSDXL_v17.safetensors",
      "weight": 1620
    },
    {
      "name": "corEpicaByZovya_v0Beta.safetensors",
      "weight": 483
    },
    {
      "name": "eldritchPhotography_v10.safetensors",
      "weight": 369
    },
    {
      "name": "enjoyXLSuperRealistic_v30ModifiedVersion.safetensors",
      "weight": 933
    }
  ],
  "normalize": 1.0,
  "cache": false
}

Don't take this model too serious. It is simply a test-merge of the top 5 models, weighted according to the number of downloads they've had.