add use_full_z_range flag, fix vae missing key warnings

README.md

@@ -14,8 +14,8 @@ tags:
 # Marigold Normals (LCM) Model Card
 
 This model belongs to the family of diffusion-based Marigold models for solving various computer vision tasks.
-The Marigold Normals model focuses on the surface normals task. 
-It takes an input image and computes surface normals in each pixel. 
+The Marigold Normals model focuses on the surface normals task.
+It takes an input image and computes surface normals in each pixel.
 The LCM stands for Latent Consistency Models, which is a technique for making the diffusion model fast.
 The Marigold Normals model is trained from Stable Diffusion with synthetic data, and the LCM model is further fine-tuned from it.
 Thanks to the rich visual knowledge stored in Stable Diffusion, Marigold models possess deep scene understanding and excel at solving computer vision tasks.

model_index.json

@@ -1,6 +1,7 @@
 {
   "_class_name":"MarigoldPipeline",
   "_diffusers_version":"0.24.0",
+  "use_full_z_range": true,
   "unet":[
      "diffusers",
      "UNet2DConditionModel"

vae/config.json

@@ -15,12 +15,16 @@
     "DownEncoderBlock2D",
     "DownEncoderBlock2D"
   ],
+  "force_upcast": true,
   "in_channels": 3,
   "latent_channels": 4,
+  "latents_mean": null,
+  "latents_std": null,
   "layers_per_block": 2,
   "norm_num_groups": 32,
   "out_channels": 3,
   "sample_size": 768,
+  "scaling_factor": 0.18215,
   "up_block_types": [
     "UpDecoderBlock2D",
     "UpDecoderBlock2D",