Upload folder using huggingface_hub

core/pipelines/workflow_recipes/_partials/conditioning/cosmos-predict2.yaml

@@ -0,0 +1,47 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "cosmos"
+      device: "default"
+
+connections:
+  - from: "unet_loader:0"
+    to: "ksampler:model"
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_vae_chains:
+  vae_chain:
+    targets:
+      - "vae_decode:vae"
+      - "vae_encode:vae"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"

requirements.txt

@@ -1,6 +1,6 @@
 comfyui-frontend-package==1.45.19
 comfyui-workflow-templates==0.10.0
-comfyui-embedded-docs==0.5.4
+comfyui-embedded-docs==0.5.5
 torch
 torchsde
 torchvision

yaml/constants.yaml

@@ -87,6 +87,14 @@ RESOLUTION_MAP:
     "3:4 (Classic Portrait)": [896, 1152]
     "3:2 (Photography)": [1216, 832]
     "2:3 (Photography Portrait)": [832, 1216]
+  cosmos-predict2:
+    "1:1 (Square)": [1024, 1024]
+    "16:9 (Landscape)": [1344, 768]
+    "9:16 (Portrait)": [768, 1344]
+    "4:3 (Classic)": [1152, 896]
+    "3:4 (Classic Portrait)": [896, 1152]
+    "3:2 (Photography)": [1216, 832]
+    "2:3 (Photography Portrait)": [832, 1216]
   anima:
     "1:1 (Square)": [1024, 1024]
     "16:9 (Landscape)": [1344, 768]

yaml/file_list.yaml

@@ -456,6 +456,15 @@ file:
       source: "hf"
       repo_id: "Comfy-Org/Lens"
       repository_file_path: "diffusion_models/lens_turbo_mxfp8.safetensors"
+    # Cosmos-Predict2
+    - filename: "cosmos_predict2_2B_t2i.pt"
+      source: "hf"
+      repo_id: "nvidia/Cosmos-Predict2-2B-Text2Image"
+      repository_file_path: "model.pt"
+    - filename: "cosmos_predict2_14B_t2i.pt"
+      source: "hf"
+      repo_id: "nvidia/Cosmos-Predict2-14B-Text2Image"
+      repository_file_path: "model.pt"
     # Anima
     - filename: "waiANIMA_v10.safetensors"
       source: "hf"
@@ -720,6 +729,11 @@ file:
       repo_id: "alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1"
       repository_file_path: "Z-Image-Turbo-Fun-Controlnet-Tile-2.1-8steps.safetensors"
   text_encoders: 
+    # Cosmos-Predict2
+    - filename: "oldt5_xxl_fp8_e4m3fn_scaled.safetensors"
+      source: "hf"
+      repo_id: "comfyanonymous/cosmos_1.0_text_encoder_and_VAE_ComfyUI"
+      repository_file_path: "text_encoders/oldt5_xxl_fp8_e4m3fn_scaled.safetensors"
     # Ideogram-4 & Boogu-Image
     - filename: "qwen3vl_8b_nvfp4.safetensors"
       source: "hf"
@@ -812,6 +826,10 @@ file:
       repo_id: "Comfy-Org/Omnigen2_ComfyUI_repackaged"
       repository_file_path: "split_files/text_encoders/qwen_2.5_vl_fp16.safetensors"  
   vae: 
+    - filename: "wan_2.1_vae.safetensors"
+      source: "hf"
+      repo_id: "Comfy-Org/Wan_2.1_ComfyUI_repackaged"
+      repository_file_path: "split_files/vae/wan_2.1_vae.safetensors"
     - filename: "qwen_image_vae.safetensors"
       source: "hf"
       repo_id: "Comfy-Org/Qwen-Image_ComfyUI"

yaml/image_gen_features.yaml

@@ -71,6 +71,11 @@ longcat-image:
   - conditioning
   - pid
 
+cosmos-predict2:
+  enabled_chains:
+  - conditioning
+  - vae
+
 anima:
   enabled_chains:
   - lora

yaml/model_architectures.yaml

@@ -9,6 +9,7 @@ architecture_order:
   - "Z-Image"
   - "Qwen-Image"
   - "LongCat-Image"
+  - "Cosmos-Predict2"
   - "Anima"
   - "NewBie-Image"
   - "Kandinsky-5"
@@ -56,6 +57,9 @@ architectures:
   "LongCat-Image":
     model_type: "longcat-image"
     controlnet_key: "LongCat-Image"
+  "Cosmos-Predict2":
+    model_type: "cosmos-predict2"
+    controlnet_key: "Cosmos-Predict2"
   "Anima":
     model_type: "anima"
     controlnet_key: "Anima"

yaml/model_defaults.yaml

@@ -99,6 +99,13 @@ LongCat-Image:
     sampler_name: "euler"
     scheduler: "simple"
 
+Cosmos-Predict2:
+  _defaults:
+    steps: 35
+    cfg: 4.0
+    sampler_name: "euler"
+    scheduler: "karras"
+
 Anima:
   _defaults:
     steps: 30

yaml/model_list.yaml

@@ -126,6 +126,19 @@ Checkpoint:
           unet: "longcat_image_bf16.safetensors"
           vae: "ae.safetensors"
           clip: "qwen_2.5_vl_7b_nvfp4.safetensors"
+  Cosmos-Predict2:
+    latent_type: sd3_latent
+    models:
+      - display_name: "Cosmos-Predict2-2B-T2I"
+        components:
+          unet: "cosmos_predict2_2B_t2i.pt"
+          clip: "oldt5_xxl_fp8_e4m3fn_scaled.safetensors"
+          vae: "wan_2.1_vae.safetensors"
+      - display_name: "Cosmos-Predict2-14B-T2I"
+        components:
+          unet: "cosmos_predict2_14B_t2i.pt"
+          clip: "oldt5_xxl_fp8_e4m3fn_scaled.safetensors"
+          vae: "wan_2.1_vae.safetensors"
   Anima:
     latent_type: latent
     models: