Upload folder using huggingface_hub

README.md

@@ -74,6 +74,7 @@ models:
   - InstantX/Qwen-Image-ControlNet-Inpainting
   - InstantX/Qwen-Image-ControlNet-Union
   - InstantX/SD3.5-Large-IP-Adapter
+  - kandinskylab/Kandinsky-5.0-T2I-Lite
   - Kijai/flux-fp8
   - kohya-ss/Anima-LLLite
   - Laxhar/noob_openpose

core/pipelines/workflow_recipes/_partials/conditioning/kandinsky-5.yaml

@@ -0,0 +1,72 @@
+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: DualCLIPLoader
+    title: "DualCLIPLoader"
+    params:
+      type: "kandinsky5_image"
+      device: "default"
+  model_sampler:
+    class_type: ModelSamplingSD3
+    title: "ModelSamplingSD3"
+    params:
+      shift: 3.0
+
+connections:
+  - from: "unet_loader:0"
+    to: "model_sampler:model"
+  - from: "model_sampler:0"
+    to: "ksampler:model"
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    output_map:
+      "unet_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["model_sampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_controlnet_chains:
+  controlnet_chain:
+    template: "ControlNetApplyAdvanced"
+    ksampler_node: "ksampler"
+    vae_source: "vae_loader:0"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+dynamic_pid_chains:
+  pid_chain:
+    ksampler_node: "ksampler"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip1_name: "clip_loader:clip_name1"
+  clip2_name: "clip_loader:clip_name2"

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]
+  kandinsky-5:
+    "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]
   omnigen2:
     "1:1 (Square)": [1024, 1024]
     "16:9 (Landscape)": [1344, 768]

yaml/file_list.yaml

@@ -536,6 +536,11 @@ file:
       source: "hf"
       repo_id: "Comfy-Org/Qwen-Image_ComfyUI"
       repository_file_path: "split_files/diffusion_models/qwen_image_fp8_e4m3fn.safetensors"
+    # Kandinsky-5
+    - filename: "kandinsky5lite_t2i.safetensors"
+      source: "hf"
+      repo_id: "kandinskylab/Kandinsky-5.0-T2I-Lite"
+      repository_file_path: "model/kandinsky5lite_t2i.safetensors"
     # Flux.1
     - filename: "flux1-dev-fp8-e4m3fn.safetensors"
       source: "hf"
@@ -796,10 +801,10 @@ file:
       source: "hf"
       repo_id: "Comfy-Org/HiDream-I1_ComfyUI"
       repository_file_path: "split_files/text_encoders/llama_3.1_8b_instruct_fp8_scaled.safetensors"
-    - filename: "qwen_2.5_vl_7b_fp8_scaled.safetensors"
+    - filename: "qwen_2.5_vl_7b_nvfp4.safetensors"
       source: "hf"
       repo_id: "Comfy-Org/Qwen-Image_ComfyUI"
-      repository_file_path: "split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors"
+      repository_file_path: "split_files/text_encoders/qwen_2.5_vl_7b_nvfp4.safetensors"
     - filename: "byt5_small_glyphxl_fp16.safetensors"
       source: "hf"
       repo_id: "Comfy-Org/HunyuanImage_2.1_ComfyUI"

yaml/image_gen_features.yaml

@@ -141,6 +141,12 @@ ovis-image:
   - vae
   - pid
 
+kandinsky-5:
+  enabled_chains:
+  - conditioning
+  - vae
+  - pid
+
 chroma1:
   enabled_chains:
   - conditioning

yaml/model_architectures.yaml

@@ -10,6 +10,7 @@ architecture_order:
   - "LongCat-Image"
   - "Anima"
   - "NewBie-Image"
+  - "Kandinsky-5"
   - "Ovis-Image"
   - "HunyuanImage"
   - "Chroma1-Radiance"
@@ -66,6 +67,9 @@ architectures:
   "Lumina":
     model_type: "lumina"
     controlnet_key: "Lumina"
+  "Kandinsky-5":
+    model_type: "kandinsky-5"
+    controlnet_key: "Kandinsky-5"
   "Ovis-Image":
     model_type: "ovis-image"
     controlnet_key: "Ovis-Image"

yaml/model_defaults.yaml

@@ -105,6 +105,13 @@ NewBie-Image:
     positive_prompt: "You are an assistant designed to generate high-quality anime images with the highest degree of image-text alignment based on xml format textual prompts. <Prompt Start>"
     negative_prompt: "You are an assistant designed to generate low-quality images based on textual prompts. <Prompt Start>"
 
+Kandinsky-5:
+  _defaults:
+    steps: 50
+    cfg: 3.5
+    sampler_name: "euler"
+    scheduler: "simple"
+
 Ovis-Image:
   _defaults:
     steps: 20

yaml/model_list.yaml

@@ -98,13 +98,13 @@ Checkpoint:
         components:
           unet: "qwen_image_2512_fp8_e4m3fn.safetensors"
           vae: "qwen_image_vae.safetensors"
-          clip: "qwen_2.5_vl_7b_fp8_scaled.safetensors"
+          clip: "qwen_2.5_vl_7b_nvfp4.safetensors"
           lora: "Qwen-Image-2512-Lightning-4steps-V1.0-bf16.safetensors"
       - display_name: "Qwen/Qwen-Image + Lightning-4steps-V1.0 LoRA"
         components:
           unet: "qwen_image_fp8_e4m3fn.safetensors"
           vae: "qwen_image_vae.safetensors"
-          clip: "qwen_2.5_vl_7b_fp8_scaled.safetensors"
+          clip: "qwen_2.5_vl_7b_nvfp4.safetensors"
           lora: "Qwen-Image-fp8-e4m3fn-Lightning-4steps-V1.0-bf16.safetensors"
   LongCat-Image:
     latent_type: sd3_latent
@@ -113,7 +113,7 @@ Checkpoint:
         components:
           unet: "longcat_image_bf16.safetensors"
           vae: "ae.safetensors"
-          clip: "qwen_2.5_vl_7b_fp8_scaled.safetensors"
+          clip: "qwen_2.5_vl_7b_nvfp4.safetensors"
   Anima:
     latent_type: latent
     models:
@@ -146,6 +146,15 @@ Checkpoint:
           vae: "ae.safetensors"
           clip1: "gemma_3_4b_it_bf16.safetensors"
           clip2: "jina_clip_v2_bf16.safetensors"
+  Kandinsky-5:
+    latent_type: latent
+    models:
+      - display_name: "Kandinsky-5.0-T2I-Lite"
+        components:
+          unet: "kandinsky5lite_t2i.safetensors"
+          vae: "ae.safetensors"
+          clip1: "qwen_2.5_vl_7b_nvfp4.safetensors"
+          clip2: "clip_l.safetensors"
   Ovis-Image:
     latent_type: sd3_latent
     models:
@@ -161,13 +170,13 @@ Checkpoint:
         components:
           unet: "hunyuanimage2.1_fp8_e4m3fn.safetensors"
           vae: "hunyuan_image_2.1_vae_fp16.safetensors"
-          clip1: "qwen_2.5_vl_7b_fp8_scaled.safetensors"
+          clip1: "qwen_2.5_vl_7b_nvfp4.safetensors"
           clip2: "byt5_small_glyphxl_fp16.safetensors"
       - display_name: "HunyuanImage-2.1-Distilled"
         components:
           unet: "hunyuanimage2.1_distilled_fp8_e4m3fn.safetensors"
           vae: "hunyuan_image_2.1_vae_fp16.safetensors"
-          clip1: "qwen_2.5_vl_7b_fp8_scaled.safetensors"
+          clip1: "qwen_2.5_vl_7b_nvfp4.safetensors"
           clip2: "byt5_small_glyphxl_fp16.safetensors"
   Chroma1-Radiance:
     latent_type: chroma_radiance_latent

yaml/pid.yaml

@@ -7,7 +7,7 @@ PiD:
     architectures: ["anima", "qwen-image"]
   - filepath: "pid_flux1_1024_to_4096_4step_mxfp8.safetensors"
     latent_format: "flux"
-    architectures: ["longcat-image", "newbie-image", "z-image", "ovis-image", "omnigen2", "chroma1", "hidream-i1", "flux1"]
+    architectures: ["longcat-image", "newbie-image", "z-image", "kandinsky-5", "ovis-image", "omnigen2", "chroma1", "hidream-i1", "flux1"]
   - filepath: "pid_sd3_1024_to_4096_4step_bf16.safetensors"
     latent_format: "sd3"
     architectures: ["sd35"]