Add Mage-Flow architecture model.

README.md

@@ -52,6 +52,7 @@ models:
   - Comfy-Org/Lens
   - Comfy-Org/LongCat-Image
   - Comfy-Org/Lumina_Image_2.0_Repackaged
+  - Comfy-Org/Mage-Flow
   - Comfy-Org/NewBie-image-Exp0.1_repackaged
   - Comfy-Org/Omnigen2_ComfyUI_repackaged
   - Comfy-Org/Ovis-Image
@@ -144,6 +145,10 @@ models:
   - meituan-longcat/LongCat-Image
   - microsoft/Lens
   - microsoft/Lens-Turbo
+  - microsoft/Mage-Flow
+  - microsoft/Mage-Flow-Edit
+  - microsoft/Mage-Flow-Edit-Turbo
+  - microsoft/Mage-Flow-Turbo
   - NewBie-AI/NewBie-image-Exp0.1
   - nvidia/PiD
   - nvidia/PixelDiT-1300M-1024px

chain_injectors/reference_image_injector.py

@@ -0,0 +1,64 @@
+import os
+
+def inject(assembler, chain_definition, chain_items):
+    if not chain_items:
+        return
+
+    valid_images = []
+    for item in chain_items:
+        if not item:
+            continue
+        img_path = item
+        if isinstance(item, dict):
+            img_path = item.get('image') or item.get('filename') or item.get('path')
+        if img_path:
+            valid_images.append(img_path)
+
+    if not valid_images:
+        return
+
+    text_encode_name = chain_definition.get('text_encode_node')
+    text_encode_id = None
+    if text_encode_name and text_encode_name in assembler.node_map:
+        text_encode_id = assembler.node_map[text_encode_name]
+    else:
+        for node_id, node in assembler.workflow.items():
+            if isinstance(node, dict) and node.get('class_type') == 'TextEncodeMageFlowEdit':
+                text_encode_id = node_id
+                break
+
+    if not text_encode_id or text_encode_id not in assembler.workflow:
+        print("Warning: TextEncodeMageFlowEdit node not found for Reference Image chain. Skipping.")
+        return
+
+    vae_node_name = chain_definition.get('vae_node', 'vae_loader')
+    vae_node_id = assembler.node_map.get(vae_node_name)
+    if not vae_node_id:
+        for node_id, node in assembler.workflow.items():
+            if isinstance(node, dict) and node.get('class_type') == 'VAELoader':
+                vae_node_id = node_id
+                break
+
+    if vae_node_id:
+        assembler.workflow[text_encode_id]['inputs']['vae'] = [vae_node_id, 0]
+
+    for i, img_filename in enumerate(valid_images):
+        load_id = assembler._get_unique_id()
+        load_node = assembler._get_node_template("LoadImage")
+        load_node['inputs']['image'] = img_filename
+        load_node['_meta']['title'] = f"Load Reference Image {i+1}"
+        assembler.workflow[load_id] = load_node
+
+        scale_id = assembler._get_unique_id()
+        scale_node = assembler._get_node_template("ImageScaleToTotalPixels")
+        scale_node['inputs']['megapixels'] = 1.0
+        scale_node['inputs']['upscale_method'] = "nearest-exact"
+        scale_node['inputs']['resolution_steps'] = 1
+        scale_node['inputs']['image'] = [load_id, 0]
+        scale_node['_meta']['title'] = f"Scale Reference {i+1}"
+        assembler.workflow[scale_id] = scale_node
+
+        input_key = f"images.image_{i+1}"
+        assembler.workflow[text_encode_id]['inputs'][input_key] = [scale_id, 0]
+
+    print(f"Reference Image injector applied. Injected {len(valid_images)} reference images to TextEncodeMageFlowEdit node '{text_encode_id}'.")

core/pipelines/pipeline_input_processor.py

@@ -351,6 +351,17 @@ def process_pipeline_inputs(ui_inputs: Dict[str, Any], progress: gr.Progress, wo
                 temp_files_to_clean.append(temp_path)
                 active_joyai_reference.append(os.path.basename(temp_path))
 
+    reference_image_data = ui_inputs.get('reference_image_data', [])
+    active_reference_images = []
+    if reference_image_data:
+        for img in reference_image_data:
+            if img:
+                if not os.path.exists(INPUT_DIR): os.makedirs(INPUT_DIR)
+                temp_path = os.path.join(INPUT_DIR, f"temp_ref_img_{random.randint(1000, 9999)}.png")
+                img.save(temp_path, "PNG")
+                temp_files_to_clean.append(temp_path)
+                active_reference_images.append(os.path.basename(temp_path))
+
     vae_source = ui_inputs.get('vae_source')
     vae_id = ui_inputs.get('vae_id')
     vae_name_override = None
@@ -390,6 +401,7 @@ def process_pipeline_inputs(ui_inputs: Dict[str, Any], progress: gr.Progress, wo
         "active_reference_latents": active_reference_latents,
         "active_hidream_o1_reference": active_hidream_o1_reference,
         "active_joyai_reference": active_joyai_reference,
+        "active_reference_images": active_reference_images,
         "active_conditioning": active_conditioning,
         "temp_files_to_clean": temp_files_to_clean
     }

core/pipelines/sd_image_pipeline.py

@@ -117,6 +117,7 @@ class SdImagePipeline(BasePipeline):
             active_reference_latents = processed["active_reference_latents"]
             active_hidream_o1_reference = processed["active_hidream_o1_reference"]
             active_joyai_reference = processed.get("active_joyai_reference", [])
+            active_reference_images = processed.get("active_reference_images", [])
             active_conditioning = processed["active_conditioning"]
 
             loras_string = f"LoRAs: [{', '.join(active_loras_for_meta)}]" if active_loras_for_meta else ""
@@ -174,6 +175,7 @@ class SdImagePipeline(BasePipeline):
                 "reference_latent_chain": active_reference_latents,
                 "hidream_o1_reference_chain": active_hidream_o1_reference,
                 "joyai_reference_chain": active_joyai_reference,
+                "reference_image_chain": active_reference_images,
                 "vae_chain": [ui_inputs.get('vae_name')] if ui_inputs.get('vae_name') else [],
                 "hidream_o1_smoothing_chain": hidream_o1_smoothing_data,
                 "pid_chain": [ui_inputs.get('pid_settings', 'OFF')] if is_pid_enabled else [],

core/pipelines/workflow_executor.py

@@ -42,6 +42,30 @@ class WorkflowExecutor:
             
             try:
                 sorted_node_ids = WorkflowExecutor.topological_sort(workflow)
+                
+                final_node_id = None
+                for node_id in reversed(sorted_node_ids):
+                    if workflow[node_id].get('class_type') == 'SaveImage':
+                        final_node_id = node_id
+                        break
+                
+                if final_node_id:
+                    required_nodes = set()
+                    nodes_to_visit = [final_node_id]
+                    while nodes_to_visit:
+                        curr_id = nodes_to_visit.pop()
+                        if curr_id in required_nodes:
+                            continue
+                        required_nodes.add(curr_id)
+                        curr_info = workflow.get(curr_id, {})
+                        for input_val in curr_info.get('inputs', {}).values():
+                            if isinstance(input_val, list) and len(input_val) == 2 and isinstance(input_val[0], str):
+                                src_id = input_val[0]
+                                if src_id in workflow and src_id not in required_nodes:
+                                    nodes_to_visit.append(src_id)
+                    
+                    sorted_node_ids = [nid for nid in sorted_node_ids if nid in required_nodes]
+
                 print(f"--- [Workflow Executor] Execution order: {sorted_node_ids}")
             except RuntimeError as e:
                 print("--- [Workflow Executor] ERROR: Failed to sort workflow. Dumping graph details. ---")

core/pipelines/workflow_recipes/_partials/_base_sampler_sd.yaml

@@ -1,10 +1,4 @@
 nodes:
-  pos_prompt:
-    class_type: CLIPTextEncode
-    title: "CLIP Text Encode (Positive)"
-  neg_prompt:
-    class_type: CLIPTextEncode
-    title: "CLIP Text Encode (Negative)"
   ksampler:
     class_type: KSampler
     title: "KSampler"
@@ -25,8 +19,6 @@ connections:
     to: "save_image:images"
 
 ui_map:
-  positive_prompt: "pos_prompt:text"
-  negative_prompt: "neg_prompt:text"
   seed: "ksampler:seed"
   steps: "ksampler:steps"
   cfg: "ksampler:cfg"

core/pipelines/workflow_recipes/_partials/conditioning/anima.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -57,6 +63,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/auraflow.yaml

@@ -1,48 +1,56 @@
-nodes:
-  ckpt_loader:
-    class_type: CheckpointLoaderSimple
-    title: "Load Checkpoint"
-
-connections:
-  - from: "ckpt_loader:0"
-    to: "ksampler:model"
-  - from: "ckpt_loader:1"
-    to: "pos_prompt:clip"
-  - from: "ckpt_loader:1"
-    to: "neg_prompt:clip"
-  - from: "pos_prompt:0"
-    to: "ksampler:positive"
-  - from: "neg_prompt:0"
-    to: "ksampler:negative"
-  - from: "ckpt_loader:2"
-    to: "vae_decode:vae"
-  - from: "ckpt_loader:2"
-    to: "vae_encode:vae"
-
-dynamic_vae_chains:
-  vae_chain:
-    targets:
-      - "vae_decode:vae"
-      - "vae_encode:vae"
-
-dynamic_lora_chains:
-  lora_chain:
-    template: "LoraLoader"
-    start: "ckpt_loader" 
-    output_map:
-      "0": "model" 
-      "1": "clip"  
-    input_map:
-      "model": "model"
-      "clip": "clip"
-    end_input_map:
-      "model": ["ksampler:model"]
-      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
-
-dynamic_conditioning_chains:
-  conditioning_chain:
-    ksampler_node: "ksampler"
-    clip_source: "ckpt_loader:1"
-
-ui_map:
-  model_name: "ckpt_loader:ckpt_name"
+nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
+  ckpt_loader:
+    class_type: CheckpointLoaderSimple
+    title: "Load Checkpoint"
+
+connections:
+  - from: "ckpt_loader:0"
+    to: "ksampler:model"
+  - from: "ckpt_loader:1"
+    to: "pos_prompt:clip"
+  - from: "ckpt_loader:1"
+    to: "neg_prompt:clip"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  - from: "ckpt_loader:2"
+    to: "vae_decode:vae"
+  - from: "ckpt_loader:2"
+    to: "vae_encode:vae"
+
+dynamic_vae_chains:
+  vae_chain:
+    targets:
+      - "vae_decode:vae"
+      - "vae_encode:vae"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    start: "ckpt_loader" 
+    output_map:
+      "0": "model" 
+      "1": "clip"  
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["ksampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "ckpt_loader:1"
+
+ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
+  model_name: "ckpt_loader:ckpt_name"

core/pipelines/workflow_recipes/_partials/conditioning/boogu-image.yaml

@@ -1,58 +1,66 @@
-nodes:
-  unet_loader:
-    class_type: UNETLoader
-    title: "Load Diffusion Model"
-    params:
-      weight_dtype: "default"
-  clip_loader:
-    class_type: CLIPLoader
-    title: "Load CLIP"
-    params:
-      type: "boogu"
-      device: "default"
-  vae_loader:
-    class_type: VAELoader
-    title: "Load VAE"
-
-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_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": ["ksampler:model"]
-      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
-
-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"
-  clip_name: "clip_loader:clip_name"
-  vae_name: "vae_loader:vae_name"
+nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "boogu"
+      device: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+
+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_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": ["ksampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+dynamic_pid_chains:
+  pid_chain:
+    ksampler_node: "ksampler"
+
+ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
+  unet_name: "unet_loader:unet_name"
+  clip_name: "clip_loader:clip_name"
+  vae_name: "vae_loader:vae_name"

core/pipelines/workflow_recipes/_partials/conditioning/chroma1-radiance.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -55,5 +61,7 @@ dynamic_conditioning_chains:
     clip_source: "t5_tokenizer:0"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/chroma1.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -60,6 +66,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip_name: "clip_loader:clip_name"

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

@@ -1,47 +1,55 @@
-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"
+nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
+  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:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/ernie-image.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -53,6 +59,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   clip_name: "clip_loader:clip_name"
   vae_name: "vae_loader:vae_name"

core/pipelines/workflow_recipes/_partials/conditioning/flux1.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load FLUX UNET"
@@ -61,6 +67,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip1_name: "clip_loader:clip_name1"

core/pipelines/workflow_recipes/_partials/conditioning/hidream-i1.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load HiDream-I1 UNET"
@@ -49,6 +55,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip1_name: "clip_loader:clip_name1"

core/pipelines/workflow_recipes/_partials/conditioning/hidream-o1.yaml

@@ -1,68 +1,76 @@
-nodes:
-  ckpt_loader:
-    class_type: CheckpointLoaderSimple
-    title: "Load Checkpoint"
-    
-  model_noise_scale:
-    class_type: ModelNoiseScale
-    title: "ModelNoiseScale"
-    params:
-      noise_scale: 8
-
-connections:
-  - from: "ckpt_loader:0"
-    to: "model_noise_scale:model"
-    
-  - from: "model_noise_scale:0"
-    to: "ksampler:model"
-
-  - from: "ckpt_loader:1"
-    to: "pos_prompt:clip"
-  - from: "ckpt_loader:1"
-    to: "neg_prompt:clip"
-
-  - from: "pos_prompt:0"
-    to: "ksampler:positive"
-  - from: "neg_prompt:0"
-    to: "ksampler:negative"
-    
-  - from: "ckpt_loader:2"
-    to: "vae_decode:vae"
-  - from: "ckpt_loader:2"
-    to: "vae_encode:vae"
-
-dynamic_vae_chains:
-  vae_chain:
-    targets:
-      - "vae_decode:vae"
-      - "vae_encode:vae"
-
-dynamic_lora_chains:
-  lora_chain:
-    template: "LoraLoader"
-    start: "ckpt_loader" 
-    output_map:
-      "0": "model" 
-      "1": "clip"  
-    input_map:
-      "model": "model"
-      "clip": "clip"
-    end_input_map:
-      "model": ["model_noise_scale:model"]
-      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
-
-dynamic_hidream_o1_smoothing_chains:
-  hidream_o1_smoothing_chain:
-    target_node: "ksampler"
-
-dynamic_conditioning_chains:
-  conditioning_chain:
-    ksampler_node: "ksampler"
-    clip_source: "ckpt_loader:1"
-
-dynamic_hidream_o1_reference_chains:
-  hidream_o1_reference_chain:
-    ksampler_node: "ksampler"
-
-ui_map:
+nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
+  ckpt_loader:
+    class_type: CheckpointLoaderSimple
+    title: "Load Checkpoint"
+    
+  model_noise_scale:
+    class_type: ModelNoiseScale
+    title: "ModelNoiseScale"
+    params:
+      noise_scale: 8
+
+connections:
+  - from: "ckpt_loader:0"
+    to: "model_noise_scale:model"
+    
+  - from: "model_noise_scale:0"
+    to: "ksampler:model"
+
+  - from: "ckpt_loader:1"
+    to: "pos_prompt:clip"
+  - from: "ckpt_loader:1"
+    to: "neg_prompt:clip"
+
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+    
+  - from: "ckpt_loader:2"
+    to: "vae_decode:vae"
+  - from: "ckpt_loader:2"
+    to: "vae_encode:vae"
+
+dynamic_vae_chains:
+  vae_chain:
+    targets:
+      - "vae_decode:vae"
+      - "vae_encode:vae"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    start: "ckpt_loader" 
+    output_map:
+      "0": "model" 
+      "1": "clip"  
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["model_noise_scale:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_hidream_o1_smoothing_chains:
+  hidream_o1_smoothing_chain:
+    target_node: "ksampler"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "ckpt_loader:1"
+
+dynamic_hidream_o1_reference_chains:
+  hidream_o1_reference_chain:
+    ksampler_node: "ksampler"
+
+ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   model_name: "ckpt_loader:ckpt_name"

core/pipelines/workflow_recipes/_partials/conditioning/hunyuanimage.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Hunyuan UNET"
@@ -36,6 +42,8 @@ dynamic_conditioning_chains:
     clip_source: "clip_loader:0"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip1_name: "clip_loader:clip_name1"

core/pipelines/workflow_recipes/_partials/conditioning/ideogram-4.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -95,6 +101,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   unet_uncond_name: "unet_uncond_loader:unet_name"
   clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/joyai-image.yaml

@@ -1,86 +1,86 @@
-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: "joyimage"
-      device: "default"
-
-  pos_prompt:
-    class_type: TextEncodeJoyImageEdit
-    title: "TextEncodeJoyImageEdit (Positive)"
-  neg_prompt:
-    class_type: TextEncodeJoyImageEdit
-    title: "TextEncodeJoyImageEdit (Negative)"
-
-  cfg_norm:
-    class_type: CFGNorm
-    title: "CFGNorm"
-    params:
-      strength: 1.0
-      pre_cfg: true
-
-connections:
-  - from: "unet_loader:0"
-    to: "cfg_norm:model"
-  - from: "cfg_norm: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_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": ["cfg_norm:model"]
-      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
-
-dynamic_conditioning_chains:
-  conditioning_chain:
-    ksampler_node: "ksampler"
-    clip_source: "clip_loader:0"
-
-dynamic_pid_chains:
-  pid_chain:
-    ksampler_node: "ksampler"
-
-dynamic_joyai_reference_chains:
-  joyai_reference_chain:
-    pos_prompt_node: "pos_prompt"
-    neg_prompt_node: "neg_prompt"
-    vae_node: "vae_loader"
-
-ui_map:
-  unet_name: "unet_loader:unet_name"
-  vae_name: "vae_loader:vae_name"
-  clip_name: "clip_loader:clip_name"
-  positive_prompt: "pos_prompt:prompt"
-  negative_prompt: "neg_prompt:prompt"
-
+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: "joyimage"
+      device: "default"
+
+  pos_prompt:
+    class_type: TextEncodeJoyImageEdit
+    title: "TextEncodeJoyImageEdit (Positive)"
+  neg_prompt:
+    class_type: TextEncodeJoyImageEdit
+    title: "TextEncodeJoyImageEdit (Negative)"
+
+  cfg_norm:
+    class_type: CFGNorm
+    title: "CFGNorm"
+    params:
+      strength: 1.0
+      pre_cfg: true
+
+connections:
+  - from: "unet_loader:0"
+    to: "cfg_norm:model"
+  - from: "cfg_norm: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_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": ["cfg_norm:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+dynamic_pid_chains:
+  pid_chain:
+    ksampler_node: "ksampler"
+
+dynamic_joyai_reference_chains:
+  joyai_reference_chain:
+    pos_prompt_node: "pos_prompt"
+    neg_prompt_node: "neg_prompt"
+    vae_node: "vae_loader"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"
+  positive_prompt: "pos_prompt:prompt"
+  negative_prompt: "neg_prompt:prompt"
+

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

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -66,6 +72,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip1_name: "clip_loader:clip_name1"

core/pipelines/workflow_recipes/_partials/conditioning/krea-2.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -58,6 +64,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   clip_name: "clip_loader:clip_name"
   vae_name: "vae_loader:vae_name"

core/pipelines/workflow_recipes/_partials/conditioning/lens.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -53,6 +59,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   clip_name: "clip_loader:clip_name"
   vae_name: "vae_loader:vae_name"

core/pipelines/workflow_recipes/_partials/conditioning/longcat-image.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -82,6 +88,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/lumina.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   ckpt_loader:
     class_type: CheckpointLoaderSimple
     title: "Load Checkpoint"
@@ -58,4 +64,6 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   model_name: "ckpt_loader:ckpt_name"

core/pipelines/workflow_recipes/_partials/conditioning/mage-flow.yaml

@@ -0,0 +1,77 @@
+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: "mage"
+      device: "default"
+  text_encode_latent:
+    class_type: TextEncodeMageFlowEdit
+    title: "TextEncodeMageFlowEdit"
+
+connections:
+  - from: "unet_loader:0"
+    to: "ksampler:model"
+  - from: "clip_loader:0"
+    to: "text_encode_latent:clip"
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+  - from: "text_encode_latent:0"
+    to: "ksampler:positive"
+  - from: "text_encode_latent:1"
+    to: "ksampler:negative"
+  - from: "text_encode_latent:2"
+    to: "ksampler:latent_image"
+
+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": ["ksampler:model"]
+      "clip": ["text_encode_latent: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"
+
+dynamic_reference_image_chains:
+  reference_image_chain:
+    text_encode_node: "text_encode_latent"
+    vae_node: "vae_loader"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"
+  positive_prompt: "text_encode_latent:prompt"
+  negative_prompt: "text_encode_latent:negative_prompt"
+  width: "text_encode_latent:width"
+  height: "text_encode_latent:height"
+  batch_size: "text_encode_latent:batch_size"

core/pipelines/workflow_recipes/_partials/conditioning/newbie-image.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -63,6 +69,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip1_name: "clip_loader:clip_name1"

core/pipelines/workflow_recipes/_partials/conditioning/omnigen2.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -58,6 +64,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/ovis-image.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -49,6 +55,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/pixeldit.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -38,6 +44,8 @@ dynamic_conditioning_chains:
     clip_source: "clip_loader:0"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/qwen-image.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Qwen UNET"
@@ -70,6 +76,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/sd15.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   ckpt_loader:
     class_type: CheckpointLoaderSimple
     title: "Load Checkpoint"
@@ -65,5 +71,7 @@ dynamic_conditioning_chains:
     clip_source: "clip_set_last_layer:0"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   model_name: "ckpt_loader:ckpt_name"
   clip_skip: "clip_set_last_layer:stop_at_clip_layer"

core/pipelines/workflow_recipes/_partials/conditioning/sd35.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   ckpt_loader:
     class_type: CheckpointLoaderSimple
     title: "Load Checkpoint"
@@ -59,4 +65,6 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   model_name: "ckpt_loader:ckpt_name"

core/pipelines/workflow_recipes/_partials/conditioning/sdxl.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   ckpt_loader:
     class_type: CheckpointLoaderSimple
     title: "Load Checkpoint"
@@ -64,4 +70,6 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   model_name: "ckpt_loader:ckpt_name"

core/pipelines/workflow_recipes/_partials/conditioning/z-image.yaml

@@ -1,4 +1,10 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   unet_loader:
     class_type: UNETLoader
     title: "Load Diffusion Model"
@@ -64,6 +70,8 @@ dynamic_pid_chains:
     ksampler_node: "ksampler"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   unet_name: "unet_loader:unet_name"
   vae_name: "vae_loader:vae_name"
   clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/input/txt2img_hidream_o1_latent.yaml

@@ -1,11 +1,11 @@
-nodes:
-  latent_source:
-    class_type: "EmptyHiDreamO1LatentImage"
-    title: "Empty HiDream-O1 Latent Image"
-
-connections: []
-
-ui_map:
-  width: "latent_source:width"
-  height: "latent_source:height"
-  batch_size: "latent_source:batch_size"
+nodes:
+  latent_source:
+    class_type: "EmptyHiDreamO1LatentImage"
+    title: "Empty HiDream-O1 Latent Image"
+
+connections: []
+
+ui_map:
+  width: "latent_source:width"
+  height: "latent_source:height"
+  batch_size: "latent_source:batch_size"

core/pipelines/workflow_recipes/_partials/input/txt2img_none.yaml

@@ -0,0 +1,5 @@
+nodes: {}
+
+connections: []
+
+ui_map: {}

ui/events/chain_handlers.py

@@ -748,6 +748,39 @@ def create_joyai_reference_event_handlers(prefix, ui_components):
     del_button.click(fn=del_ref_row, inputs=[count_state], outputs=del_outputs, show_progress=False)
 
 
+def create_reference_image_event_handlers(prefix, ui_components):
+    ref_rows = ui_components.get(f'reference_image_rows_{prefix}')
+    if not ref_rows: return
+    count_state = ui_components[f'reference_image_count_state_{prefix}']
+    add_button = ui_components[f'add_reference_image_button_{prefix}']
+    del_button = ui_components[f'delete_reference_image_button_{prefix}']
+    images = ui_components[f'reference_image_images_{prefix}']
+
+    def add_ref_row(c):
+        c += 1
+        return {
+            count_state: c,
+            ref_rows[c - 1]: gr.update(visible=True),
+            add_button: gr.update(visible=c < 10),
+            del_button: gr.update(visible=True),
+        }
+
+    def del_ref_row(c):
+        c -= 1
+        return {
+            count_state: c,
+            ref_rows[c]: gr.update(visible=False),
+            images[c]: None,
+            add_button: gr.update(visible=True),
+            del_button: gr.update(visible=c > 0),
+        }
+
+    add_outputs = [count_state, add_button, del_button] + ref_rows
+    del_outputs = [count_state, add_button, del_button] + ref_rows + images
+    add_button.click(fn=add_ref_row, inputs=[count_state], outputs=add_outputs, show_progress=False)
+    del_button.click(fn=del_ref_row, inputs=[count_state], outputs=del_outputs, show_progress=False)
+
+
 def create_embedding_event_handlers(prefix, ui_components):
     rows = ui_components.get(f'embedding_rows_{prefix}')
     if not rows: return

ui/events/change_handlers.py

@@ -17,7 +17,7 @@ from .config_loaders import (
     load_ipadapter_config
 )
 
-def make_update_fn(m_comp, cat_comp, cs_comp, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, anima_cn_types, anima_cn_series, anima_cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, krea2_cn_types, krea2_cn_series, krea2_cn_filepaths, ipa_preset, lora_acc, cn_acc, anima_cn_acc, diffsynth_cn_acc, krea2_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, hidream_o1_ref_acc, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp, pid_acc=None, vae_acc=None, joyai_ref_acc=None):
+def make_update_fn(m_comp, cat_comp, cs_comp, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, anima_cn_types, anima_cn_series, anima_cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, krea2_cn_types, krea2_cn_series, krea2_cn_filepaths, ipa_preset, lora_acc, cn_acc, anima_cn_acc, diffsynth_cn_acc, krea2_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, hidream_o1_ref_acc, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp, pid_acc=None, vae_acc=None, joyai_ref_acc=None, ref_img_acc=None):
     def update_fn(*args):
         arch = args[0]
         category = args[1]
@@ -69,6 +69,7 @@ def make_update_fn(m_comp, cat_comp, cs_comp, ar_comp, width_comp, height_comp,
         if ref_latent_acc: updates[ref_latent_acc] = gr.update(visible=('reference_latent' in enabled_chains))
         if hidream_o1_ref_acc: updates[hidream_o1_ref_acc] = gr.update(visible=('hidream_o1_reference' in enabled_chains))
         if joyai_ref_acc: updates[joyai_ref_acc] = gr.update(visible=('joyai_reference' in enabled_chains))
+        if ref_img_acc: updates[ref_img_acc] = gr.update(visible=('reference_image' in enabled_chains))
         if pid_acc: updates[pid_acc] = gr.update(visible=('pid' in enabled_chains))
         if vae_acc: updates[vae_acc] = gr.update(visible=('vae' in enabled_chains))
 
@@ -143,7 +144,7 @@ def make_update_fn(m_comp, cat_comp, cs_comp, ar_comp, width_comp, height_comp,
     return update_fn
 
 
-def make_model_change_fn(cat_comp_ref, cs_comp, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, anima_cn_types, anima_cn_series, anima_cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, krea2_cn_types, krea2_cn_series, krea2_cn_filepaths, arch_comp_ref, ipa_preset, lora_acc, cn_acc, anima_cn_acc, diffsynth_cn_acc, krea2_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, hidream_o1_ref_acc, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp, pid_acc=None, vae_acc=None, joyai_ref_acc=None):
+def make_model_change_fn(cat_comp_ref, cs_comp, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, anima_cn_types, anima_cn_series, anima_cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, krea2_cn_types, krea2_cn_series, krea2_cn_filepaths, arch_comp_ref, ipa_preset, lora_acc, cn_acc, anima_cn_acc, diffsynth_cn_acc, krea2_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, hidream_o1_ref_acc, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp, pid_acc=None, vae_acc=None, joyai_ref_acc=None, ref_img_acc=None):
     def change_fn(*args):
         model_name = args[0]
         idx = 1
@@ -200,9 +201,9 @@ def make_model_change_fn(cat_comp_ref, cs_comp, ar_comp, width_comp, height_comp
         if ref_latent_acc: updates[ref_latent_acc] = gr.update(visible=('reference_latent' in enabled_chains))
         if hidream_o1_ref_acc: updates[hidream_o1_ref_acc] = gr.update(visible=('hidream_o1_reference' in enabled_chains))
         if joyai_ref_acc: updates[joyai_ref_acc] = gr.update(visible=('joyai_reference' in enabled_chains))
+        if ref_img_acc: updates[ref_img_acc] = gr.update(visible=('reference_image' in enabled_chains))
         if pid_acc: updates[pid_acc] = gr.update(visible=('pid' in enabled_chains))
         if vae_acc: updates[vae_acc] = gr.update(visible=('vae' in enabled_chains))
-        if vae_acc: updates[vae_acc] = gr.update(visible=('vae' in enabled_chains))
 
         if cs_comp:
             updates[cs_comp] = gr.update(visible=(arch_model_type == "sd15"))

ui/events/main.py

@@ -13,7 +13,8 @@ from .chain_handlers import (
     create_style_event_handlers,
     create_reference_latent_event_handlers,
     create_hidream_o1_reference_event_handlers,
-    create_joyai_reference_event_handlers
+    create_joyai_reference_event_handlers,
+    create_reference_image_event_handlers
 )
 from .change_handlers import (
     make_update_fn,
@@ -68,6 +69,7 @@ def attach_event_handlers(ui_components, demo):
         ref_latent_accordion = ui_components.get(f'reference_latent_accordion_{prefix}')
         hidream_o1_ref_accordion = ui_components.get(f'hidream_o1_reference_accordion_{prefix}')
         joyai_ref_accordion = ui_components.get(f'joyai_reference_accordion_{prefix}')
+        ref_img_accordion = ui_components.get(f'reference_image_accordion_{prefix}')
         pid_accordion = ui_components.get(f'pid_accordion_{prefix}')
         vae_accordion = ui_components.get(f'vae_accordion_{prefix}')
         
@@ -106,6 +108,7 @@ def attach_event_handlers(ui_components, demo):
             if ref_latent_accordion: outputs.append(ref_latent_accordion)
             if hidream_o1_ref_accordion: outputs.append(hidream_o1_ref_accordion)
             if joyai_ref_accordion: outputs.append(joyai_ref_accordion)
+            if ref_img_accordion: outputs.append(ref_img_accordion)
             if pid_accordion: outputs.append(pid_accordion)
             if vae_accordion: outputs.append(vae_accordion)
             if ipa_preset_list: outputs.append(ipa_preset_list)
@@ -120,7 +123,7 @@ def attach_event_handlers(ui_components, demo):
                 krea2_cn_types_list, krea2_cn_series_list, krea2_cn_filepaths_list,
                 ipa_preset_list, lora_accordion, cn_accordion, anima_cn_accordion, diffsynth_cn_accordion, krea2_cn_accordion, ipa_accordion, sd3_ipa_accordion, flux1_ipa_accordion, style_accordion, embedding_accordion, conditioning_accordion,
                 ref_latent_accordion, hidream_o1_ref_accordion, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp,
-                pid_acc=pid_accordion, vae_acc=vae_accordion, joyai_ref_acc=joyai_ref_accordion
+                pid_acc=pid_accordion, vae_acc=vae_accordion, joyai_ref_acc=joyai_ref_accordion, ref_img_acc=ref_img_accordion
             )
             inputs = [arch_comp, cat_comp]
             if aspect_ratio_comp:
@@ -153,6 +156,7 @@ def attach_event_handlers(ui_components, demo):
             if ref_latent_accordion: outputs2.append(ref_latent_accordion)
             if hidream_o1_ref_accordion: outputs2.append(hidream_o1_ref_accordion)
             if joyai_ref_accordion: outputs2.append(joyai_ref_accordion)
+            if ref_img_accordion: outputs2.append(ref_img_accordion)
             if pid_accordion: outputs2.append(pid_accordion)
             if vae_accordion: outputs2.append(vae_accordion)
             if ipa_preset_list: outputs2.append(ipa_preset_list)
@@ -172,7 +176,7 @@ def attach_event_handlers(ui_components, demo):
                     krea2_cn_types_list, krea2_cn_series_list, krea2_cn_filepaths_list,
                     arch_comp, ipa_preset_list, lora_accordion, cn_accordion, anima_cn_accordion, diffsynth_cn_accordion, krea2_cn_accordion, ipa_accordion, sd3_ipa_accordion, flux1_ipa_accordion, style_accordion, embedding_accordion, conditioning_accordion,
                     ref_latent_accordion, hidream_o1_ref_accordion, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp,
-                    pid_acc=pid_accordion, vae_acc=vae_accordion, joyai_ref_acc=joyai_ref_accordion
+                    pid_acc=pid_accordion, vae_acc=vae_accordion, joyai_ref_acc=joyai_ref_accordion, ref_img_acc=ref_img_accordion
                 )
                 model_comp.change(fn=change_fn, inputs=inputs2, outputs=outputs2)
 
@@ -190,6 +194,7 @@ def attach_event_handlers(ui_components, demo):
         create_reference_latent_event_handlers(prefix, ui_components)
         create_hidream_o1_reference_event_handlers(prefix, ui_components)
         create_joyai_reference_event_handlers(prefix, ui_components)
+        create_reference_image_event_handlers(prefix, ui_components)
         create_run_event(prefix, task_type, ui_components)
 
     if 'view_mode_inpaint' in ui_components:

ui/events/run_handlers.py

@@ -53,6 +53,7 @@ def create_run_event(prefix: str, task_type: str, ui_components: dict):
     reference_latent_data_components = ui_components.get(f'all_reference_latent_components_flat_{prefix}', [])
     hidream_o1_reference_data_components = ui_components.get(f'all_hidream_o1_reference_components_flat_{prefix}', [])
     joyai_reference_data_components = ui_components.get(f'all_joyai_reference_components_flat_{prefix}', [])
+    reference_image_data_components = ui_components.get(f'all_reference_image_components_flat_{prefix}', [])
     
     run_inputs_map['vae_source'] = ui_components.get(f'vae_source_{prefix}')
     run_inputs_map['vae_id'] = ui_components.get(f'vae_id_{prefix}')
@@ -63,7 +64,7 @@ def create_run_event(prefix: str, task_type: str, ui_components: dict):
     all_chains = [
         lora_data_components, controlnet_data_components, anima_controlnet_lllite_data_components, diffsynth_controlnet_data_components, krea2_controlnet_data_components, ipadapter_data_components,
         sd3_ipadapter_data_components, flux1_ipadapter_data_components, style_data_components,
-        embedding_data_components, conditioning_data_components, reference_latent_data_components, hidream_o1_reference_data_components, joyai_reference_data_components
+        embedding_data_components, conditioning_data_components, reference_latent_data_components, hidream_o1_reference_data_components, joyai_reference_data_components, reference_image_data_components
     ]
     for chain in all_chains:
         if chain:
@@ -94,6 +95,7 @@ def create_run_event(prefix: str, task_type: str, ui_components: dict):
         assign_chain_data('reference_latent_data', reference_latent_data_components)
         assign_chain_data('hidream_o1_reference_data', hidream_o1_reference_data_components)
         assign_chain_data('joyai_reference_data', joyai_reference_data_components)
+        assign_chain_data('reference_image_data', reference_image_data_components)
 
         return ui_dict
 

ui/shared/hires_fix_ui.py

@@ -7,7 +7,7 @@ from .ui_components import (
     create_conditioning_ui, create_vae_override_ui,
     create_model_architecture_filter_ui, create_category_filter_ui,
     create_sd3_ipadapter_ui, create_flux1_ipadapter_ui, create_style_ui,
-    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui
+    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui, create_reference_image_ui
 )
 
 default_vals = MODEL_DEFAULTS_CONFIG.get('Default', {})
@@ -102,6 +102,7 @@ def create_ui():
         components.update(create_reference_latent_ui(prefix))
         components.update(create_hidream_o1_reference_ui(prefix))
         components.update(create_joyai_reference_ui(prefix))
+        components.update(create_reference_image_ui(prefix))
         components.update(create_vae_override_ui(prefix))
             
     return components

ui/shared/img2img_ui.py

@@ -7,7 +7,7 @@ from .ui_components import (
     create_conditioning_ui, create_vae_override_ui,
     create_model_architecture_filter_ui, create_category_filter_ui,
     create_sd3_ipadapter_ui, create_flux1_ipadapter_ui, create_style_ui,
-    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui
+    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui, create_reference_image_ui
 )
 
 default_vals = MODEL_DEFAULTS_CONFIG.get('Default', {})
@@ -83,6 +83,7 @@ def create_ui():
         components.update(create_reference_latent_ui(prefix))
         components.update(create_hidream_o1_reference_ui(prefix))
         components.update(create_joyai_reference_ui(prefix))
+        components.update(create_reference_image_ui(prefix))
         components.update(create_vae_override_ui(prefix))
             
     return components

ui/shared/inpaint_ui.py

@@ -6,7 +6,7 @@ from .ui_components import (
     create_conditioning_ui, create_vae_override_ui,
     create_model_architecture_filter_ui, create_category_filter_ui,
     create_sd3_ipadapter_ui, create_flux1_ipadapter_ui, create_style_ui,
-    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui
+    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui, create_reference_image_ui
 )
 
 default_vals = MODEL_DEFAULTS_CONFIG.get('Default', {})
@@ -115,6 +115,7 @@ def create_ui():
             components.update(create_reference_latent_ui(prefix))
             components.update(create_hidream_o1_reference_ui(prefix))
             components.update(create_joyai_reference_ui(prefix))
+            components.update(create_reference_image_ui(prefix))
             components.update(create_vae_override_ui(prefix))
         components[f'accordion_wrapper_{prefix}'] = accordion_wrapper
             

ui/shared/outpaint_ui.py

@@ -7,7 +7,7 @@ from .ui_components import (
     create_conditioning_ui, create_vae_override_ui,
     create_model_architecture_filter_ui, create_category_filter_ui,
     create_sd3_ipadapter_ui, create_flux1_ipadapter_ui, create_style_ui,
-    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui
+    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui, create_reference_image_ui
 )
 
 default_vals = MODEL_DEFAULTS_CONFIG.get('Default', {})
@@ -98,6 +98,7 @@ def create_ui():
         components.update(create_reference_latent_ui(prefix))
         components.update(create_hidream_o1_reference_ui(prefix))
         components.update(create_joyai_reference_ui(prefix))
+        components.update(create_reference_image_ui(prefix))
         components.update(create_vae_override_ui(prefix))
             
     return components

ui/shared/txt2img_ui.py

@@ -6,7 +6,7 @@ from .ui_components import (
     create_conditioning_ui, create_vae_override_ui,
     create_model_architecture_filter_ui, create_category_filter_ui,
     create_sd3_ipadapter_ui, create_flux1_ipadapter_ui, create_style_ui,
-    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui,
+    create_reference_latent_ui, create_hidream_o1_reference_ui, create_joyai_reference_ui, create_reference_image_ui,
     create_pid_ui
 )
 
@@ -55,6 +55,7 @@ def create_ui():
         components.update(create_reference_latent_ui(prefix))
         components.update(create_hidream_o1_reference_ui(prefix))
         components.update(create_joyai_reference_ui(prefix))
+        components.update(create_reference_image_ui(prefix))
         components.update(create_vae_override_ui(prefix))
         components.update(create_pid_ui(prefix))
         

ui/shared/ui_components.py

@@ -725,6 +725,35 @@ def create_joyai_reference_ui(prefix: str, max_units=2):
 
     return components
 
+def create_reference_image_ui(prefix: str, max_units=10):
+    components = {}
+    key = lambda name: f"{name}_{prefix}"
+
+    with gr.Accordion("Mage-Flow Reference Edit Settings", open=False, visible=('reference_image' in default_enabled_chains)) as ref_accordion:
+        components[key('reference_image_accordion')] = ref_accordion
+        gr.Markdown("💡 **Tip:** (Mage-Flow-Edit-Turbo/Mage-Flow-Edit recommended) For multimodal models, this feature enables powerful editing and combining capabilities. In txt2img mode, adding a single reference image performs an **Image Edit**, while adding multiple images performs an **Image Combine**.")
+        
+        ref_image_groups = []
+        ref_image_inputs = []
+        with gr.Row():
+            for i in range(max_units):
+                with gr.Column(visible=(i < 1), min_width=160) as img_col:
+                    img_comp = gr.Image(type="pil", label=f"Ref. {i+1}", sources=["upload"], height=150)
+                    ref_image_groups.append(img_col)
+                    ref_image_inputs.append(img_comp)
+        
+        components[key('reference_image_rows')] = ref_image_groups
+        components[key('reference_image_images')] = ref_image_inputs
+
+        with gr.Row():
+            components[key('add_reference_image_button')] = gr.Button("✚ Add Reference Image")
+            components[key('delete_reference_image_button')] = gr.Button("➖ Delete Reference Image", visible=False)
+        components[key('reference_image_count_state')] = gr.State(1)
+        
+        components[key('all_reference_image_components_flat')] = ref_image_inputs
+
+    return components
+
 def create_pid_ui(prefix: str):
     components = {}
     key = lambda name: f"{name}_{prefix}"

yaml/constants.yaml

@@ -15,6 +15,14 @@ RESOLUTION_MAP:
     "3:4 (Classic Portrait)": [896, 1152]
     "3:2 (Photography)": [1216, 832]
     "2:3 (Photography Portrait)": [832, 1216]
+  mage-flow:
+    "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]
   joyai-image:
     "1:1 (Square)": [1024, 1024]
     "16:9 (Landscape)": [1344, 768]
@@ -241,8 +249,9 @@ RESOLUTION_MAP:
     "2:3 (Portrait)": [512, 768]
 
 MULTIPLIERS_MAP:
-  joyai-image: 1
   krea-2: 1
+  mage-flow: 1
+  joyai-image: 1
   boogu-image: 1
   pixeldit: 1
   ideogram-4: 1

yaml/file_list.yaml

@@ -405,6 +405,23 @@ file:
       source: "hf"
       repo_id: "Comfy-Org/Krea-2"
       repository_file_path: "diffusion_models/krea2_raw_fp8_scaled.safetensors"
+    # Mage-Flow
+    - filename: "mage_flow_turbo_int8_convrot.safetensors"
+      source: "hf"
+      repo_id: "Comfy-Org/Mage-Flow"
+      repository_file_path: "diffusion_models/mage_flow_turbo_int8_convrot.safetensors"
+    - filename: "mage_flow_int8_convrot.safetensors"
+      source: "hf"
+      repo_id: "Comfy-Org/Mage-Flow"
+      repository_file_path: "diffusion_models/mage_flow_int8_convrot.safetensors"
+    - filename: "mage_flow_edit_turbo_int8_convrot.safetensors"
+      source: "hf"
+      repo_id: "Comfy-Org/Mage-Flow"
+      repository_file_path: "diffusion_models/mage_flow_edit_turbo_int8_convrot.safetensors"
+    - filename: "mage_flow_edit_int8_convrot.safetensors"
+      source: "hf"
+      repo_id: "Comfy-Org/Mage-Flow"
+      repository_file_path: "diffusion_models/mage_flow_edit_int8_convrot.safetensors"
     # JoyAI-Image
     - filename: "joyai_image_edit_int8_convrot.safetensors"
       source: "hf"
@@ -916,4 +933,8 @@ file:
     - filename: "flux2-vae.safetensors"
       source: "hf"
       repo_id: "Comfy-Org/flux2-dev"
-      repository_file_path: "split_files/vae/flux2-vae.safetensors"
+      repository_file_path: "split_files/vae/flux2-vae.safetensors"
+    - filename: "mage_flow_vae_bf16.safetensors"
+      source: "hf"
+      repo_id: "Comfy-Org/Mage-Flow"
+      repository_file_path: "vae/mage_flow_vae_bf16.safetensors"

yaml/image_gen_features.yaml

@@ -8,6 +8,10 @@ default:
   - conditioning
   - vae
 
+mage-flow:
+  enabled_chains:
+  - reference_image
+
 joyai-image:
   enabled_chains:
   - joyai_reference

yaml/injectors.yaml

@@ -31,6 +31,8 @@ injector_definitions:
     module: "chain_injectors.hidream_o1_reference_injector"
   dynamic_joyai_reference_chains:
     module: "chain_injectors.joyai_reference_injector"
+  dynamic_reference_image_chains:
+    module: "chain_injectors.reference_image_injector"
   dynamic_pid_chains:
     module: "chain_injectors.pid_injector"
   
@@ -51,4 +53,5 @@ injector_order:
   - dynamic_hidream_o1_smoothing_chains
   - dynamic_hidream_o1_reference_chains
   - dynamic_joyai_reference_chains
+  - dynamic_reference_image_chains
   - dynamic_pid_chains