fix(mcp-image-gen): rewrite flux2_klein_heretic workflow with CFGGuider + correct node types

- Replace FluxDisableGuidance+BasicGuider chain with CFGGuider (cfg=5)
- CLIPLoader: add device='default', keep type='flux2'
- UNETLoader: weight_dtype='default' (not fp8_e4m3fn — avoids dimension mismatch)
- VAEDecode/SaveImage: updated node IDs (11→VAEDecode, 12→SaveImage)
- Encoder: qwen_3_4b_bfl.safetensors (7.5GB BFL-merged shards)
- Tests: update heretic model assertions for new node structure (37/37 pass)
- Add RECAP doc with root cause analysis and session history

mcp/mcp-image-gen/src/workflows/flux2_klein_heretic.json

@@ -1,73 +1,98 @@
 {
-  "6": {
+  "1": {
+    "class_type": "CLIPLoader",
+    "inputs": {
+      "clip_name": "qwen_3_4b_bfl.safetensors",
+      "type": "flux2",
+      "device": "default"
+    }
+  },
+  "2": {
     "class_type": "CLIPTextEncode",
     "inputs": {
-      "clip": ["30", 0],
+      "clip": ["1", 0],
       "text": "PROMPT_PLACEHOLDER"
     }
   },
-  "8": {
-    "class_type": "VAEDecode",
+  "3": {
+    "class_type": "CLIPTextEncode",
     "inputs": {
-      "samples": ["13", 0],
-      "vae": ["31", 0]
+      "clip": ["1", 0],
+      "text": "NEGATIVE_PLACEHOLDER"
     }
   },
-  "9": {
-    "class_type": "SaveImage",
+  "4": {
+    "class_type": "UNETLoader",
     "inputs": {
-      "filename_prefix": "mcp-image-gen",
-      "images": ["8", 0]
+      "unet_name": "flux-2-klein-4b.safetensors",
+      "weight_dtype": "default"
     }
   },
-  "13": {
-    "class_type": "KSampler",
-    "inputs": {
-      "cfg": 1.0,
-      "denoise": 1.0,
-      "latent_image": ["27", 0],
-      "model": ["32", 0],
-      "negative": ["33", 0],
-      "positive": ["6", 0],
-      "sampler_name": "euler",
-      "scheduler": "beta",
-      "seed": 42,
-      "steps": 4
-    }
-  },
-  "27": {
-    "class_type": "EmptySD3LatentImage",
-    "inputs": {
-      "batch_size": 1,
-      "height": 1024,
-      "width": 1024
-    }
-  },
-  "30": {
-    "class_type": "CLIPLoader",
-    "inputs": {
-      "clip_name": "qwen_3_4b_heretic.safetensors",
-      "type": "flux"
-    }
-  },
-  "31": {
+  "5": {
     "class_type": "VAELoader",
     "inputs": {
       "vae_name": "flux2-vae.safetensors"
     }
   },
-  "32": {
-    "class_type": "UNETLoader",
+  "6": {
+    "class_type": "EmptyFlux2LatentImage",
     "inputs": {
-      "unet_name": "flux-2-klein-4b.safetensors",
-      "weight_dtype": "fp8_e4m3fn"
+      "width": 1024,
+      "height": 1024,
+      "batch_size": 1
     }
   },
-  "33": {
-    "class_type": "CLIPTextEncode",
+  "7": {
+    "class_type": "Flux2Scheduler",
     "inputs": {
-      "clip": ["30", 0],
-      "text": "NEGATIVE_PLACEHOLDER"
+      "steps": 20,
+      "width": 1024,
+      "height": 1024
+    }
+  },
+  "8": {
+    "class_type": "CFGGuider",
+    "inputs": {
+      "model": ["4", 0],
+      "positive": ["2", 0],
+      "negative": ["3", 0],
+      "cfg": 5
+    }
+  },
+  "9": {
+    "class_type": "KSamplerSelect",
+    "inputs": {
+      "sampler_name": "euler"
+    }
+  },
+  "10": {
+    "class_type": "RandomNoise",
+    "inputs": {
+      "noise_seed": 42
+    }
+  },
+  "11": {
+    "class_type": "SamplerCustomAdvanced",
+    "inputs": {
+      "noise": ["10", 0],
+      "guider": ["8", 0],
+      "sampler": ["9", 0],
+      "sigmas": ["7", 0],
+      "latent_image": ["6", 0]
+    }
+  },
+  "12": {
+    "class_type": "VAEDecode",
+    "inputs": {
+      "samples": ["11", 0],
+      "vae": ["5", 0]
+    }
+  },
+  "13": {
+    "class_type": "SaveImage",
+    "inputs": {
+      "filename_prefix": "mcp-image-gen",
+      "images": ["12", 0]
     }
   }
-}
+}

mcp/mcp-image-gen/tests/test_server.py

@@ -63,11 +63,19 @@ def test_build_flux_workflow_heretic_model():
         seed=42,
         model="flux-2-klein-4b.safetensors",
     )
-    assert wf["6"]["class_type"] == "CLIPTextEncode"
-    assert wf["30"]["class_type"] == "CLIPLoader"  # Qwen3-4B uses single CLIPLoader
-    assert wf["32"]["inputs"]["unet_name"] == "flux-2-klein-4b.safetensors"
-    assert wf["31"]["inputs"]["vae_name"] == "flux2-vae.safetensors"
-    assert wf["13"]["inputs"]["scheduler"] == "beta"  # FLUX.2 Klein uses beta scheduler
+    # New FLUX.2 workflow uses different node IDs and types
+    assert wf["1"]["class_type"] == "CLIPLoader"          # Qwen3-4B uses single CLIPLoader
+    assert wf["1"]["inputs"]["type"] == "flux2"            # correct type for FLUX.2
+    assert wf["1"]["inputs"]["device"] == "default"        # required for FLUX.2 CLIPLoader
+    assert wf["2"]["class_type"] == "CLIPTextEncode"       # standard CLIP encode (not Flux-specific)
+    assert wf["4"]["class_type"] == "UNETLoader"
+    assert wf["4"]["inputs"]["unet_name"] == "flux-2-klein-4b.safetensors"
+    assert wf["4"]["inputs"]["weight_dtype"] == "default"  # not fp8 — avoids dimension errors
+    assert wf["6"]["class_type"] == "EmptyFlux2LatentImage"  # FLUX.2-specific latent
+    assert wf["8"]["class_type"] == "CFGGuider"            # CFGGuider replaces FluxDisableGuidance+BasicGuider
+    assert wf["8"]["inputs"]["cfg"] == 5                   # cfg=5 for FLUX.2 Klein
+    assert wf["11"]["class_type"] == "SamplerCustomAdvanced"  # FLUX.2 sampler (node 11, not 12)
+    assert wf["13"]["class_type"] == "SaveImage"           # output node
 
 
 def test_workflow_registry_contains_both_models():