Merge 29938f760f12443b70820d4c5b0c0461463ff00d into 56fa7dbe380cb5591c5542f8aa51ce2fc26beedf

There is still one of the text encoders missing and I didn't actually test it.

.github/workflows/test-ci.yml

@@ -5,6 +5,7 @@ on:
   push:
     branches:
       - master
+      - release/**
     paths-ignore:
       - 'app/**'
       - 'input/**'

.github/workflows/test-execution.yml

@@ -2,9 +2,9 @@ name: Execution Tests
 
 on:
   push:
-    branches: [ main, master ]
+    branches: [ main, master, release/** ]
   pull_request:
-    branches: [ main, master ]
+    branches: [ main, master, release/** ]
 
 jobs:
   test:

.github/workflows/test-launch.yml

@@ -2,9 +2,9 @@ name: Test server launches without errors
 
 on:
   push:
-    branches: [ main, master ]
+    branches: [ main, master, release/** ]
   pull_request:
-    branches: [ main, master ]
+    branches: [ main, master, release/** ]
 
 jobs:
   test:

.github/workflows/test-unit.yml

@@ -2,9 +2,9 @@ name: Unit Tests
 
 on:
   push:
-    branches: [ main, master ]
+    branches: [ main, master, release/** ]
   pull_request:
-    branches: [ main, master ]
+    branches: [ main, master, release/** ]
 
 jobs:
   test:

.github/workflows/update-version.yml

@@ -6,6 +6,7 @@ on:
       - "pyproject.toml"
     branches:
       - master
+      - release/**
 
 jobs:
   update-version:

.github/workflows/windows_release_nightly_pytorch.yml

@@ -9,6 +9,12 @@ on:
         type: string
         default: "129"
 
+      release_tag:
+        description: 'Git tag to attach nightly package (e.g., v0.37.0); leave empty to upload to latest'
+        required: false
+        type: string
+        default: ""
+
       python_minor:
         description: 'python minor version'
         required: true
@@ -34,6 +40,7 @@ jobs:
     steps:
         - uses: actions/checkout@v4
           with:
+            ref: ${{ inputs.release_tag || 'master' }}
             fetch-depth: 30
             persist-credentials: false
         - uses: actions/setup-python@v5
@@ -89,5 +96,5 @@ jobs:
           with:
                 repo_token: ${{ secrets.GITHUB_TOKEN }}
                 file: ComfyUI_windows_portable_nvidia_or_cpu_nightly_pytorch.7z
-                tag: "latest"
+                tag: ${{ inputs.release_tag || 'latest' }}
                 overwrite: true

.github/workflows/windows_release_package.yml

@@ -9,6 +9,12 @@ on:
         type: string
         default: "129"
 
+      release_tag:
+        description: 'Git tag to package and attach (e.g., v0.37.0); leave empty to build from master and upload to latest'
+        required: false
+        type: string
+        default: ""
+
       python_minor:
         description: 'python minor version'
         required: true
@@ -50,6 +56,7 @@ jobs:
 
         - uses: actions/checkout@v4
           with:
+            ref: ${{ inputs.release_tag || 'master' }}
             fetch-depth: 150
             persist-credentials: false
         - shell: bash
@@ -101,6 +108,5 @@ jobs:
           with:
                 repo_token: ${{ secrets.GITHUB_TOKEN }}
                 file: new_ComfyUI_windows_portable_nvidia_cu${{ inputs.cu }}_or_cpu.7z
-                tag: "latest"
+                tag: ${{ inputs.release_tag || 'latest' }}
                 overwrite: true
-

comfy/ldm/lumina/model.py

@@ -377,6 +377,7 @@ class NextDiT(nn.Module):
         z_image_modulation=False,
         time_scale=1.0,
         pad_tokens_multiple=None,
+        clip_text_dim=None,
         image_model=None,
         device=None,
         dtype=None,
@@ -447,6 +448,31 @@ class NextDiT(nn.Module):
             ),
         )
 
+        self.clip_text_pooled_proj = None
+
+        if clip_text_dim is not None:
+            self.clip_text_dim = clip_text_dim
+            self.clip_text_pooled_proj = nn.Sequential(
+                operation_settings.get("operations").RMSNorm(clip_text_dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype")),
+                operation_settings.get("operations").Linear(
+                    clip_text_dim,
+                    clip_text_dim,
+                    bias=True,
+                    device=operation_settings.get("device"),
+                    dtype=operation_settings.get("dtype"),
+                ),
+            )
+            self.time_text_embed = nn.Sequential(
+                nn.SiLU(),
+                operation_settings.get("operations").Linear(
+                    min(dim, 1024) + clip_text_dim,
+                    min(dim, 1024),
+                    bias=True,
+                    device=operation_settings.get("device"),
+                    dtype=operation_settings.get("dtype"),
+                ),
+            )
+
         self.layers = nn.ModuleList(
             [
                 JointTransformerBlock(
@@ -585,6 +611,15 @@ class NextDiT(nn.Module):
 
         cap_feats = self.cap_embedder(cap_feats)  # (N, L, D)  # todo check if able to batchify w.o. redundant compute
 
+        if self.clip_text_pooled_proj is not None:
+            pooled = kwargs.get("clip_text_pooled", None)
+            if pooled is not None:
+                pooled = self.clip_text_pooled_proj(pooled)
+            else:
+                pooled = torch.zeros((1, self.clip_text_dim), device=x.device, dtype=x.dtype)
+
+            adaln_input = self.time_text_embed(torch.cat((t, pooled), dim=-1))
+
         patches = transformer_options.get("patches", {})
         x_is_tensor = isinstance(x, torch.Tensor)
         img, mask, img_size, cap_size, freqs_cis = self.patchify_and_embed(x, cap_feats, cap_mask, t, num_tokens, transformer_options=transformer_options)

comfy/model_base.py

@@ -1110,6 +1110,10 @@ class Lumina2(BaseModel):
             if 'num_tokens' not in out:
                 out['num_tokens'] = comfy.conds.CONDConstant(cross_attn.shape[1])
 
+        clip_text_pooled = kwargs["pooled_output"]  # Newbie
+        if clip_text_pooled is not None:
+            out['clip_text_pooled'] = comfy.conds.CONDRegular(clip_text_pooled)
+
         return out
 
 class WAN21(BaseModel):

comfy/model_detection.py

@@ -423,6 +423,9 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
             dit_config["axes_lens"] = [300, 512, 512]
             dit_config["rope_theta"] = 10000.0
             dit_config["ffn_dim_multiplier"] = 4.0
+            ctd_weight = state_dict.get('{}clip_text_pooled_proj.0.weight'.format(key_prefix), None)
+            if ctd_weight is not None:
+                dit_config["clip_text_dim"] = ctd_weight.shape[0]
         elif dit_config["dim"] == 3840:  # Z image
             dit_config["n_heads"] = 30
             dit_config["n_kv_heads"] = 30