Merge branch 'vladmandic:dev' into dev

data/reference-community.json

@@ -132,20 +132,12 @@
     "date": "2025 January",
     "skip": true
   },
-  "Anima Preview 1": {
-    "path": "CalamitousFelicitousness/Anima-sdnext-diffusers",
-    "preview": "CalamitousFelicitousness--Anima-sdnext-diffusers.jpg",
-    "desc": "Modified Cosmos-Predict-2B that replaces the T5-11B text encoder with Qwen3-0.6B. Anima is a 2 billion parameter text-to-image model created via a collaboration between CircleStone Labs and Comfy Org. It is focused mainly on anime concepts, characters, and styles, but is also capable of generating a wide variety of other non-photorealistic content. The model is designed for making illustrations and artistic images, and will not work well at realism.",
+  "Anima Preview 3": {
+    "path": "CalamitousFelicitousness/Anima-Preview-3-sdnext-diffusers",
+    "preview": "CalamitousFelicitousness--Anima-Preview-3-sdnext-diffusers.jpg",
+    "desc": "Anima Preview V3 with extended 1024-resolution training and expanded dataset coverage for less common artists. A 2B parameter anime-focused text-to-image model based on modified Cosmos-Predict-2B with Qwen3-0.6B text encoder, created by CircleStone Labs and Comfy Org.",
     "tags": "community",
-    "date": "2026 January",
-    "skip": true
-  },
-  "Anima Preview 2": {
-    "path": "CalamitousFelicitousness/Anima-Preview-2-sdnext-diffusers",
-    "preview": "CalamitousFelicitousness--Anima-Preview-2-sdnext-diffusers.jpg",
-    "desc": "Anima Preview V2 with improved hyperparameters, extended medium-resolution training for more character knowledge, and a regularization dataset for better natural language comprehension. A 2B parameter anime-focused text-to-image model based on modified Cosmos-Predict-2B with Qwen3-0.6B text encoder.",
-    "tags": "community",
-    "date": "2026 March",
+    "date": "2026 April",
     "skip": true
   },
   "FireRed Image Edit 1.0": {

installer.py

@@ -513,7 +513,7 @@ def check_transformers():
     pkg_tokenizers = package_spec('tokenizers')
     # target_commit = '753d61104116eefc8ffc977327b441ee0c8d599f' # transformers commit hash == 4.57.6
     # target_commit = "aad13b87ed59f2afcfaebc985f403301887a35fc" # transformers commit hash == 5.3.0
-    target_commit = "2dba8e0495974930af02274d75bd182d22cc1686" # transformers commit hash == 5.3.0.dev-0331
+    target_commit = "380e3cc5d59912a48508cb6d4959a31cd460e12e" # transformers commit hash == 5.5.0.dev-0409
     if args.use_directml:
         target_transformers = '4.52.4'
         target_tokenizers = '0.21.4'

modules/caption/models_def.py

@@ -5,6 +5,8 @@ vlm_models = {
     "Google Gemma 3 4B": "google/gemma-3-4b-it",
     "Google Gemma 3n E2B": "google/gemma-3n-E2B-it", # 1.5GB
     "Google Gemma 3n E4B": "google/gemma-3n-E4B-it", # 1.5GB
+    "Google Gemma 4 E2B": "google/gemma-4-E2B-it",
+    "Google Gemma 4 E4B": "google/gemma-4-E4B-it",
     "Nidum Gemma 3 4B Uncensored": "nidum/Nidum-Gemma-3-4B-it-Uncensored",
     "Allura Gemma 3 Glitter 4B": "allura-org/Gemma-3-Glitter-4B",
     # Qwen3.5

modules/caption/vqa.py

@@ -619,11 +619,13 @@ class VQA:
         return response
 
     def _load_gemma(self, repo: str):
-        """Load Gemma 3 model and processor."""
+        """Load Gemma model and processor."""
         if self.model is None or self.loaded != repo:
             log.debug(f'Caption load: vlm="{repo}"')
             self._unload_current()
-            if '3n' in repo:
+            if 'gemma-4' in repo.lower():
+                cls = transformers.Gemma4ForConditionalGeneration
+            elif '3n' in repo:
                 cls = transformers.Gemma3nForConditionalGeneration  # pylint: disable=no-member
             else:
                 cls = transformers.Gemma3ForConditionalGeneration

modules/processing_prompt.py

@@ -87,8 +87,8 @@ def set_prompt(p,
         ('StableDiffusion' in cls) or
         ('StableCascade' in cls) or
         ('Flux' in cls and 'Flux2' not in cls) or
-        ('Chroma' in cls) or
-        ('HiDreamImagePipeline' in cls)
+        ('Chroma' in cls)
+        # or ('HiDreamImagePipeline' in cls)
     ):
         jobid = shared.state.begin('TE Encode')
         try:

scripts/prompt_enhance.py

@@ -109,6 +109,8 @@ class Options:
         'google/gemma-3-4b-it',
         'google/gemma-3n-E2B-it',
         'google/gemma-3n-E4B-it',
+        'google/gemma-4-E2B-it',
+        'google/gemma-4-E4B-it',
         # Gemma Finetunes
         'nidum/Nidum-Gemma-3-4B-it-Uncensored',
         'allura-org/Gemma-3-Glitter-4B',
@@ -173,6 +175,8 @@ class Options:
         'google/gemma-3-4b-it': {},
         'google/gemma-3n-E2B-it': {},
         'google/gemma-3n-E4B-it': {},
+        'google/gemma-4-E2B-it': {},
+        'google/gemma-4-E4B-it': {},
         # Gemma Finetunes
         'nidum/Nidum-Gemma-3-4B-it-Uncensored': {},
         'allura-org/Gemma-3-Glitter-4B': {},
@@ -267,6 +271,7 @@ class Options:
         'qwen2_5_vl': 'Qwen2_5_VLForConditionalGeneration',
         'qwen2_vl': 'Qwen2VLForConditionalGeneration',
         'mistral3': 'Mistral3ForConditionalGeneration',
+        'gemma4': 'Gemma4ForConditionalGeneration',
     }
 
     # default = list(models)[1] # gemma-3-4b-it

test/test-detailer-api.py

@@ -30,7 +30,7 @@ FACE_TEST_IMAGES = [
     'models/Reference/ponyRealism_V23.jpg',                               # realistic woman, clear face
     'models/Reference/HiDream-ai--HiDream-I1-Fast.jpg',                   # realistic man, clear face + text
     'models/Reference/stabilityai--stable-diffusion-xl-base-1.0.jpg',     # realistic woman portrait
-    'models/Reference/CalamitousFelicitousness--Anima-sdnext-diffusers.jpg',  # anime face (non-realistic test)
+    'models/Reference/CalamitousFelicitousness--Anima-Preview-3-sdnext-diffusers.jpg',  # anime face (non-realistic test)
 ]
 
 # Fallback images (no guaranteed faces)