add gemma3n to caption/vlm and promptenhance

Signed-off-by: Vladimir Mandic <mandic00@live.com>

CHANGELOG.md

@@ -12,7 +12,8 @@ We also switched project license to **Apache-2.0** which means that SD.Next is n
 Feature highlights include:  
 - **ModernUI** layout redesign which should make it more user friendly and easier to navigate  
 - New background replacement and relightning methods using **Latent Bridge Matching** and new **PixelArt** processing filter  
-- Compute improvements with support for **SageAttention2++**  
+- New LLM/VLM models available for captioning and prompt enhance  
+- Compute improvements  
 
 And (as always) many bugfixes and improvements to existing features!  
 
@@ -35,7 +36,8 @@ Although upgrades and existing installations are tested and should work fine!
     available in *img2img or control -> scripts*  
   - Add **FLUX.1-Kontext-Dev** inpaint workflow  
   - Support **FLUX.1** all-in-one safetensors  
-  - Support TAESD preview and remote VAE for **HunyuanDit**  
+  - Support **TAESD** preview and remote VAE for **HunyuanDit**  
+  - Support for [Gemma 3n](https://huggingface.co/google/gemma-3n-E4B-it) E2B and E4B LLM/VLM models in **prompt enhance** and process **captioning**  
 - **UI**  
   - major update to modernui layout  
   - redesign of the Flat UI theme  

modules/interrogate/vqa.py

@@ -25,6 +25,8 @@ vlm_models = {
     "CogFlorence 2.2 Large": "thwri/CogFlorence-2.2-Large", # 1.6GB
     "Moondream 2": "vikhyatk/moondream2", # 3.7GB
     "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 Pix Textcaps": "google/pix2struct-textcaps-base", # 1.1GB
     "Google PaliGemma 2 3B": "google/paligemma2-3b-pt-224",
     "Alibaba Qwen VL2 2B": "Qwen/Qwen2-VL-2B-Instruct",
@@ -168,7 +170,11 @@ def gemma(question: str, image: Image.Image, repo: str = None, system_prompt: st
     if model is None or loaded != repo:
         shared.log.debug(f'Interrogate load: vlm="{repo}"')
         model = None
-        model = transformers.Gemma3ForConditionalGeneration.from_pretrained(
+        if '3n' in repo:
+            cls = transformers.Gemma3nForConditionalGeneration
+        else:
+            cls = transformers.Gemma3ForConditionalGeneration
+        model = cls.from_pretrained(
             repo,
             torch_dtype=devices.dtype,
             cache_dir=shared.opts.hfcache_dir,

scripts/prompt_enhance.py

@@ -35,6 +35,8 @@ class Options:
     models = {
         'google/gemma-3-1b-it': {},
         'google/gemma-3-4b-it': {},
+        'google/gemma-3n-E2B-it': {},
+        'google/gemma-3n-E4B-it': {},
         'Qwen/Qwen3-0.6B-FP8': {},
         'Qwen/Qwen3-1.7B-FP8': {},
         'Qwen/Qwen3-4B-FP8': {},