add prx

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

CHANGELOG.md

@@ -17,6 +17,8 @@ And a first cloud model with **Google's Nano Banana**
   - [Google Gemini 2.5 Flash Nano Banana](https://blog.google/products/gemini/gemini-nano-banana-examples/)  
     first cloud-based model directly supported in SD.Next UI  
     *note*: need to set `GOOGLE_API_KEY` environment variable with your key to use this model  
+  - [Photoroom PRX 1024 Beta](https://huggingface.co/Photoroom/prx-1024-t2i-beta)  
+    PRX (Photoroom Experimental) is a small 1.3-billion-parameter text-to-image model trained entirely from scratch, it uses T5-Gemma text-encoder  
 - **Features**
   - **kanvas**: new module for native canvas-based image manipulation  
     kanvas is a full replacement for *img2img, inpaint and outpaint* controls  

html/reference.json

@@ -912,6 +912,12 @@
     "size": 15.48,
     "date": "2023 April"
   },
+  "Photoroom PRX 1024": {
+    "path": "Photoroom/prx-1024-t2i-beta",
+    "desc": "PRX (Photoroom Experimental) is a 1.3-billion-parameter text-to-image model trained entirely from scratch and released under an Apache 2.0 license.",
+    "preview": "gemini-2.5-flash-image.jpg",
+    "skip": true
+  },
 
   "FLUX.1-Dev sdnq-svd-uint4": {
     "path": "Disty0/FLUX.1-dev-SDNQ-uint4-svd-r32",

modules/modeldata.py

@@ -66,6 +66,8 @@ def get_model_type(pipe):
         model_type = 'nextstep'
     elif 'X-Omni' in name:
         model_type = 'x-omni'
+    elif 'Photoroom' in name:
+        model_type = 'prx'
     # video models
     elif "CogVideo" in name:
         model_type = 'cogvideo'

modules/sd_detect.py

@@ -129,6 +129,8 @@ def guess_by_name(fn, current_guess):
         new_guess = 'Stable Diffusion XL'
     elif 'stable-video-diffusion' in fn.lower():
         new_guess = 'StableVideoDiffusion'
+    elif 'prx-' in fn.lower():
+        new_guess = 'PRX'
     elif 'gemini-2.5-flash-image' in fn.lower():
         new_guess = 'NanoBanana'
     if debug_load:

modules/sd_models.py

@@ -426,6 +426,10 @@ def load_diffuser_force(model_type, checkpoint_info, diffusers_load_config, op='
             from pipelines.model_google import load_nanobanana
             sd_model = load_nanobanana(checkpoint_info, diffusers_load_config)
             allow_post_quant = False
+        elif model_type in ['PRX']:
+            from pipelines.model_prx import load_prx
+            sd_model = load_prx(checkpoint_info, diffusers_load_config)
+            allow_post_quant = False
     except Exception as e:
         shared.log.error(f'Load {op}: path="{checkpoint_info.path}" {e}')
         if debug_load:

pipelines/model_prx.py

@@ -0,0 +1,32 @@
+import diffusers
+from modules import shared, devices, sd_models, model_quant, sd_hijack_te
+from pipelines import generic
+
+
+def load_prx(checkpoint_info, diffusers_load_config=None):
+    if diffusers_load_config is None:
+        diffusers_load_config = {}
+    repo_id = sd_models.path_to_repo(checkpoint_info)
+    sd_models.hf_auth_check(checkpoint_info)
+
+    load_args, _quant_args = model_quant.get_dit_args(diffusers_load_config, allow_quant=False)
+    shared.log.debug(f'Load model: type=PRX repo="{repo_id}" config={diffusers_load_config} offload={shared.opts.diffusers_offload_mode} dtype={devices.dtype} args={load_args}')
+
+    from transformers.models.t5gemma.modeling_t5gemma import T5GemmaEncoder
+    transformer = generic.load_transformer(repo_id, cls_name=diffusers.PRXTransformer2DModel, load_config=diffusers_load_config)
+    text_encoder = generic.load_text_encoder(repo_id, cls_name=T5GemmaEncoder, load_config=diffusers_load_config)
+
+    pipe = diffusers.PRXPipeline.from_pretrained(
+        repo_id,
+        transformer=transformer,
+        text_encoder=text_encoder,
+        cache_dir=shared.opts.diffusers_dir,
+        **load_args,
+    )
+
+    del text_encoder
+    del transformer
+    sd_hijack_te.init_hijack(pipe)
+
+    devices.torch_gc()
+    return pipe