add qwen-lightning

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

CHANGELOG.md

@@ -4,7 +4,7 @@
 
 ### Highlights for 2025-08-10
 
-Several new models: [Qwen-Image](https://qwenlm.github.io/blog/qwen-image/) and [FLUX.1-Krea-Dev](https://www.krea.ai/blog/flux-krea-open-source-release)  
+Several new models: [Qwen-Image](https://qwenlm.github.io/blog/qwen-image/) (plus *Lightning* variant) and [FLUX.1-Krea-Dev](https://www.krea.ai/blog/flux-krea-open-source-release)  
 Several updated models: [Chroma](https://huggingface.co/lodestones/Chroma), [SkyReels-V2](https://huggingface.co/Skywork/SkyReels-V2-DF-14B-720P-Diffusers) and [Wan-VACE](https://huggingface.co/Wan-AI/Wan2.1-VACE-14B-diffusers)  
 Plus continuing with major **UI** work, we have new embedded **Docs/Wiki** search, redesigned real-time **hints**, built-in **GPU monitor**, **CivitAI** integration and more!  
 On the compute side, new profiles for high-vram GPUs, offloading improvements and support for new `torch` release    
@@ -16,11 +16,13 @@ And (*as always*) many bugfixes and improvements to existing features!
 
 - **Models**  
   - [Qwen-Image](https://qwenlm.github.io/blog/qwen-image/)  
-    new image foundational model with 20B params DiT and using Qwen2.5-VL-7B as the text-encoder!  
-    available for text-to-image workflows, image-editing workflows will follow soon  
-    *note*: this model is almost 2x the size of Flux, quantization and offloading are highly recommended!  
-    recommended params: *steps=50, attention-guidance=4*  
+    new image foundational model with *20B* params DiT and using *Qwen2.5-VL-7B* as the text-encoder!  
     available via *networks -> models -> reference*  
+    *note*: this model is almost 2x the size of Flux, quantization and offloading are highly recommended!  
+    *note* qwen-image supports text-to-image workflows as image-editing model is not yet available  
+    *recommended* params: *steps=50, attention-guidance=4*  
+    also available is pre-packaged [Qwen-Lightning](https://huggingface.co/vladmandic/Qwen-Lightning)  
+    which is an unofficial merge of [Qwen-Image](https://qwenlm.github.io/blog/qwen-image/) with [Qwen-Lightning-LoRA](https://github.com/ModelTC/Qwen-Image-Lightning/) to improve quality and allow for generating in 8-steps!  
   - [FLUX.1-Krea-Dev](https://www.krea.ai/blog/flux-krea-open-source-release)  
     new 12B base model compatible with FLUX.1-Dev from *Black Forest Labs* with opinionated aesthetics and aesthetic preferences in mind  
     available via *networks -> models -> reference*  
@@ -38,7 +40,7 @@ And (*as always*) many bugfixes and improvements to existing features!
     optimized support with granular guidance control will follow soon  
 **Torch**  
   - Set default to `torch==2.8.0` for *CUDA, ROCm and OpenVINO*  
-  - Add support for `torch==2.9.0`  
+  - Add support for `torch==2.9.0-nightly`  
 - **UI**  
   - new embedded docs/wiki search!  
     **Docs** search: fully-local and works in real-time on all document pages  

html/reference.json

@@ -169,6 +169,13 @@
     "skip": true,
     "extras": ""
   },
+  "Qwen-Lightning": {
+    "path": "vladmandic/Qwen-Lightning",
+    "preview": "Qwen--Qwen-Image.jpg",
+    "desc": " Qwen-Lightning is step-distilled from Qwen-Image to allow for generation in 8 steps.",
+    "skip": true,
+    "extras": "steps: 8"
+  },
 
   "Ostris Flex.2 Preview": {
     "path": "ostris/Flex.2-preview",

modules/ui_extra_networks.py

@@ -348,9 +348,9 @@ class ExtraNetworksPage:
                 "color": random_bright_color(),
                 "reference": "reference" if 'Reference' in item.get('name', '') else "",
             }
-            alias = item.get("alias", None)
-            if alias is not None:
-                args['title'] += f'\nAlias: {alias}'
+            # alias = item.get("alias", None)
+            # if alias is not None:
+            #     args['title'] += f'\nAlias: {alias}'
             return self.card.format(**args)
         except Exception as e:
             shared.log.error(f'Networks: item error: page={tabname} item={item["name"]} {e}')

pipelines/chroma/convert_chroma.py

@@ -1,94 +0,0 @@
-
-import os
-import torch
-import transformers
-import diffusers
-import huggingface_hub as hf
-from rich import print as rprint
-from rich.traceback import install as install_traceback
-
-
-convert = True
-test = False
-upload = True
-input_files = [
-    'chroma-unlocked-v50.safetensors',
-    'chroma-unlocked-v50-annealed.safetensors',
-]
-input_folder = '/mnt/models/UNET'
-output_folder = '/mnt/models/Diffusers'
-cache_dir = '/mnt/models/huggingface'
-hf_token = ''
-dtype = torch.bfloat16
-device = torch.device('cuda')
-
-
-rprint('starting chroma conversion')
-install_traceback(show_locals=False)
-rprint(f'torch={torch.__version__} diffusers={diffusers.__version__} transformers={transformers.__version__}')
-for input_file in input_files:
-    input_basename = os.path.splitext(input_file)[0]
-    input_model = os.path.join(input_folder, input_file)
-    output_model = os.path.join(output_folder, input_basename)
-
-    if convert:
-        rprint(f'load transformer: {input_model}')
-        transformer = diffusers.ChromaTransformer2DModel.from_single_file(
-            input_model,
-            torch_dtype=dtype,
-            cache_dir=cache_dir,
-        ).to(device)
-
-        rprint('load text-encoder')
-        text_encoder = transformers.T5EncoderModel.from_pretrained(
-            "black-forest-labs/FLUX.1-schnell",
-            subfolder="text_encoder_2",
-            torch_dtype=dtype,
-            cache_dir=cache_dir,
-        ).to(device)
-
-        rprint('load tokenizer')
-        tokenizer = transformers.T5Tokenizer.from_pretrained(
-            "black-forest-labs/FLUX.1-schnell",
-            subfolder="tokenizer_2",
-            cache_dir=cache_dir,
-        )
-
-        rprint('load pipeline')
-        pipe = diffusers.ChromaPipeline.from_pretrained(
-            "black-forest-labs/FLUX.1-dev",
-            transformer=transformer,
-            text_encoder=text_encoder,
-            tokenizer=tokenizer,
-            torch_dtype=dtype,
-            cache_dir=cache_dir,
-        ).to(device)
-
-
-        rprint(f'save pipeline: {output_model}')
-        pipe.save_pretrained(
-            output_model,
-        )
-
-    if test:
-        rprint('test load')
-        pipe = diffusers.ChromaPipeline.from_pretrained(
-            output_model,
-            torch_dtype=dtype,
-            cache_dir=cache_dir,
-        )
-
-    if upload:
-        rprint('hf login')
-        hf.logout()
-        hf.login(token=hf_token, add_to_git_credential=False, write_permission=True)
-        rprint('upload model')
-        pipe.push_to_hub(
-            input_basename,
-            private=False,
-            token=hf_token,
-        )
-
-    pipe = None
-
-rprint('done')