Upload 5 files

app.py

@@ -2,16 +2,17 @@ import spaces
 import gradio as gr
 import json
 import torch
-from diffusers import DiffusionPipeline, AutoencoderTiny, AutoencoderKL
+from diffusers import DiffusionPipeline, AutoencoderTiny, AutoencoderKL, AutoPipelineForImage2Image
 from live_preview_helpers import flux_pipe_call_that_returns_an_iterable_of_images
-from diffusers import FluxControlNetPipeline, FluxControlNetModel, FluxMultiControlNetModel
+from diffusers.utils import load_image
+from diffusers import FluxControlNetPipeline, FluxControlNetModel, FluxMultiControlNetModel, FluxControlNetImg2ImgPipeline
 from huggingface_hub import HfFileSystem, ModelCard
 import random
 import time
 
 from env import models, num_loras, num_cns
 from mod import (clear_cache, get_repo_safetensors, is_repo_name, is_repo_exists, get_model_trigger,
-                 description_ui, compose_lora_json, is_valid_lora, fuse_loras, save_image,
+                 description_ui, compose_lora_json, is_valid_lora, fuse_loras, save_image, preprocess_i2i_image,
                  get_trigger_word, enhance_prompt, deselect_lora, set_control_union_image,
                  get_control_union_mode, set_control_union_mode, get_control_params, translate_to_en)
 from flux import (search_civitai_lora, select_civitai_lora, search_civitai_lora_json,
@@ -34,6 +35,8 @@ dtype = torch.bfloat16
 taef1 = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype)
 good_vae = AutoencoderKL.from_pretrained(base_model, subfolder="vae", torch_dtype=dtype)
 pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=dtype, vae=taef1)
+pipe_i2i = AutoPipelineForImage2Image.from_pretrained(base_model, vae=good_vae, transformer=pipe.transformer, text_encoder=pipe.text_encoder,
+ tokenizer=pipe.tokenizer, text_encoder_2=pipe.text_encoder_2, tokenizer_2=pipe.tokenizer_2, torch_dtype=dtype)
 controlnet_union = None
 controlnet = None
 last_model = models[0]
@@ -45,9 +48,11 @@ MAX_SEED = 2**32-1
 
 # https://huggingface.co/InstantX/FLUX.1-dev-Controlnet-Union
 # https://huggingface.co/spaces/jiuface/FLUX.1-dev-Controlnet-Union
+# https://huggingface.co/docs/diffusers/main/en/api/pipelines/flux
 #@spaces.GPU()
-def change_base_model(repo_id: str, cn_on: bool, progress=gr.Progress(track_tqdm=True)): # , progress=gr.Progress(track_tqdm=True) # gradio.exceptions.Error: 'Model load Error: too many values to unpack (expected 2)'
+def change_base_model(repo_id: str, cn_on: bool, disable_model_cache: bool, progress=gr.Progress(track_tqdm=True)):
     global pipe
+    global pipe_i2i
     global taef1
     global good_vae
     global controlnet_union
@@ -56,8 +61,9 @@ def change_base_model(repo_id: str, cn_on: bool, progress=gr.Progress(track_tqdm
     global last_cn_on
     global dtype
     try:
-        if (repo_id == last_model and cn_on is last_cn_on) or not is_repo_name(repo_id) or not is_repo_exists(repo_id): return gr.update(visible=True)
+        if not disable_model_cache and (repo_id == last_model and cn_on is last_cn_on) or not is_repo_name(repo_id) or not is_repo_exists(repo_id): return gr.update(visible=True)
         pipe.to("cpu")
+        pipe_i2i.to("cpu")
         good_vae.to("cpu")
         taef1.to("cpu")
         if controlnet is not None: controlnet.to("cpu")
@@ -69,6 +75,8 @@ def change_base_model(repo_id: str, cn_on: bool, progress=gr.Progress(track_tqdm
             controlnet_union = FluxControlNetModel.from_pretrained(controlnet_model_union_repo, torch_dtype=dtype)
             controlnet = FluxMultiControlNetModel([controlnet_union])
             pipe = FluxControlNetPipeline.from_pretrained(repo_id, controlnet=controlnet, torch_dtype=dtype)
+            pipe_i2i = FluxControlNetImg2ImgPipeline.from_pretrained(repo_id, controlnet=controlnet, vae=None, transformer=pipe.transformer, text_encoder=pipe.text_encoder,
+             tokenizer=pipe.tokenizer, text_encoder_2=pipe.text_encoder_2, tokenizer_2=pipe.tokenizer_2, torch_dtype=dtype)
             last_model = repo_id
             last_cn_on = cn_on
             progress(1, desc=f"Model loaded: {repo_id} / ControlNet Loaded: {controlnet_model_union_repo}")
@@ -77,6 +85,8 @@ def change_base_model(repo_id: str, cn_on: bool, progress=gr.Progress(track_tqdm
             progress(0, desc=f"Loading model: {repo_id}")
             print(f"Loading model: {repo_id}")
             pipe = DiffusionPipeline.from_pretrained(repo_id, torch_dtype=dtype)
+            pipe_i2i = AutoPipelineForImage2Image.from_pretrained(repo_id, vae=None, transformer=pipe.transformer, text_encoder=pipe.text_encoder,
+             tokenizer=pipe.tokenizer, text_encoder_2=pipe.text_encoder_2, tokenizer_2=pipe.tokenizer_2, torch_dtype=dtype)
             last_model = repo_id
             last_cn_on = cn_on
             progress(1, desc=f"Model loaded: {repo_id}")
@@ -183,7 +193,67 @@ def generate_image(prompt_mash, steps, seed, cfg_scale, width, height, lora_scal
         print(e)
         raise gr.Error(f"Inference Error: {e}") from e
 
-def run_lora(prompt, cfg_scale, steps, selected_index, randomize_seed, seed, width, height,
+@spaces.GPU(duration=70)
+def generate_image_to_image(prompt_mash, image_input_path, image_strength, steps, cfg_scale, width, height, lora_scale, seed, cn_on, progress=gr.Progress(track_tqdm=True)):
+    global pipe_i2i
+    global good_vae
+    global controlnet
+    global controlnet_union
+    try:
+        good_vae.to("cuda")
+        generator = torch.Generator(device="cuda").manual_seed(int(float(seed)))
+        image_input = load_image(image_input_path)
+        
+        with calculateDuration("Generating image"):
+            # Generate image
+            modes, images, scales = get_control_params()
+            if True or not cn_on or len(modes) == 0:
+                pipe_i2i.to("cuda")
+                pipe_i2i.vae = good_vae
+                image_input = load_image(image_input_path)
+                progress(0, desc="Start I2I Inference.")
+                final_image = pipe_i2i(
+                    prompt=prompt_mash,
+                    image=image_input,
+                    strength=image_strength,
+                    num_inference_steps=steps,
+                    guidance_scale=cfg_scale,
+                    width=width,
+                    height=height,
+                    generator=generator,
+                    joint_attention_kwargs={"scale": lora_scale},
+                    output_type="pil",
+                ).images[0]
+                return final_image 
+            else: # omitted
+                pipe_i2i.to("cuda")
+                pipe_i2i.vae = good_vae
+                image_input = load_image(image_input_path)
+                if controlnet_union is not None: controlnet_union.to("cuda")
+                if controlnet is not None: controlnet.to("cuda")
+                pipe_i2i.enable_model_cpu_offload()
+                progress(0, desc="Start I2I Inference with ControlNet.")
+                final_image = pipe_i2i(
+                    prompt=prompt_mash,
+                    control_image=images,
+                    control_mode=modes,
+                    image=image_input,
+                    strength=image_strength,
+                    num_inference_steps=steps,
+                    guidance_scale=cfg_scale,
+                    width=width,
+                    height=height,
+                    controlnet_conditioning_scale=scales,
+                    generator=generator,
+                    joint_attention_kwargs={"scale": lora_scale},
+                    output_type="pil",
+                ).images[0]
+                return final_image
+    except Exception as e:
+        print(e)
+        raise gr.Error(f"I2I Inference Error: {e}") from e
+
+def run_lora(prompt, image_input, image_strength, cfg_scale, steps, selected_index, randomize_seed, seed, width, height,
               lora_scale, lora_json, cn_on, translate_on, progress=gr.Progress(track_tqdm=True)):
     global pipe
     if selected_index is None and not is_valid_lora(lora_json):
@@ -195,6 +265,8 @@ def run_lora(prompt, cfg_scale, steps, selected_index, randomize_seed, seed, wid
         try:
             pipe.unfuse_lora()
             pipe.unload_lora_weights()
+            pipe_i2i.unfuse_lora()
+            pipe_i2i.unload_lora_weights()
         except Exception as e:
             print(e)
     
@@ -225,10 +297,16 @@ def run_lora(prompt, cfg_scale, steps, selected_index, randomize_seed, seed, wid
             prompt_mash = prompt_mash
         # Load LoRA weights
         with calculateDuration(f"Loading LoRA weights for {selected_lora['title']}"):
-            if "weights" in selected_lora:
-                pipe.load_lora_weights(lora_path, weight_name=selected_lora["weights"])
+            if(image_input is not None):
+                if "weights" in selected_lora:
+                    pipe_i2i.load_lora_weights(lora_path, weight_name=selected_lora["weights"])
+                else:
+                    pipe_i2i.load_lora_weights(lora_path)
             else:
-                pipe.load_lora_weights(lora_path)
+                if "weights" in selected_lora:
+                    pipe.load_lora_weights(lora_path, weight_name=selected_lora["weights"])
+                else:
+                    pipe.load_lora_weights(lora_path)
         
     # Set random seed for reproducibility
     with calculateDuration("Randomizing seed"):
@@ -236,17 +314,20 @@ def run_lora(prompt, cfg_scale, steps, selected_index, randomize_seed, seed, wid
             seed = random.randint(0, MAX_SEED)
     
     progress(0, desc="Running Inference.")
-    image_generator = generate_image(prompt_mash, steps, seed, cfg_scale, width, height, lora_scale, cn_on, progress)
-    # Consume the generator to get the final image
-    final_image = None
-    step_counter = 0
-    for image in image_generator:
-        step_counter+=1
-        final_image = image
-        progress_bar = f'<div class="progress-container"><div class="progress-bar" style="--current: {step_counter}; --total: {steps};"></div></div>'
-        yield image, seed, gr.update(value=progress_bar, visible=True)
-        
-    yield save_image(final_image, None, last_model, prompt_mash, height, width, steps, cfg_scale, seed), seed, gr.update(value=progress_bar, visible=False)
+    if(image_input is not None):
+        final_image = generate_image_to_image(prompt_mash, image_input, image_strength, steps, cfg_scale, width, height, lora_scale, seed, cn_on, progress)
+        yield save_image(final_image, None, last_model, prompt_mash, height, width, steps, cfg_scale, seed), seed, gr.update(visible=False)
+    else:
+        image_generator = generate_image(prompt_mash, steps, seed, cfg_scale, width, height, lora_scale, cn_on, progress)
+        # Consume the generator to get the final image
+        final_image = None
+        step_counter = 0
+        for image in image_generator:
+            step_counter+=1
+            final_image = image
+            progress_bar = f'<div class="progress-container"><div class="progress-bar" style="--current: {step_counter}; --total: {steps};"></div></div>'
+            yield image, seed, gr.update(value=progress_bar, visible=True)
+        yield save_image(final_image, None, last_model, prompt_mash, height, width, steps, cfg_scale, seed), seed, gr.update(value=progress_bar, visible=False)
 
 def get_huggingface_safetensors(link):
   split_link = link.split("/")
@@ -392,8 +473,14 @@ with gr.Blocks(theme='Nymbo/Nymbo_Theme', fill_width=True, css=css, delete_cache
             model_info = gr.Markdown(elem_classes="info")
         with gr.Row():
             with gr.Accordion("Advanced Settings", open=False):
+                with gr.Row():
+                    input_image = gr.Image(label="Input image", type="filepath", height=256, sources=["upload", "clipboard"], show_share_button=False)
+                    with gr.Column():
+                        image_strength = gr.Slider(label="Image Strength", minimum=0.1, maximum=1.0, step=0.01, value=0.75)
+                        input_image_preprocess = gr.Checkbox(True, label="Preprocess Input image")
                 with gr.Column():
                     with gr.Row():
+                        lora_scale = gr.Slider(label="LoRA Scale", minimum=-3, maximum=3, step=0.01, value=0.95)
                         width = gr.Slider(label="Width", minimum=256, maximum=1536, step=64, value=1024)
                         height = gr.Slider(label="Height", minimum=256, maximum=1536, step=64, value=1024)
                     with gr.Row():
@@ -402,7 +489,7 @@ with gr.Blocks(theme='Nymbo/Nymbo_Theme', fill_width=True, css=css, delete_cache
                     with gr.Row():
                         randomize_seed = gr.Checkbox(True, label="Randomize seed")
                         seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0, randomize=True)
-                        lora_scale = gr.Slider(label="LoRA Scale", minimum=-3, maximum=3, step=0.01, value=0.95)
+                        disable_model_cache = gr.Checkbox(False, label="Disable model caching")
                     with gr.Accordion("External LoRA", open=True):
                         with gr.Column():
                             lora_repo_json = gr.JSON(value=[{}] * num_loras, visible=False)
@@ -486,20 +573,21 @@ with gr.Blocks(theme='Nymbo/Nymbo_Theme', fill_width=True, css=css, delete_cache
     gr.on(
         triggers=[generate_button.click, prompt.submit],
         fn=change_base_model,
-        inputs=[model_name, cn_on],
+        inputs=[model_name, cn_on, disable_model_cache],
         outputs=[result],
         queue=True,
         show_api=False,
         trigger_mode="once",
     ).success(
         fn=run_lora,
-        inputs=[prompt, cfg_scale, steps, selected_index, randomize_seed, seed, width, height,
+        inputs=[prompt, input_image, image_strength, cfg_scale, steps, selected_index, randomize_seed, seed, width, height,
                  lora_scale, lora_repo_json, cn_on, auto_trans], 
         outputs=[result, seed, progress_bar],
         queue=True,
         show_api=True,
     )
 
+    input_image.upload(preprocess_i2i_image, [input_image, input_image_preprocess, height, width], [input_image], queue=False, show_api=False)
     deselect_lora_button.click(deselect_lora, None, [prompt, selected_info, selected_index, width, height], queue=False, show_api=False)
     gr.on(
         triggers=[model_name.change, cn_on.change],
@@ -509,7 +597,7 @@ with gr.Blocks(theme='Nymbo/Nymbo_Theme', fill_width=True, css=css, delete_cache
         queue=False,
         show_api=False,
         trigger_mode="once",
-    ).then(change_base_model, [model_name, cn_on], [result], queue=True, show_api=False)
+    ).then(change_base_model, [model_name, cn_on, disable_model_cache], [result], queue=True, show_api=False)
     prompt_enhance.click(enhance_prompt, [prompt], [prompt], queue=False, show_api=False)
 
     gr.on(

loras.json

@@ -48,12 +48,30 @@
     "trigger_word": "in the style of TOK a trtcrd, tarot style",
     "aspect": "portrait"
   },
+  {
+    "repo": "alvdansen/pola-photo-flux",
+    "image": "https://huggingface.co/alvdansen/pola-photo-flux/resolve/main/images/out-2%20(83).webp",
+    "trigger_word": ", polaroid style",
+    "title": "Polaroid Style"
+  },
+  {
+    "image": "https://huggingface.co/dvyio/flux-lora-the-sims/resolve/main/images/dunBAVBsALOepaE_dsWFI_6b0fef6b0fc4472aa07d00edea7c75b3.jpg",
+    "repo": "dvyio/flux-lora-the-sims",
+    "trigger_word": ", video game screenshot in the style of THSMS",
+    "title": "The Sims style" 
+  },
   {
     "image": "https://huggingface.co/alvdansen/softpasty-flux-dev/resolve/main/images/ComfyUI_00814_%20(2).png",
     "title": "SoftPasty",
     "repo": "alvdansen/softpasty-flux-dev",
     "trigger_word": "araminta_illus illustration style"
   },
+  {
+    "image": "https://huggingface.co/dvyio/flux-lora-film-noir/resolve/main/images/S8iWMa0GamEcFkanHHmI8_a232d8b83bb043808742d661dac257f7.jpg",
+    "title": "Film Noir",
+    "repo": "dvyio/flux-lora-film-noir",
+    "trigger_word": "in the style of FLMNR"
+  },
   {
     "image": "https://huggingface.co/AIWarper/RubberCore1920sCartoonStyle/resolve/main/images/Rub_00006_.png",
     "title": "1920s cartoon",
@@ -61,6 +79,13 @@
     "trigger_word": "RU883R style",
     "trigger_position": "prepend"
   },
+  {
+    "image": "https://huggingface.co/Norod78/JojosoStyle-flux-lora/resolve/main/samples/1725244218477__000004255_1.jpg",
+    "title": "JoJo Style",
+    "repo": "Norod78/JojosoStyle-flux-lora",
+    "trigger_word": "JojosoStyle",
+    "trigger_position": "prepend"
+  },
   {
     "image": "https://github.com/XLabs-AI/x-flux/blob/main/assets/readme/examples/picture-6-rev1.png?raw=true",
     "title": "flux-Realism",
@@ -81,6 +106,13 @@
     "trigger_word": "mgwr/cine",
     "trigger_position": "prepend"
   },
+  {
+    "image": "https://huggingface.co/sWizad/pokemon-trainer-sprites-pixelart-flux/resolve/main/26578915.jpeg",
+    "repo": "sWizad/pokemon-trainer-sprites-pixelart-flux",
+    "title": "Pokemon Trainer Sprites",
+    "trigger_word": "white background, a pixel image of",
+    "trigger_position": "prepend" 
+  },
   {
     "image": "https://huggingface.co/nerijs/animation2k-flux/resolve/main/images/Q8-oVxNnXvZ9HNrgbNpGw_02762aaaba3b47859ee5fe9403a371e3.png",
     "title": "animation2k",
@@ -100,12 +132,6 @@
     "trigger_word": "ps1 game screenshot,",
     "trigger_position": "prepend"
   },
-  {
-    "image": "https://huggingface.co/Shakker-Labs/FLUX.1-dev-LoRA-blended-realistic-illustration/resolve/main/images/example3.png",
-    "title": "Blended Realistic Illustration",
-    "repo": "Shakker-Labs/FLUX.1-dev-LoRA-blended-realistic-illustration",
-    "trigger_word": "artistic style blends reality and illustration elements"
-  },
   {
     "image": "https://huggingface.co/alvdansen/flux-koda/resolve/main/images/ComfyUI_00566_%20(2).png",
     "title": "flux koda",
@@ -119,7 +145,7 @@
     "trigger_word": ""
   },
   {
-    "image": "https://pbs.twimg.com/media/GU7NsZPa8AA4Ddl?format=jpg&name=4096x4096",
+    "image": "https://huggingface.co/davisbro/half_illustration/resolve/main/images/example3.webp",
     "title": "Half Illustration",
     "repo": "davisbro/half_illustration",
     "trigger_word": "in the style of TOK"
@@ -171,8 +197,7 @@
   {
     "image": "https://huggingface.co/kudzueye/Boreal/resolve/main/images/ComfyUI_00845_.png",
     "title": "Boreal",
-    "repo": "kudzueye/Boreal",
-    "weights": "boreal-flux-dev-lora-v04_1000_steps.safetensors",
+    "repo": "kudzueye/boreal-flux-dev-v2",
     "trigger_word": "phone photo"
   },
   {

mod.py

@@ -239,6 +239,18 @@ def set_control_union_image(i: int, mode: str, image: Image.Image | None, height
     return control_images[i]
 
 
+def preprocess_i2i_image(image_path: str, is_preprocess: bool, height: int, width: int):
+    try:
+        if not is_preprocess: return image_path
+        image_resolution = max(width, height) 
+        image = Image.open(image_path)
+        image_resized = resize_image(expand2square(image.convert("RGB")), image_resolution, image_resolution, False)
+        image_resized.save(image_path)
+    except Exception as e:
+        raise gr.Error(f"Error: {e}")
+    return image_path
+
+
 def compose_lora_json(lorajson: list[dict], i: int, name: str, scale: float, filename: str, trigger: str):
     lorajson[i]["name"] = str(name) if name != "None" else ""
     lorajson[i]["scale"] = float(scale)

requirements.txt

@@ -1,6 +1,6 @@
 spaces
 torch
-git+https://github.com/huggingface/diffusers
+git+https://github.com/huggingface/diffusers@aa73072f1f7014635e3de916cbcf47858f4c37a0
 transformers
 peft
 sentencepiece