Remove arrays from prompts

DOCS.md

@@ -19,12 +19,6 @@ This is the same syntax used in [InvokeAI](https://invoke-ai.github.io/InvokeAI/
 | `(blue)1.2` | `(blue:1.2)`  |
 | `(blue)0.8` | `(blue:0.8)`  |
 
-#### Arrays
-
-Arrays allow you to generate multiple different images from a single prompt. For example, `an adult [[blonde,brunette]] [[man,woman]]` will expand into **4** different prompts. This implementation was inspired by [Fooocus](https://github.com/lllyasviel/Fooocus/pull/1503).
-
-> NB: Make sure to set `Images` to the number of images you want to generate. Otherwise, only the first prompt will be used.
-
 ### Models
 
 Each model checkpoint has a different aesthetic:
@@ -61,7 +55,7 @@ Select one or more [textual inversion](https://huggingface.co/docs/diffusers/en/
 
 [Styles](https://huggingface.co/spaces/adamelliotfields/diffusion/blob/main/data/styles.json) are prompt templates that wrap your positive and negative prompts. They were originally derived from the [twri/sdxl_prompt_styler](https://github.com/twri/sdxl_prompt_styler) Comfy node, but have since been entirely rewritten.
 
-Start by framing a simple subject like `portrait of a young adult woman` or `landscape of a mountain range` and experiment.
+Start by framing a simple subject like `portrait of a cat` or `landscape of a mountain range` and experiment.
 
 #### Anime
 
@@ -79,7 +73,7 @@ Rescale up to 4x using [Real-ESRGAN](https://github.com/xinntao/Real-ESRGAN) wit
 
 ### Image-to-Image
 
-The `🖼️ Image` tab enables the image-to-image and IP-Adapter pipelines.
+The `Image-to-Image` settings allows you to provide input images for the initial latents, ControlNet, and IP-Adapter.
 
 #### Strength
 

lib/__init__.py

@@ -1,39 +1,19 @@
-from .annotators import CannyAnnotator
 from .config import Config
 from .inference import generate
-from .loader import Loader
-from .logger import Logger
-from .upscaler import RealESRGAN
 from .utils import (
-    annotate_image,
     async_call,
     disable_progress_bars,
     download_civit_file,
     download_repo_files,
-    enable_progress_bars,
-    load_json,
     read_file,
-    resize_image,
-    safe_progress,
-    timer,
 )
 
 __all__ = [
-    "CannyAnnotator",
     "Config",
-    "Loader",
-    "Logger",
-    "RealESRGAN",
-    "annotate_image",
     "async_call",
     "disable_progress_bars",
     "download_civit_file",
     "download_repo_files",
-    "enable_progress_bars",
     "generate",
-    "load_json",
     "read_file",
-    "resize_image",
-    "safe_progress",
-    "timer",
 ]

lib/inference.py

@@ -1,9 +1,6 @@
-import gc
 import os
-import re
 import time
 from datetime import datetime
-from itertools import product
 
 import torch
 from compel import Compel, DiffusersTextualInversionManager, ReturnedEmbeddingsType
@@ -16,6 +13,7 @@ from .loader import Loader
 from .logger import Logger
 from .utils import (
     annotate_image,
+    clear_cuda_cache,
     load_json,
     resize_image,
     safe_progress,
@@ -23,25 +21,7 @@ from .utils import (
 )
 
 
-def parse_prompt_with_arrays(prompt: str) -> list[str]:
-    arrays = re.findall(r"\[\[(.*?)\]\]", prompt)
-
-    if not arrays:
-        return [prompt]
-
-    tokens = [item.split(",") for item in arrays]  # [("a", "b"), ("1", "2")]
-    combinations = list(product(*tokens))  # [("a", "1"), ("a", "2"), ("b", "1"), ("b", "2")]
-
-    # find all the arrays in the prompt and replace them with tokens
-    prompts = []
-    for combo in combinations:
-        current_prompt = prompt
-        for i, token in enumerate(combo):
-            current_prompt = current_prompt.replace(f"[[{arrays[i]}]]", token.strip(), 1)
-        prompts.append(current_prompt)
-    return prompts
-
-
+# Inject prompts into style templates
 def apply_style(positive_prompt, negative_prompt, style_id="none"):
     if style_id.lower() == "none":
         return (positive_prompt, negative_prompt)
@@ -78,6 +58,7 @@ def gpu_duration(**kwargs):
     return loading + (duration * num_images)
 
 
+# Request GPU when deployed to Hugging Face
 @GPU(duration=gpu_duration)
 def generate(
     positive_prompt,
@@ -143,7 +124,7 @@ def generate(
     else:
         IP_ADAPTER = ""
 
-    # custom progress bar for multiple images
+    # Custom progress bar for multiple images
     def callback_on_step_end(pipeline, step, timestep, latents):
         nonlocal CURRENT_STEP, CURRENT_IMAGE
         if progress is not None:
@@ -224,7 +205,7 @@ def generate(
         except (EnvironmentError, HFValidationError, RepositoryNotFoundError):
             raise Error(f"Invalid embedding: {embedding}")
 
-    # prompt embeds
+    # Embed prompts with weights
     compel = Compel(
         device=pipe.device,
         tokenizer=pipe.tokenizer,
@@ -241,15 +222,9 @@ def generate(
     for i in range(num_images):
         try:
             generator = torch.Generator(device=pipe.device).manual_seed(current_seed)
+            positive_styled, negative_styled = apply_style(positive_prompt, negative_prompt, style)
 
-            positive_prompts = parse_prompt_with_arrays(positive_prompt)
-            index = i % len(positive_prompts)
-            positive_styled, negative_styled = apply_style(
-                positive_prompts[index],
-                negative_prompt,
-                style,
-            )
-
+            # User didn't provide a negative prompt
             if negative_styled.startswith("(), "):
                 negative_styled = negative_styled[4:]
 
@@ -305,6 +280,7 @@ def generate(
             CURRENT_STEP = 0
             CURRENT_IMAGE += 1
 
+    # Upscale
     if scale > 1:
         msg = f"Upscaling {scale}x"
         with timer(msg, logger=log.info):
@@ -314,13 +290,15 @@ def generate(
                 images[i] = image
                 safe_progress(progress, i + 1, num_images, desc=msg)
 
-    # cleanup
-    loader.collect()
-    gc.collect()
+    # Flush memory after generating
+    clear_cuda_cache()
 
     end = time.perf_counter()
     msg = f"Generating {len(images)} image{'s' if len(images) > 1 else ''} took {end - start:.2f}s"
     log.info(msg)
+
+    # Alert if notifier provided
     if Info:
         Info(msg)
+
     return images

lib/loader.py

@@ -10,7 +10,7 @@ from diffusers.models.attention_processor import AttnProcessor2_0, IPAdapterAttn
 from .config import Config
 from .logger import Logger
 from .upscaler import RealESRGAN
-from .utils import safe_progress, timer
+from .utils import clear_cuda_cache, safe_progress, timer
 
 
 class Loader:
@@ -184,10 +184,10 @@ class Loader:
             to_unload.append("model")
             to_unload.append("pipe")
 
-        self.collect()
+        clear_cuda_cache()
         for component in to_unload:
             setattr(self, component, None)
-            gc.collect()
+        gc.collect()
 
     def _should_load_upscaler(self, scale=1):
         if self.upscaler is None and scale > 1:
@@ -311,12 +311,6 @@ class Loader:
                         variant="fp16",
                     ).to(self.pipe.device)
 
-    def collect(self):
-        torch.cuda.empty_cache()
-        torch.cuda.ipc_collect()
-        torch.cuda.reset_peak_memory_stats()
-        torch.cuda.synchronize()
-
     def load(
         self,
         kind,

lib/utils.py

@@ -9,6 +9,7 @@ from typing import Callable, Tuple, TypeVar
 import anyio
 import httpx
 import numpy as np
+import torch
 from anyio import Semaphore
 from diffusers.utils import logging as diffusers_logging
 from huggingface_hub._snapshot_download import snapshot_download
@@ -66,6 +67,14 @@ def safe_progress(progress, current=0, total=0, desc=""):
         progress((current, total), desc=desc)
 
 
+def clear_cuda_cache():
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+        torch.cuda.ipc_collect()
+        torch.cuda.reset_peak_memory_stats()
+        torch.cuda.synchronize()
+
+
 def download_repo_files(repo_id, allow_patterns, token=None):
     was_disabled = are_progress_bars_disabled()
     enable_progress_bars()