Use image_shape for SD3 (#1979)

keras-team · Oct 26, 2024 · 17518e1 · 17518e1
1 parent b3dfa08
commit 17518e1
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 9 deletions.
diff --git a/guides/ipynb/keras_hub/stable_diffusion_3_in_keras_hub.ipynb b/guides/ipynb/keras_hub/stable_diffusion_3_in_keras_hub.ipynb
@@ -10,7 +10,7 @@
     "\n",
     "**Author:** [Hongyu Chiu](https://github.com/james77777778), [fchollet](https://twitter.com/fchollet), [lukewood](https://twitter.com/luke_wood_ml), [divamgupta](https://github.com/divamgupta)<br>\n",
     "**Date created:** 2024/10/09<br>\n",
-    "**Last modified:** 2024/10/09<br>\n",
+    "**Last modified:** 2024/10/24<br>\n",
     "**Description:** Image generation using KerasHub's Stable Diffusion 3 model."
    ]
   },
@@ -96,7 +96,7 @@
     "That will automatically load and configure trained `backbone` and `preprocessor`\n",
     "for you.\n",
     "\n",
-    "Note that in this guide, we'll use `height=512` and `width=512` for faster\n",
+    "Note that in this guide, we'll use `image_shape=(512, 512, 3)` for faster\n",
     "image generation. For higher-quality output, it's recommended to use the default\n",
     "size of `1024`. Since the entire backbone has about 3 billion parameters, which\n",
     "can be challenging to fit into a consumer-level GPU, we set `dtype=\"float16\"` to\n",
@@ -148,7 +148,7 @@
     "\n",
     "\n",
     "backbone = keras_hub.models.StableDiffusion3Backbone.from_preset(\n",
-    "    \"stable_diffusion_3_medium\", height=512, width=512, dtype=\"float16\"\n",
+    "    \"stable_diffusion_3_medium\", image_shape=(512, 512, 3), dtype=\"float16\"\n",
     ")\n",
     "preprocessor = keras_hub.models.StableDiffusion3TextToImagePreprocessor.from_preset(\n",
     "    \"stable_diffusion_3_medium\"\n",

diff --git a/guides/keras_hub/stable_diffusion_3_in_keras_hub.py b/guides/keras_hub/stable_diffusion_3_in_keras_hub.py
@@ -2,7 +2,7 @@
 Title: Stable Diffusion 3 in KerasHub!
 Author: [Hongyu Chiu](https://github.com/james77777778), [fchollet](https://twitter.com/fchollet), [lukewood](https://twitter.com/luke_wood_ml), [divamgupta](https://github.com/divamgupta)
 Date created: 2024/10/09
-Last modified: 2024/10/09
+Last modified: 2024/10/24
 Description: Image generation using KerasHub's Stable Diffusion 3 model.
 Accelerator: GPU
 """
@@ -63,7 +63,7 @@
 That will automatically load and configure trained `backbone` and `preprocessor`
 for you.
 
-Note that in this guide, we'll use `height=512` and `width=512` for faster
+Note that in this guide, we'll use `image_shape=(512, 512, 3)` for faster
 image generation. For higher-quality output, it's recommended to use the default
 size of `1024`. Since the entire backbone has about 3 billion parameters, which
 can be challenging to fit into a consumer-level GPU, we set `dtype="float16"` to
@@ -107,7 +107,7 @@ def display_generated_images(images):
 
 
 backbone = keras_hub.models.StableDiffusion3Backbone.from_preset(
-    "stable_diffusion_3_medium", height=512, width=512, dtype="float16"
+    "stable_diffusion_3_medium", image_shape=(512, 512, 3), dtype="float16"
 )
 preprocessor = keras_hub.models.StableDiffusion3TextToImagePreprocessor.from_preset(
     "stable_diffusion_3_medium"

diff --git a/guides/md/keras_hub/stable_diffusion_3_in_keras_hub.md b/guides/md/keras_hub/stable_diffusion_3_in_keras_hub.md
@@ -2,7 +2,7 @@
 
 **Author:** [Hongyu Chiu](https://github.com/james77777778), [fchollet](https://twitter.com/fchollet), [lukewood](https://twitter.com/luke_wood_ml), [divamgupta](https://github.com/divamgupta)<br>
 **Date created:** 2024/10/09<br>
-**Last modified:** 2024/10/09<br>
+**Last modified:** 2024/10/24<br>
 **Description:** Image generation using KerasHub's Stable Diffusion 3 model.
 
 
@@ -71,7 +71,7 @@ text_to_image = keras_hub.models.StableDiffusion3TextToImage.from_preset(
 That will automatically load and configure trained `backbone` and `preprocessor`
 for you.
 
-Note that in this guide, we'll use `height=512` and `width=512` for faster
+Note that in this guide, we'll use `image_shape=(512, 512, 3)` for faster
 image generation. For higher-quality output, it's recommended to use the default
 size of `1024`. Since the entire backbone has about 3 billion parameters, which
 can be challenging to fit into a consumer-level GPU, we set `dtype="float16"` to
@@ -116,7 +116,7 @@ def display_generated_images(images):
 
 
 backbone = keras_hub.models.StableDiffusion3Backbone.from_preset(
-    "stable_diffusion_3_medium", height=512, width=512, dtype="float16"
+    "stable_diffusion_3_medium", image_shape=(512, 512, 3), dtype="float16"
 )
 preprocessor = keras_hub.models.StableDiffusion3TextToImagePreprocessor.from_preset(
     "stable_diffusion_3_medium"