Make height and width optional (huggingface#1401)

* fix

* add test

* fix test

* uP

* up

* fix some tests

Author: patrickvonplaten

src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion.py

@@ -390,8 +390,8 @@ def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype
     def __call__(
         self,
         prompt: Union[str, List[str]],
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -411,9 +411,9 @@ def __call__(
         Args:
             prompt (`str` or `List[str]`):
                 The prompt or prompts to guide the image generation.
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -459,6 +459,9 @@ def __call__(
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
 
         # 1. Check inputs. Raise error if not correct
         self.check_inputs(prompt, height, width, callback_steps)

src/diffusers/pipelines/latent_diffusion/pipeline_latent_diffusion.py

@@ -65,8 +65,8 @@ def __init__(
     def __call__(
         self,
         prompt: Union[str, List[str]],
-        height: Optional[int] = 256,
-        width: Optional[int] = 256,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: Optional[int] = 50,
         guidance_scale: Optional[float] = 1.0,
         eta: Optional[float] = 0.0,
@@ -79,9 +79,9 @@ def __call__(
         Args:
             prompt (`str` or `List[str]`):
                 The prompt or prompts to guide the image generation.
-            height (`int`, *optional*, defaults to 256):
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 256):
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -106,6 +106,9 @@ def __call__(
             `return_dict` is True, otherwise a `tuple. When returning a tuple, the first element is a list with the
             generated images.
         """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
 
         if isinstance(prompt, str):
             batch_size = 1

src/diffusers/pipelines/stable_diffusion/pipeline_flax_stable_diffusion.py

@@ -160,13 +160,17 @@ def _generate(
         params: Union[Dict, FrozenDict],
         prng_seed: jax.random.PRNGKey,
         num_inference_steps: int = 50,
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         guidance_scale: float = 7.5,
         latents: Optional[jnp.array] = None,
         debug: bool = False,
         neg_prompt_ids: jnp.array = None,
     ):
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
+
         if height % 8 != 0 or width % 8 != 0:
             raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
 
@@ -249,8 +253,8 @@ def __call__(
         params: Union[Dict, FrozenDict],
         prng_seed: jax.random.PRNGKey,
         num_inference_steps: int = 50,
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         guidance_scale: float = 7.5,
         latents: jnp.array = None,
         return_dict: bool = True,
@@ -265,9 +269,9 @@ def __call__(
         Args:
             prompt (`str` or `List[str]`):
                 The prompt or prompts to guide the image generation.
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -302,6 +306,10 @@ def __call__(
             element is a list of `bool`s denoting whether the corresponding generated image likely represents
             "not-safe-for-work" (nsfw) content, according to the `safety_checker`.
         """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
+
         if jit:
             images = _p_generate(
                 self,

src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py

@@ -172,8 +172,8 @@ def _encode_prompt(self, prompt, num_images_per_prompt, do_classifier_free_guida
     def __call__(
         self,
         prompt: Union[str, List[str]],
-        height: Optional[int] = 512,
-        width: Optional[int] = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: Optional[int] = 50,
         guidance_scale: Optional[float] = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -187,6 +187,10 @@ def __call__(
         callback_steps: Optional[int] = 1,
         **kwargs,
     ):
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
+
         if isinstance(prompt, str):
             batch_size = 1
         elif isinstance(prompt, list):

src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion_inpaint.py

@@ -236,8 +236,8 @@ def __call__(
         prompt: Union[str, List[str]],
         image: PIL.Image.Image,
         mask_image: PIL.Image.Image,
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -265,9 +265,9 @@ def __call__(
                 repainted, while black pixels will be preserved. If `mask_image` is a PIL image, it will be converted
                 to a single channel (luminance) before use. If it's a tensor, it should contain one color channel (L)
                 instead of 3, so the expected shape would be `(B, H, W, 1)`.
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -312,6 +312,10 @@ def __call__(
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
+
         if isinstance(prompt, str):
             batch_size = 1
         elif isinstance(prompt, list):

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

@@ -389,8 +389,8 @@ def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype
     def __call__(
         self,
         prompt: Union[str, List[str]],
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -410,9 +410,9 @@ def __call__(
         Args:
             prompt (`str` or `List[str]`):
                 The prompt or prompts to guide the image generation.
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -458,6 +458,9 @@ def __call__(
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
 
         # 1. Check inputs. Raise error if not correct
         self.check_inputs(prompt, height, width, callback_steps)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_image_variation.py

@@ -292,8 +292,8 @@ def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype
     def __call__(
         self,
         image: Union[PIL.Image.Image, List[PIL.Image.Image], torch.FloatTensor],
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
         num_images_per_prompt: Optional[int] = 1,
@@ -315,9 +315,9 @@ def __call__(
                 configuration of
                 [this](https://huggingface.co/lambdalabs/sd-image-variations-diffusers/blob/main/feature_extractor/preprocessor_config.json)
                 `CLIPFeatureExtractor`
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -360,6 +360,9 @@ def __call__(
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
 
         # 1. Check inputs. Raise error if not correct
         self.check_inputs(image, height, width, callback_steps)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -509,8 +509,8 @@ def __call__(
         prompt: Union[str, List[str]],
         image: Union[torch.FloatTensor, PIL.Image.Image],
         mask_image: Union[torch.FloatTensor, PIL.Image.Image],
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -538,9 +538,9 @@ def __call__(
                 repainted, while black pixels will be preserved. If `mask_image` is a PIL image, it will be converted
                 to a single channel (luminance) before use. If it's a tensor, it should contain one color channel (L)
                 instead of 3, so the expected shape would be `(B, H, W, 1)`.
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -586,6 +586,9 @@ def __call__(
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
 
         # 1. Check inputs
         self.check_inputs(prompt, height, width, callback_steps)

src/diffusers/pipelines/stable_diffusion_safe/pipeline_stable_diffusion_safe.py

@@ -495,8 +495,8 @@ def perform_safety_guidance(
     def __call__(
         self,
         prompt: Union[str, List[str]],
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -521,9 +521,9 @@ def __call__(
         Args:
             prompt (`str` or `List[str]`):
                 The prompt or prompts to guide the image generation.
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -589,6 +589,9 @@ def __call__(
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * 8
+        width = width or self.unet.config.sample_size * 8
 
         # 1. Check inputs. Raise error if not correct
         self.check_inputs(prompt, height, width, callback_steps)

src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion.py

@@ -111,8 +111,8 @@ def disable_attention_slicing(self):
     def image_variation(
         self,
         image: Union[torch.FloatTensor, PIL.Image.Image],
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -131,9 +131,9 @@ def image_variation(
         Args:
             image (`PIL.Image.Image`, `List[PIL.Image.Image]` or `torch.Tensor`):
                 The image prompt or prompts to guide the image generation.
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.image_unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.image_unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -193,7 +193,7 @@ def image_variation(
         >>> pipe = pipe.to("cuda")
 
         >>> generator = torch.Generator(device="cuda").manual_seed(0)
-        >>> image = pipe(image, generator=generator).images[0]
+        >>> image = pipe.image_variation(image, generator=generator).images[0]
         >>> image.save("./car_variation.png")
         ```
 
@@ -227,8 +227,8 @@ def image_variation(
     def text_to_image(
         self,
         prompt: Union[str, List[str]],
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -247,9 +247,9 @@ def text_to_image(
         Args:
             prompt (`str` or `List[str]`):
                 The prompt or prompts to guide the image generation.
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.image_unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.image_unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
@@ -341,8 +341,8 @@ def dual_guided(
         prompt: Union[PIL.Image.Image, List[PIL.Image.Image]],
         image: Union[str, List[str]],
         text_to_image_strength: float = 0.5,
-        height: int = 512,
-        width: int = 512,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
         num_images_per_prompt: Optional[int] = 1,
@@ -360,9 +360,9 @@ def dual_guided(
         Args:
             prompt (`str` or `List[str]`):
                 The prompt or prompts to guide the image generation.
-            height (`int`, *optional*, defaults to 512):
+            height (`int`, *optional*, defaults to self.image_unet.config.sample_size * 8):
                 The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to 512):
+            width (`int`, *optional*, defaults to self.image_unet.config.sample_size * 8):
                 The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the