IPAdapterTesterMixin

src/diffusers/pipelines/animatediff/pipeline_animatediff.py

@@ -1066,7 +1066,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 7. Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 8. Denoising loop
         num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order

src/diffusers/pipelines/animatediff/pipeline_animatediff_video2video.py

@@ -437,6 +437,41 @@ def encode_image(self, image, device, num_images_per_prompt, output_hidden_state
 
             return image_embeds, uncond_image_embeds
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+    def prepare_ip_adapter_image_embeds(
+        self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt
+    ):
+        if ip_adapter_image_embeds is None:
+            if not isinstance(ip_adapter_image, list):
+                ip_adapter_image = [ip_adapter_image]
+
+            if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+                raise ValueError(
+                    f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+                )
+
+            image_embeds = []
+            for single_ip_adapter_image, image_proj_layer in zip(
+                ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+            ):
+                output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+                single_image_embeds, single_negative_image_embeds = self.encode_image(
+                    single_ip_adapter_image, device, 1, output_hidden_state
+                )
+                single_image_embeds = torch.stack([single_image_embeds] * num_images_per_prompt, dim=0)
+                single_negative_image_embeds = torch.stack(
+                    [single_negative_image_embeds] * num_images_per_prompt, dim=0
+                )
+
+                if self.do_classifier_free_guidance:
+                    single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds])
+                    single_image_embeds = single_image_embeds.to(device)
+
+                image_embeds.append(single_image_embeds)
+        else:
+            image_embeds = ip_adapter_image_embeds
+        return image_embeds
+
     # Copied from diffusers.pipelines.text_to_video_synthesis/pipeline_text_to_video_synth.TextToVideoSDPipeline.decode_latents
     def decode_latents(self, latents):
         latents = 1 / self.vae.config.scaling_factor * latents
@@ -731,6 +766,7 @@ def __call__(
         prompt_embeds: Optional[torch.FloatTensor] = None,
         negative_prompt_embeds: Optional[torch.FloatTensor] = None,
         ip_adapter_image: Optional[PipelineImageInput] = None,
+        ip_adapter_image_embeds: Optional[List[torch.FloatTensor]] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
@@ -780,6 +816,9 @@ def __call__(
                 not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
             ip_adapter_image: (`PipelineImageInput`, *optional*):
                 Optional image input to work with IP Adapters.
+            ip_adapter_image_embeds (`List[torch.FloatTensor]`, *optional*):
+                Pre-generated image embeddings for IP-Adapter. If not
+                provided, embeddings are computed from the `ip_adapter_image` input argument.
             output_type (`str`, *optional*, defaults to `"pil"`):
                 The output format of the generated video. Choose between `torch.FloatTensor`, `PIL.Image` or
                 `np.array`.
@@ -866,13 +905,10 @@ def __call__(
         if self.do_classifier_free_guidance:
             prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
 
-        if ip_adapter_image is not None:
-            output_hidden_state = False if isinstance(self.unet.encoder_hid_proj, ImageProjection) else True
-            image_embeds, negative_image_embeds = self.encode_image(
-                ip_adapter_image, device, num_videos_per_prompt, output_hidden_state
+        if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+            image_embeds = self.prepare_ip_adapter_image_embeds(
+                ip_adapter_image, ip_adapter_image_embeds, device, batch_size * num_videos_per_prompt
             )
-            if self.do_classifier_free_guidance:
-                image_embeds = torch.cat([negative_image_embeds, image_embeds])
 
         # 4. Prepare timesteps
         timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
@@ -899,7 +935,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 7. Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 8. Denoising loop
         num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order

src/diffusers/pipelines/controlnet/pipeline_controlnet.py

@@ -1206,7 +1206,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 7.1 Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 7.2 Create tensor stating which controlnets to keep
         controlnet_keep = []

src/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py

@@ -1200,7 +1200,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 7.1 Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 7.2 Create tensor stating which controlnets to keep
         controlnet_keep = []

src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py

@@ -1495,7 +1495,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 7.1 Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 7.2 Create tensor stating which controlnets to keep
         controlnet_keep = []

src/diffusers/pipelines/latent_consistency_models/pipeline_latent_consistency_img2img.py

@@ -477,8 +477,9 @@ def encode_image(self, image, device, num_images_per_prompt, output_hidden_state
 
             return image_embeds, uncond_image_embeds
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
     def prepare_ip_adapter_image_embeds(
-        self, ip_adapter_image, ip_adapter_image_embeds, do_classifier_free_guidance, device, num_images_per_prompt
+        self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt
     ):
         if ip_adapter_image_embeds is None:
             if not isinstance(ip_adapter_image, list):
@@ -502,7 +503,7 @@ def prepare_ip_adapter_image_embeds(
                     [single_negative_image_embeds] * num_images_per_prompt, dim=0
                 )
 
-                if do_classifier_free_guidance:
+                if self.do_classifier_free_guidance:
                     single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds])
                     single_image_embeds = single_image_embeds.to(device)
 
@@ -699,6 +700,10 @@ def cross_attention_kwargs(self):
     def clip_skip(self):
         return self._clip_skip
 
+    @property
+    def do_classifier_free_guidance(self):
+        return False
+
     @property
     def num_timesteps(self):
         return self._num_timesteps
@@ -845,7 +850,7 @@ def __call__(
 
         if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
             image_embeds = self.prepare_ip_adapter_image_embeds(
-                ip_adapter_image, ip_adapter_image_embeds, False, device, batch_size * num_images_per_prompt
+                ip_adapter_image, ip_adapter_image_embeds, device, batch_size * num_images_per_prompt
             )
 
         # 3. Encode input prompt
@@ -860,7 +865,7 @@ def __call__(
             prompt,
             device,
             num_images_per_prompt,
-            False,
+            self.do_classifier_free_guidance,
             negative_prompt=None,
             prompt_embeds=prompt_embeds,
             negative_prompt_embeds=None,
@@ -906,7 +911,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, None)
 
         # 7.1 Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 8. LCM Multistep Sampling Loop
         num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order

src/diffusers/pipelines/latent_consistency_models/pipeline_latent_consistency_text2img.py

@@ -461,6 +461,41 @@ def encode_image(self, image, device, num_images_per_prompt, output_hidden_state
 
             return image_embeds, uncond_image_embeds
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+    def prepare_ip_adapter_image_embeds(
+        self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt
+    ):
+        if ip_adapter_image_embeds is None:
+            if not isinstance(ip_adapter_image, list):
+                ip_adapter_image = [ip_adapter_image]
+
+            if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+                raise ValueError(
+                    f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+                )
+
+            image_embeds = []
+            for single_ip_adapter_image, image_proj_layer in zip(
+                ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+            ):
+                output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+                single_image_embeds, single_negative_image_embeds = self.encode_image(
+                    single_ip_adapter_image, device, 1, output_hidden_state
+                )
+                single_image_embeds = torch.stack([single_image_embeds] * num_images_per_prompt, dim=0)
+                single_negative_image_embeds = torch.stack(
+                    [single_negative_image_embeds] * num_images_per_prompt, dim=0
+                )
+
+                if self.do_classifier_free_guidance:
+                    single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds])
+                    single_image_embeds = single_image_embeds.to(device)
+
+                image_embeds.append(single_image_embeds)
+        else:
+            image_embeds = ip_adapter_image_embeds
+        return image_embeds
+
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
     def run_safety_checker(self, image, device, dtype):
         if self.safety_checker is None:
@@ -590,6 +625,10 @@ def cross_attention_kwargs(self):
     def clip_skip(self):
         return self._clip_skip
 
+    @property
+    def do_classifier_free_guidance(self):
+        return False
+
     @property
     def num_timesteps(self):
         return self._num_timesteps
@@ -610,6 +649,7 @@ def __call__(
         latents: Optional[torch.FloatTensor] = None,
         prompt_embeds: Optional[torch.FloatTensor] = None,
         ip_adapter_image: Optional[PipelineImageInput] = None,
+        ip_adapter_image_embeds: Optional[List[torch.FloatTensor]] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
@@ -660,6 +700,9 @@ def __call__(
                 provided, text embeddings are generated from the `prompt` input argument.
             ip_adapter_image: (`PipelineImageInput`, *optional*):
                 Optional image input to work with IP Adapters.
+            ip_adapter_image_embeds (`List[torch.FloatTensor]`, *optional*):
+                Pre-generated image embeddings for IP-Adapter. If not
+                provided, embeddings are computed from the `ip_adapter_image` input argument.
             output_type (`str`, *optional*, defaults to `"pil"`):
                 The output format of the generated image. Choose between `PIL.Image` or `np.array`.
             return_dict (`bool`, *optional*, defaults to `True`):
@@ -726,12 +769,10 @@ def __call__(
             batch_size = prompt_embeds.shape[0]
 
         device = self._execution_device
-        # do_classifier_free_guidance = guidance_scale > 1.0
 
-        if ip_adapter_image is not None:
-            output_hidden_state = False if isinstance(self.unet.encoder_hid_proj, ImageProjection) else True
-            image_embeds, negative_image_embeds = self.encode_image(
-                ip_adapter_image, device, num_images_per_prompt, output_hidden_state
+        if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+            image_embeds = self.prepare_ip_adapter_image_embeds(
+                ip_adapter_image, ip_adapter_image_embeds, device, batch_size * num_images_per_prompt
             )
 
         # 3. Encode input prompt
@@ -746,7 +787,7 @@ def __call__(
             prompt,
             device,
             num_images_per_prompt,
-            False,
+            self.do_classifier_free_guidance,
             negative_prompt=None,
             prompt_embeds=prompt_embeds,
             negative_prompt_embeds=None,
@@ -786,7 +827,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, None)
 
         # 7.1 Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 8. LCM MultiStep Sampling Loop:
         num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order

src/diffusers/pipelines/pia/pipeline_pia.py

@@ -1188,7 +1188,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 7. Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 8. Denoising loop
         num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py

@@ -1111,7 +1111,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 7.1 Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 7.2 Optionally get Guidance Scale Embedding
         timestep_cond = None

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -1397,7 +1397,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 9.1 Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 9.2 Optionally get Guidance Scale Embedding
         timestep_cond = None

src/diffusers/pipelines/stable_diffusion_panorama/pipeline_stable_diffusion_panorama.py

@@ -777,7 +777,11 @@ def __call__(
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 7.1 Add image embeds for IP-Adapter
-        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+            else None
+        )
 
         # 8. Denoising loop
         # Each denoising step also includes refinement of the latents with respect to the

tests/models/unets/test_models_unet_2d_condition.py

@@ -62,7 +62,10 @@ def create_ip_adapter_state_dict(model):
     key_id = 1
 
     for name in model.attn_processors.keys():
-        cross_attention_dim = None if name.endswith("attn1.processor") else model.config.cross_attention_dim
+        cross_attention_dim = (
+            None if name.endswith("attn1.processor") or "motion_module" in name else model.config.cross_attention_dim
+        )
+
         if name.startswith("mid_block"):
             hidden_size = model.config.block_out_channels[-1]
         elif name.startswith("up_blocks"):
@@ -71,6 +74,7 @@ def create_ip_adapter_state_dict(model):
         elif name.startswith("down_blocks"):
             block_id = int(name[len("down_blocks.")])
             hidden_size = model.config.block_out_channels[block_id]
+
         if cross_attention_dim is not None:
             sd = IPAdapterAttnProcessor(
                 hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, scale=1.0