Support K-LMS in img2img (#270)

* Support K-LMS in img2img

* Apply review suggestions

examples/inference/image_to_image.py

@@ -5,7 +5,14 @@ import numpy as np
 import torch
 
 import PIL
-from diffusers import AutoencoderKL, DDIMScheduler, DiffusionPipeline, PNDMScheduler, UNet2DConditionModel
+from diffusers import (
+    AutoencoderKL,
+    DDIMScheduler,
+    DiffusionPipeline,
+    LMSDiscreteScheduler,
+    PNDMScheduler,
+    UNet2DConditionModel,
+)
 from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
 from tqdm.auto import tqdm
 from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
@@ -87,12 +94,17 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
         # get the original timestep using init_timestep
         init_timestep = int(num_inference_steps * strength) + offset
         init_timestep = min(init_timestep, num_inference_steps)
-        timesteps = self.scheduler.timesteps[-init_timestep]
-        timesteps = torch.tensor([timesteps] * batch_size, dtype=torch.long, device=self.device)
+        if isinstance(self.scheduler, LMSDiscreteScheduler):
+            timesteps = torch.tensor(
+                [num_inference_steps - init_timestep] * batch_size, dtype=torch.long, device=self.device
+            )
+        else:
+            timesteps = self.scheduler.timesteps[-init_timestep]
+            timesteps = torch.tensor([timesteps] * batch_size, dtype=torch.long, device=self.device)
 
         # add noise to latents using the timesteps
         noise = torch.randn(init_latents.shape, generator=generator, device=self.device)
-        init_latents = self.scheduler.add_noise(init_latents, noise, timesteps)
+        init_latents = self.scheduler.add_noise(init_latents, noise, timesteps).to(self.device)
 
         # get prompt text embeddings
         text_input = self.tokenizer(
@@ -133,8 +145,15 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
         latents = init_latents
         t_start = max(num_inference_steps - init_timestep + offset, 0)
         for i, t in tqdm(enumerate(self.scheduler.timesteps[t_start:])):
+            t_index = t_start + i
             # expand the latents if we are doing classifier free guidance
             latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+            if isinstance(self.scheduler, LMSDiscreteScheduler):
+                sigma = self.scheduler.sigmas[t_index]
+                # the model input needs to be scaled to match the continuous ODE formulation in K-LMS
+                latent_model_input = latent_model_input / ((sigma**2 + 1) ** 0.5)
+                latent_model_input = latent_model_input.to(self.unet.dtype)
+                t = t.to(self.unet.dtype)
 
             # predict the noise residual
             noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=text_embeddings)["sample"]
@@ -145,11 +164,14 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
                 noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
 
             # compute the previous noisy sample x_t -> x_t-1
-            latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs)["prev_sample"]
+            if isinstance(self.scheduler, LMSDiscreteScheduler):
+                latents = self.scheduler.step(noise_pred, t_index, latents, **extra_step_kwargs)["prev_sample"]
+            else:
+                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs)["prev_sample"]
 
         # scale and decode the image latents with vae
         latents = 1 / 0.18215 * latents
-        image = self.vae.decode(latents)
+        image = self.vae.decode(latents.to(self.vae.dtype))
 
         image = (image / 2 + 0.5).clamp(0, 1)
         image = image.cpu().permute(0, 2, 3, 1).numpy()

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

@@ -138,6 +138,7 @@ class StableDiffusionPipeline(DiffusionPipeline):
             latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
             if isinstance(self.scheduler, LMSDiscreteScheduler):
                 sigma = self.scheduler.sigmas[i]
+                # the model input needs to be scaled to match the continuous ODE formulation in K-LMS
                 latent_model_input = latent_model_input / ((sigma**2 + 1) ** 0.5)
 
             # predict the noise residual

src/diffusers/schedulers/scheduling_lms_discrete.py

@@ -124,8 +124,9 @@ class LMSDiscreteScheduler(SchedulerMixin, ConfigMixin):
         return {"prev_sample": prev_sample}
 
     def add_noise(self, original_samples, noise, timesteps):
-        sigmas = self.match_shape(self.sigmas, noise)
-        noisy_samples = original_samples + noise * sigmas[timesteps]
+        sigmas = self.match_shape(self.sigmas[timesteps], noise)
+        noisy_samples = original_samples + noise * sigmas
+
         return noisy_samples
 
     def __len__(self):