fix: k-diff samplers made more stable by skipping second to last step

crowsonkb/k-diffusion#43 (comment)

See discussion here as well: AUTOMATIC1111/stable-diffusion-webui#3483

imaginairy/samplers/ddim.py

@@ -65,9 +65,10 @@ def sample(
         time_range = np.flip(timesteps)
         total_steps = timesteps.shape[0]
 
-        if orig_latent is not None:
+        # t_start is none if init image strength set to 0
+        if orig_latent is not None and t_start is not None:
             noisy_latent = self.noise_an_image(
-                init_latent=orig_latent, t=t_start, schedule=schedule, noise=noise
+                init_latent=orig_latent, t=t_start - 1, schedule=schedule, noise=noise
             )
         else:
             noisy_latent = noise

imaginairy/samplers/kdiff.py

@@ -14,6 +14,7 @@
 from imaginairy.utils import get_device
 from imaginairy.vendored.k_diffusion import sampling as k_sampling
 from imaginairy.vendored.k_diffusion.external import CompVisDenoiser, CompVisVDenoiser
+from imaginairy.vendored.k_diffusion.sampling import get_sigmas_karras
 
 
 class StandardCompVisDenoiser(CompVisDenoiser):
@@ -96,12 +97,17 @@ def sample(
             t_start = num_steps - t_start + 1
         sigmas = self.cv_denoiser.get_sigmas(num_steps)[t_start:]
 
+        # see https://github.com/crowsonkb/k-diffusion/issues/43#issuecomment-1305195666
+        if self.short_name in (SamplerName.K_DPM_2, SamplerName.K_DPMPP_2M, SamplerName.K_DPM_2_ANCESTRAL):
+            sigmas = torch.cat([sigmas[:-2], sigmas[-1:]])
+
         # if our number of steps is zero, just return the initial latent
         if sigmas.nelement() == 0:
             if orig_latent is not None:
                 return orig_latent
             return noise
 
+        # t_start is none if init image strength set to 0
         if orig_latent is not None and t_start is not None:
             noisy_latent = noise * sigmas[0] + orig_latent
         else:
@@ -141,12 +147,6 @@ def callback(data):
 
         return samples
 
-    @torch.no_grad()
-    def noise_an_image(self, init_latent, t, sigmas, noise=None):
-        if isinstance(t, int):
-            t = torch.tensor([t], device=get_device())
-        t = t.clamp(0, 1000)
-
 
 class DPMFastSampler(KDiffusionSampler):
     short_name = SamplerName.K_DPM_FAST

imaginairy/samplers/plms.py

@@ -75,9 +75,10 @@ def sample(
 
         old_eps = []
 
-        if orig_latent is not None:
+        # t_start is none if init image strength set to 0
+        if orig_latent is not None and t_start is not None:
             noisy_latent = self.noise_an_image(
-                init_latent=orig_latent, t=t_start, schedule=schedule, noise=noise
+                init_latent=orig_latent, t=t_start - 1, schedule=schedule, noise=noise
             )
         else:
             noisy_latent = noise