inpaint model progress

- working with plain prompts, weighted prompts and merge prompts
- not tested with prompt2prompt

configs/stable-diffusion/v1-inpainting-inference.yaml

@@ -76,4 +76,4 @@ model:
           target: torch.nn.Identity
 
     cond_stage_config:
-      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder
+      target: ldm.modules.encoders.modules.WeightedFrozenCLIPEmbedder

ldm/models/diffusion/ksampler.py

@@ -43,14 +43,7 @@ class CFGDenoiser(nn.Module):
 
 
     def forward(self, x, sigma, uncond, cond, cond_scale):
-        if isinstance(cond,dict): # hybrid model
-            x_in = torch.cat([x] * 2)
-            sigma_in = torch.cat([sigma] * 2)
-            cond_in = self.sampler.make_cond_in(uncond,cond)
-            uncond, cond = self.inner_model(x_in, sigma_in, cond=cond_in).chunk(2)
-            next_x = uncond + (cond - uncond) * cond_scale
-        else: # cross attention model
-            next_x = self.invokeai_diffuser.do_diffusion_step(x, sigma, uncond, cond, cond_scale)
+        next_x = self.invokeai_diffuser.do_diffusion_step(x, sigma, uncond, cond, cond_scale)
         if self.warmup < self.warmup_max:
             thresh = max(1, 1 + (self.threshold - 1) * (self.warmup / self.warmup_max))
             self.warmup += 1

ldm/models/diffusion/shared_invokeai_diffusion.py

@@ -90,7 +90,19 @@ class InvokeAIDiffuserComponent:
             # faster batched path
             x_twice = torch.cat([x]*2)
             sigma_twice = torch.cat([sigma]*2)
-            both_conditionings = torch.cat([unconditioning, conditioning])
+            if isinstance(conditioning, dict):
+                assert isinstance(unconditioning, dict)
+                both_conditionings = dict()
+                for k in conditioning:
+                    if isinstance(conditioning[k], list):
+                        both_conditionings[k] = [
+                            torch.cat([unconditioning[k][i], conditioning[k][i]])
+                            for i in range(len(conditioning[k]))
+                        ]
+                    else:
+                        both_conditionings[k] = torch.cat([unconditioning[k], conditioning[k]])
+            else:
+                both_conditionings = torch.cat([unconditioning, conditioning])
             unconditioned_next_x, conditioned_next_x = self.model_forward_callback(x_twice, sigma_twice, both_conditionings).chunk(2)
         else:
             #print('pct', percent_through, ': doing cross attention control on', cross_attention_control_types_to_do)

ldm/modules/encoders/modules.py

@@ -439,7 +439,7 @@ class FrozenCLIPEmbedder(AbstractEncoder):
             param.requires_grad = False
 
     def forward(self, text, **kwargs):
-
+        print(f'DEBUG text={text}, max_length={self.max_length}')
         batch_encoding = self.tokenizer(
             text,
             truncation=True,