Prompting: enable long prompts and compel's new .and() concatenating feature (#3497)

this PR adds long prompt support and enables compel's new `.and()`
concatenation feature which improves image quality especially with SD2.1

example of a long prompt:
> a moist sloppy pindlesackboy sloppy hamblin' bogomadong, Clem Fandango
is pissed-off, Wario's Woods in background, making a noise like
ga-woink-a
![000075 6dfd7adf
466129594](https://github.com/invoke-ai/InvokeAI/assets/144366/051608b6-8d52-463b-af10-04b695cda9c1)

the same prompt broken into fragments and concatenated using `.and()`
(syntax works like `.blend()`):
```
("a moist sloppy pindlesackboy sloppy hamblin' bogomadong", 
"Clem Fandango is pissed-off", 
"Wario's Woods in background", 
"making a noise like ga-woink-a").and()
```
![000076 68b1c320
466129594](https://github.com/invoke-ai/InvokeAI/assets/144366/3fee291f-5562-40f9-9c3c-a73765fc893a)


and a less silly example:

> A dream of a distant galaxy, by Caspar David Friedrich, matte
painting, trending on artstation, HQ
![000129 1b33b559
2793529321](https://github.com/invoke-ai/InvokeAI/assets/144366/d4113756-ed0d-49cd-bb2e-a2fc4a09e0af)

the same prompt broken into two fragments and concatenated:
```
("A dream of a distant galaxy, by Caspar David Friedrich, matte painting", 
"trending on artstation, HQ").and()
```
![000128 b5d5cd62
2793529321](https://github.com/invoke-ai/InvokeAI/assets/144366/c373c009-05db-4c42-8a1d-c89fbdb334ec)

as with `.blend()` you can also weight the parts eg `("a man eating an
apple", "sitting on the roof of a car", "high quality, trending on
artstation, 8K UHD").and(1, 0.5, 0.5)` which will assign weight `1` to
`a man eating an apple` and `0.5` to `sitting on the roof of a car` and
`high quality, trending on artstation, 8K UHD`.

invokeai/app/invocations/compel.py

@@ -3,6 +3,7 @@ from pydantic import BaseModel, Field
 
 from invokeai.app.invocations.util.choose_model import choose_model
 from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext, InvocationConfig
+from ...backend.prompting.conditioning import try_parse_legacy_blend
 
 from ...backend.util.devices import choose_torch_device, torch_dtype
 from ...backend.stable_diffusion.diffusion import InvokeAIDiffuserComponent
@@ -13,7 +14,7 @@ from compel.prompt_parser import (
     Blend,
     CrossAttentionControlSubstitute,
     FlattenedPrompt,
-    Fragment,
+    Fragment, Conjunction,
 )
 
 
@@ -93,25 +94,22 @@ class CompelInvocation(BaseInvocation):
             text_encoder=text_encoder,
             textual_inversion_manager=pipeline.textual_inversion_manager,
             dtype_for_device_getter=torch_dtype,
-            truncate_long_prompts=True, # TODO:
+            truncate_long_prompts=False,
         )
 
-        # TODO: support legacy blend?
-
-        conjunction = Compel.parse_prompt_string(prompt_str)
-        prompt: Union[FlattenedPrompt, Blend] = conjunction.prompts[0]
+        legacy_blend = try_parse_legacy_blend(prompt_str, skip_normalize=False)
+        if legacy_blend is not None:
+            conjunction = legacy_blend
+        else:
+            conjunction = Compel.parse_prompt_string(prompt_str)
 
         if context.services.configuration.log_tokenization:
-            log_tokenization_for_prompt_object(prompt, tokenizer)
+            log_tokenization_for_conjunction(conjunction, tokenizer)
 
-        c, options = compel.build_conditioning_tensor_for_prompt_object(prompt)
-
-        # TODO: long prompt support
-        #if not self.truncate_long_prompts:
-        #    [c, uc] = compel.pad_conditioning_tensors_to_same_length([c, uc])
+        c, options = compel.build_conditioning_tensor_for_conjunction(conjunction)
 
         ec = InvokeAIDiffuserComponent.ExtraConditioningInfo(
-            tokens_count_including_eos_bos=get_max_token_count(tokenizer, prompt),
+            tokens_count_including_eos_bos=get_max_token_count(tokenizer, conjunction),
             cross_attention_control_args=options.get("cross_attention_control", None),
         )
 
@@ -128,14 +126,22 @@ class CompelInvocation(BaseInvocation):
 
 
 def get_max_token_count(
-    tokenizer, prompt: Union[FlattenedPrompt, Blend], truncate_if_too_long=False
+    tokenizer, prompt: Union[FlattenedPrompt, Blend, Conjunction], truncate_if_too_long=False
 ) -> int:
     if type(prompt) is Blend:
         blend: Blend = prompt
         return max(
             [
-                get_max_token_count(tokenizer, c, truncate_if_too_long)
-                for c in blend.prompts
+                get_max_token_count(tokenizer, p, truncate_if_too_long)
+                for p in blend.prompts
+            ]
+        )
+    elif type(prompt) is Conjunction:
+        conjunction: Conjunction = prompt
+        return sum(
+            [
+                get_max_token_count(tokenizer, p, truncate_if_too_long)
+                for p in conjunction.prompts
             ]
         )
     else:
@@ -170,6 +176,22 @@ def get_tokens_for_prompt_object(
     return tokens
 
 
+def log_tokenization_for_conjunction(
+    c: Conjunction, tokenizer, display_label_prefix=None
+):
+    display_label_prefix = display_label_prefix or ""
+    for i, p in enumerate(c.prompts):
+        if len(c.prompts)>1:
+            this_display_label_prefix = f"{display_label_prefix}(conjunction part {i + 1}, weight={c.weights[i]})"
+        else:
+            this_display_label_prefix = display_label_prefix
+        log_tokenization_for_prompt_object(
+            p,
+            tokenizer,
+            display_label_prefix=this_display_label_prefix
+        )
+
+
 def log_tokenization_for_prompt_object(
     p: Union[Blend, FlattenedPrompt], tokenizer, display_label_prefix=None
 ):

invokeai/app/invocations/latent.py

@@ -4,6 +4,7 @@ import random
 import einops
 from typing import Literal, Optional, Union, List
 
+from compel import Compel
 from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_controlnet import MultiControlNetModel
 
 from pydantic import BaseModel, Field, validator
@@ -233,6 +234,15 @@ class TextToLatentsInvocation(BaseInvocation):
         c, extra_conditioning_info = context.services.latents.get(self.positive_conditioning.conditioning_name)
         uc, _ = context.services.latents.get(self.negative_conditioning.conditioning_name)
 
+        compel = Compel(
+            tokenizer=model.tokenizer,
+            text_encoder=model.text_encoder,
+            textual_inversion_manager=model.textual_inversion_manager,
+            dtype_for_device_getter=torch_dtype,
+            truncate_long_prompts=False,
+        )
+        [c, uc] = compel.pad_conditioning_tensors_to_same_length([c, uc])
+
         conditioning_data = ConditioningData(
             uc,
             c,

invokeai/backend/prompting/conditioning.py

@@ -38,7 +38,7 @@ def get_uc_and_c_and_ec(prompt_string,
                     dtype_for_device_getter=torch_dtype,
                     truncate_long_prompts=False,
                     )
-    
+
     config = get_invokeai_config()
 
     # get rid of any newline characters
@@ -282,6 +282,8 @@ def split_weighted_subprompts(text, skip_normalize=False) -> list:
         (match.group("prompt").replace("\\:", ":"), float(match.group("weight") or 1))
         for match in re.finditer(prompt_parser, text)
     ]
+    if len(parsed_prompts) == 0:
+        return []
     if skip_normalize:
         return parsed_prompts
     weight_sum = sum(map(lambda x: x[1], parsed_prompts))

pyproject.toml

@@ -38,7 +38,7 @@ dependencies = [
   "albumentations",
   "click",
   "clip_anytorch",          # replacing "clip @ https://github.com/openai/CLIP/archive/eaa22acb90a5876642d0507623e859909230a52d.zip",
-  "compel~=1.1.5",
+  "compel>=1.2.1",
   "controlnet-aux>=0.0.4",
   "timm==0.6.13",           # needed to override timm latest in controlnet_aux, see  https://github.com/isl-org/ZoeDepth/issues/26
   "datasets",