Merge branch 'main' into feature/z-image-control

Author: Pfannkuchensack

invokeai/app/invocations/latents_to_image.py

@@ -2,12 +2,6 @@
 
 import torch
 from diffusers.image_processor import VaeImageProcessor
-from diffusers.models.attention_processor import (
-    AttnProcessor2_0,
-    LoRAAttnProcessor2_0,
-    LoRAXFormersAttnProcessor,
-    XFormersAttnProcessor,
-)
 from diffusers.models.autoencoders.autoencoder_kl import AutoencoderKL
 from diffusers.models.autoencoders.autoencoder_tiny import AutoencoderTiny
 
@@ -77,26 +71,9 @@ def invoke(self, context: InvocationContext) -> ImageOutput:
             assert isinstance(vae, (AutoencoderKL, AutoencoderTiny))
             latents = latents.to(TorchDevice.choose_torch_device())
             if self.fp32:
+                # FP32 mode: convert everything to float32 for maximum precision
                 vae.to(dtype=torch.float32)
-
-                use_torch_2_0_or_xformers = hasattr(vae.decoder, "mid_block") and isinstance(
-                    vae.decoder.mid_block.attentions[0].processor,
-                    (
-                        AttnProcessor2_0,
-                        XFormersAttnProcessor,
-                        LoRAXFormersAttnProcessor,
-                        LoRAAttnProcessor2_0,
-                    ),
-                )
-                # if xformers or torch_2_0 is used attention block does not need
-                # to be in float32 which can save lots of memory
-                if use_torch_2_0_or_xformers:
-                    vae.post_quant_conv.to(latents.dtype)
-                    vae.decoder.conv_in.to(latents.dtype)
-                    vae.decoder.mid_block.to(latents.dtype)
-                else:
-                    latents = latents.float()
-
+                latents = latents.float()
             else:
                 vae.to(dtype=torch.float16)
                 latents = latents.half()

invokeai/app/invocations/z_image_denoise.py

@@ -17,13 +17,9 @@
     Input,
     InputField,
     LatentsField,
-    WithBoard,
-    WithMetadata,
     ZImageConditioningField,
 )
-from invokeai.app.invocations.z_image_control import ZImageControlField
-from invokeai.app.invocations.z_image_image_to_latents import ZImageImageToLatentsInvocation
-from invokeai.app.invocations.model import LoRAField, TransformerField, VAEField
+from invokeai.app.invocations.model import TransformerField
 from invokeai.app.invocations.primitives import LatentsOutput
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelFormat
@@ -43,10 +39,10 @@
     title="Denoise - Z-Image",
     tags=["image", "z-image"],
     category="image",
-    version="1.0.0",
+    version="1.1.0",
     classification=Classification.Prototype,
 )
-class ZImageDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
+class ZImageDenoiseInvocation(BaseInvocation):
     """Run the denoising process with a Z-Image model."""
 
     # If latents is provided, this means we are doing image-to-image.

invokeai/app/invocations/z_image_latents_to_image.py

@@ -55,9 +55,7 @@ def invoke(self, context: InvocationContext) -> ImageOutput:
 
         # FLUX VAE doesn't support seamless, so only apply for AutoencoderKL
         seamless_context = (
-            nullcontext()
-            if is_flux_vae
-            else SeamlessExt.static_patch_model(vae_info.model, self.vae.seamless_axes)
+            nullcontext() if is_flux_vae else SeamlessExt.static_patch_model(vae_info.model, self.vae.seamless_axes)
         )
 
         with seamless_context, vae_info.model_on_device() as (_, vae):

invokeai/backend/model_manager/configs/factory.py

@@ -73,17 +73,17 @@
     Main_GGUF_ZImage_Config,
     MainModelDefaultSettings,
 )
+from invokeai.backend.model_manager.configs.qwen3_encoder import (
+    Qwen3Encoder_Checkpoint_Config,
+    Qwen3Encoder_GGUF_Config,
+    Qwen3Encoder_Qwen3Encoder_Config,
+)
 from invokeai.backend.model_manager.configs.siglip import SigLIP_Diffusers_Config
 from invokeai.backend.model_manager.configs.spandrel import Spandrel_Checkpoint_Config
 from invokeai.backend.model_manager.configs.t2i_adapter import (
     T2IAdapter_Diffusers_SD1_Config,
     T2IAdapter_Diffusers_SDXL_Config,
 )
-from invokeai.backend.model_manager.configs.qwen3_encoder import (
-    Qwen3Encoder_Checkpoint_Config,
-    Qwen3Encoder_GGUF_Config,
-    Qwen3Encoder_Qwen3Encoder_Config,
-)
 from invokeai.backend.model_manager.configs.t5_encoder import T5Encoder_BnBLLMint8_Config, T5Encoder_T5Encoder_Config
 from invokeai.backend.model_manager.configs.textual_inversion import (
     TI_File_SD1_Config,

invokeai/backend/model_manager/configs/qwen3_encoder.py

@@ -3,7 +3,6 @@
 from pydantic import Field
 
 from invokeai.backend.model_manager.configs.base import Checkpoint_Config_Base, Config_Base
-from invokeai.backend.quantization.gguf.ggml_tensor import GGMLTensor
 from invokeai.backend.model_manager.configs.identification_utils import (
     NotAMatchError,
     raise_for_class_name,
@@ -13,6 +12,7 @@
 )
 from invokeai.backend.model_manager.model_on_disk import ModelOnDisk
 from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelFormat, ModelType
+from invokeai.backend.quantization.gguf.ggml_tensor import GGMLTensor
 
 
 def _has_qwen3_keys(state_dict: dict[str | int, Any]) -> bool:

invokeai/backend/model_manager/load/model_cache/torch_module_autocast/custom_modules/custom_diffusers_rms_norm.py

@@ -1,5 +1,4 @@
 import torch
-
 from diffusers.models.normalization import RMSNorm as DiffusersRMSNorm
 
 from invokeai.backend.model_manager.load.model_cache.torch_module_autocast.cast_to_device import cast_to_device

invokeai/backend/model_manager/load/model_loaders/z_image.py

@@ -6,7 +6,6 @@
 
 import accelerate
 import torch
-
 from transformers import AutoTokenizer, Qwen3ForCausalLM
 
 from invokeai.backend.model_manager.configs.base import Checkpoint_Config_Base, Diffusers_Config_Base
@@ -288,8 +287,7 @@ def _load_from_singlefile(
 
         if not isinstance(config, Main_GGUF_ZImage_Config):
             raise TypeError(
-                f"Expected Main_GGUF_ZImage_Config, got {type(config).__name__}. "
-                "Model configuration type mismatch."
+                f"Expected Main_GGUF_ZImage_Config, got {type(config).__name__}. Model configuration type mismatch."
             )
         model_path = Path(config.path)
 
@@ -511,10 +509,11 @@ def _load_from_singlefile(
         self,
         config: AnyModelConfig,
     ) -> AnyModel:
-        from invokeai.backend.util.logging import InvokeAILogger
         from safetensors.torch import load_file
         from transformers import Qwen3Config, Qwen3ForCausalLM
 
+        from invokeai.backend.util.logging import InvokeAILogger
+
         logger = InvokeAILogger.get_logger(self.__class__.__name__)
 
         if not isinstance(config, Qwen3Encoder_Checkpoint_Config):
@@ -670,15 +669,15 @@ def _load_from_gguf(
         self,
         config: AnyModelConfig,
     ) -> AnyModel:
-        from invokeai.backend.util.logging import InvokeAILogger
         from transformers import Qwen3Config, Qwen3ForCausalLM
 
+        from invokeai.backend.util.logging import InvokeAILogger
+
         logger = InvokeAILogger.get_logger(self.__class__.__name__)
 
         if not isinstance(config, Qwen3Encoder_GGUF_Config):
             raise TypeError(
-                f"Expected Qwen3Encoder_GGUF_Config, got {type(config).__name__}. "
-                "Model configuration type mismatch."
+                f"Expected Qwen3Encoder_GGUF_Config, got {type(config).__name__}. Model configuration type mismatch."
             )
         model_path = Path(config.path)
 

invokeai/backend/patches/layer_patcher.py

@@ -175,9 +175,7 @@ def _apply_model_layer_patch(
 
         # TODO(ryand): Using torch.autocast(...) over explicit casting may offer a speed benefit on CUDA
         # devices here. Experimentally, it was found to be very slow on CPU. More investigation needed.
-        params_dict = patch.get_parameters(
-            dict(module_to_patch.named_parameters(recurse=False)), weight=patch_weight
-        )
+        params_dict = patch.get_parameters(dict(module_to_patch.named_parameters(recurse=False)), weight=patch_weight)
         if not params_dict:
             logger = InvokeAILogger.get_logger(LayerPatcher.__name__)
             logger.warning(f"LoRA patch returned no parameters for module: {module_to_patch_key}")

invokeai/backend/patches/lora_conversions/z_image_lora_conversion_utils.py

@@ -28,17 +28,18 @@ def is_state_dict_likely_z_image_lora(state_dict: dict[str | int, torch.Tensor])
     # Check for Z-Image transformer keys (S3-DiT architecture)
     # Various training frameworks use different prefixes
     has_transformer_keys = any(
-        k.startswith((
-            "transformer.",
-            "base_model.model.transformer.",
-            "diffusion_model.",
-        )) for k in str_keys
+        k.startswith(
+            (
+                "transformer.",
+                "base_model.model.transformer.",
+                "diffusion_model.",
+            )
+        )
+        for k in str_keys
     )
 
     # Check for Qwen3 text encoder keys
-    has_qwen3_keys = any(
-        k.startswith(("text_encoder.", "base_model.model.text_encoder.")) for k in str_keys
-    )
+    has_qwen3_keys = any(k.startswith(("text_encoder.", "base_model.model.text_encoder.")) for k in str_keys)
 
     return has_transformer_keys or has_qwen3_keys
 
@@ -96,15 +97,15 @@ def lora_model_from_z_image_state_dict(
         # Check and strip text encoder prefixes first
         for prefix in text_encoder_prefixes:
             if layer_key.startswith(prefix):
-                clean_key = layer_key[len(prefix):]
+                clean_key = layer_key[len(prefix) :]
                 is_text_encoder = True
                 break
 
         # If not text encoder, check transformer prefixes
         if not is_text_encoder:
             for prefix in transformer_prefixes:
                 if layer_key.startswith(prefix):
-                    clean_key = layer_key[len(prefix):]
+                    clean_key = layer_key[len(prefix) :]
                     break
 
         # Apply the appropriate internal prefix

invokeai/frontend/web/public/locales/en.json

@@ -501,6 +501,14 @@
                 "title": "Next Prompt in History",
                 "desc": "When the prompt is focused, move to the next (newer) prompt in your history."
             },
+            "promptWeightUp": {
+                "title": "Increase Weight of Prompt Selection",
+                "desc": "When the prompt is focused and text is selected, increase the weight of the selected prompt."
+            },
+            "promptWeightDown": {
+                "title": "Decrease Weight of Prompt Selection",
+                "desc": "When the prompt is focused and text is selected, decrease the weight of the selected prompt."
+            },
             "toggleLeftPanel": {
                 "title": "Toggle Left Panel",
                 "desc": "Show or hide the left panel."