refactor(lora): remove the lora_apply_te setting

Text encoder networks now apply unconditionally in the native path.

- remove the option, the per-request parameter and the control threading
- collapse activate_filtered into plain activate at all call sites
- drop the toggle from the prompt embed cache key
- register the retired key so existing configs load without warnings

modules/control/run.py

@@ -403,7 +403,7 @@ def control_run(state: str = '', # pylint: disable=keyword-arg-before-vararg
                 sequential_seed: bool | None = None,
                 # prompt/attention overrides
                 prompt_attention: str | None = None, prompt_mean_norm: bool | None = None, diffusers_zeros_prompt_pad: bool | None = None,
-                te_pooled_embeds: bool | None = None, lora_apply_te: bool | None = None, te_complex_human_instruction: str | None = None, te_use_mask: bool | None = None,
+                te_pooled_embeds: bool | None = None, te_complex_human_instruction: str | None = None, te_use_mask: bool | None = None,
                 # generation modifier overrides (hijack)
                 freeu_enabled: bool | None = None, freeu_b1: float | None = None, freeu_b2: float | None = None, freeu_s1: float | None = None, freeu_s2: float | None = None,
                 hypertile_unet_enabled: bool | None = None, hypertile_hires_only: bool | None = None, hypertile_unet_tile: int | None = None, hypertile_unet_min_tile: int | None = None,
@@ -586,7 +586,7 @@ def control_run(state: str = '', # pylint: disable=keyword-arg-before-vararg
         # prompt/attention overrides
         prompt_attention=prompt_attention, prompt_mean_norm=prompt_mean_norm,
         diffusers_zeros_prompt_pad=diffusers_zeros_prompt_pad, te_pooled_embeds=te_pooled_embeds,
-        lora_apply_te=lora_apply_te, te_complex_human_instruction=te_complex_human_instruction, te_use_mask=te_use_mask,
+        te_complex_human_instruction=te_complex_human_instruction, te_use_mask=te_use_mask,
         # generation modifier overrides (hijack)
         freeu_enabled=freeu_enabled, freeu_b1=freeu_b1, freeu_b2=freeu_b2, freeu_s1=freeu_s1, freeu_s2=freeu_s2,
         hypertile_unet_enabled=hypertile_unet_enabled, hypertile_hires_only=hypertile_hires_only,

modules/detailer/detailer.py

@@ -321,7 +321,7 @@ class Detailer():
                 pc.disable_extra_networks = True # disable processing_diffusers from handling network activation since its handled here
                 network_same = len(p.network_data.values()) == len(pc.network_data.values()) and all(x == y for x, y in zip(p.network_data.values(), pc.network_data.values()))
                 if not network_same:
-                    extra_networks.activate_filtered(pc, pc.network_data)
+                    extra_networks.activate(pc, pc.network_data)
                 log.debug(f'Detail: model="{i+1}:{name}" item={j+1}/{len(items)} box={item.box} label="{item.label}" score={item.score:.2f} seg={detailer_opt(p, "detailer_segmentation")} network={network_same} prompt="{pc.prompt}"')
                 pc.init_images = [image]
                 pc.image_mask = [item.mask]

modules/extra_networks.py

@@ -121,15 +121,6 @@ def activate(p: StableDiffusionProcessing, extra_network_data: defaultdict[str,
     p.network_data = extra_network_data
 
 
-def activate_filtered(p: StableDiffusionProcessing, extra_network_data: defaultdict[str, list[ExtraNetworkParams]] | None = None, step=0):
-    """activate with text encoder components gated on lora_apply_te; must run before prompt encode so te networks affect embeds"""
-    apply_te = getattr(p, 'lora_apply_te', None)
-    if apply_te is None:
-        apply_te = shared.opts.lora_apply_te
-    exclude = [] if apply_te else ['text_encoder', 'text_encoder_2', 'text_encoder_3']
-    activate(p, extra_network_data, step=step, exclude=exclude)
-
-
 def deactivate(p: StableDiffusionProcessing, extra_network_data: defaultdict[str, list[ExtraNetworkParams]] | None = None, force: bool | None = None):
     """call deactivate for extra networks in extra_network_data in specified order, then call deactivate for all remaining registered networks"""
     if p.disable_extra_networks:

modules/face/faceid.py

@@ -216,7 +216,7 @@ def face_id(
                 p.subseeds = p.all_subseeds[n * p.batch_size:(n+1) * p.batch_size]
                 p.prompts, p.network_data = extra_networks.parse_prompts(p.prompts, p.network_data)
 
-                extra_networks.activate_filtered(p, p.network_data)
+                extra_networks.activate(p, p.network_data)
                 ip_model_dict.update({
                         "prompt": p.prompts[0],
                         "negative_prompt": p.negative_prompts[0],

modules/options_handler.py

@@ -17,7 +17,7 @@ if TYPE_CHECKING:
     import builtins
 
 cmd_opts = cmd_args.parse_args()
-compatibility_opts = ['clip_skip', 'uni_pc_lower_order_final', 'uni_pc_order', 'xformers_options']
+compatibility_opts = ['clip_skip', 'uni_pc_lower_order_final', 'uni_pc_order', 'xformers_options', 'lora_apply_te']
 removed_values = { # a stored choice that no longer exists is kept by validate, so it has to be rewritten or it selects nothing
     'cross_attention_optimization': (['Batch matrix-matrix', 'Dynamic Attention BMM'], 'Scaled-Dot-Product'),
 }

modules/processing_class.py

@@ -254,7 +254,6 @@ class StableDiffusionProcessing:
                  prompt_mean_norm: bool | None = None,
                  diffusers_zeros_prompt_pad: bool | None = None,
                  te_pooled_embeds: bool | None = None,
-                 lora_apply_te: bool | None = None,
                  te_complex_human_instruction: str | None = None,
                  te_use_mask: bool | None = None,
                  # generation modifier overrides (hijack)
@@ -543,7 +542,6 @@ class StableDiffusionProcessing:
         self.prompt_mean_norm = prompt_mean_norm
         self.diffusers_zeros_prompt_pad = diffusers_zeros_prompt_pad
         self.te_pooled_embeds = te_pooled_embeds
-        self.lora_apply_te = lora_apply_te
         self.te_complex_human_instruction = te_complex_human_instruction
         self.te_use_mask = te_use_mask
         # generation modifier overrides (hijack)

modules/processing_diffusers.py

@@ -149,7 +149,7 @@ def process_base(p: processing.StableDiffusionProcessing):
     if 'detailer' in p.ops:
         desc = 'Detail'
     p.prompts, p.network_data = extra_networks.parse_prompts(p.prompts, p.network_data)
-    extra_networks.activate_filtered(p) # networks must patch weights before prompt encode so te loras affect embeds
+    extra_networks.activate(p) # networks must patch weights before prompt encode so te loras affect embeds
     base_args = set_pipeline_args(
         p=p,
         model=shared.sd_model,
@@ -313,7 +313,7 @@ def process_hires(p: processing.StableDiffusionProcessing, output):
                 prompts, p.network_data = extra_networks.parse_prompts(prompts)
                 reset_prompts = True
             if reset_prompts or ('base' in p.skip):
-                extra_networks.activate_filtered(p)
+                extra_networks.activate(p)
 
             hires_args = set_pipeline_args(
                 p=p,

modules/prompt_parser_diffusers.py

@@ -130,11 +130,8 @@ class PromptEmbedder:
         # unpack EN data in case of TE LoRA
         en_data = p.network_data
         en_data = [idx.items for item in en_data.values() for idx in item]
-        apply_te = getattr(p, 'lora_apply_te', None)
-        if apply_te is None:
-            apply_te = shared.opts.lora_apply_te
         effective_batch = 1 if self.allsame else self.batchsize
-        key = str([self.prompts, self.negative_prompts, effective_batch, self.clip_skip, self.steps, en_data, apply_te])
+        key = str([self.prompts, self.negative_prompts, effective_batch, self.clip_skip, self.steps, en_data])
         item = cache.get(key)
         if not item:
             if not any(flatten(emb) for emb in [self.prompt_embeds,

modules/ui_definitions.py

@@ -699,7 +699,6 @@ def create_settings(cmd_opts):
         "lora_apply_tags": OptionInfo(0, "LoRA auto-apply tags", gr.Slider, {"minimum": -1, "maximum": 32, "step": 1}),
 
         "lora_apply_sep": OptionInfo("<h2>Apply method</h2>", "", gr.HTML),
-        "lora_apply_te": OptionInfo(False, "LoRA native apply to text encoder"),
         "lora_fuse_native": OptionInfo(True, "LoRA native fuse with model"),
         "lora_fuse_diffusers": OptionInfo(False, "LoRA diffusers fuse with model"),
 

ui/locale/locale_en.json

@@ -861,7 +861,6 @@
     {"id":"","label":"Load custom Diffusers pipeline","localized":"","hint":"","ui":"settings_huggingface"},
     {"id":"","label":"LoRA force reload always","localized":"","hint":"Forces LoRA networks to reload from storage on every generation, even if already cached.<br>Useful for debugging or when LoRA files are being modified externally.<br>Disable for normal use to benefit from caching.","ui":"settings_lora"},
     {"id":"","label":"LoRA load using Diffusers method","localized":"","hint":"Alternative method uses diffusers built-in LoRA capabilities instead of native SD.Next implementation (may reduce LoRA compatibility)","ui":"settings_lora"},
-    {"id":"","label":"LoRA native apply to text encoder","localized":"","hint":"","ui":"settings_lora"},
     {"id":"","label":"LoRA native fuse with model","localized":"","hint":"Merge LoRA into the model for lower memory usage.<br><br><b style=\"color: #ef4444\">Warning:</b> After removing or switching a LoRA, you may still see its style in generated images. To get a clean model, reload it from the model selector.","ui":"settings_lora"},
     {"id":"","label":"LoRA diffusers fuse with model","localized":"","hint":"Merge LoRA into the model for lower memory usage and torch.compile compatibility.<br><br><b style=\"color: #ef4444\">Warning:</b> After removing or switching a LoRA, you may still see its style in generated images. To get a clean model, reload it from the model selector.","ui":"settings_lora"},
     {"id":"","label":"LoRA quantized apply method","localized":"","hint":"How networks are applied to SDNQ-quantized model weights:<br>- <b>exact</b>: adapters are carried alongside the quantized weights at full precision; apply and removal are exact and the quantized weights are never modified. The carried factors take additional VRAM, growing with adapter rank, size and count<br>- <b>requantize</b>: adapters are merged into the quantized weights, matching the behavior of earlier releases. Uses no additional VRAM (a weight backup for network removal is held in system RAM); on models quantized below 8 bits rounding typically loses much of the adapter effect, with strong adapters retaining more<br><br>With <b>requantize</b> selected, the host rank, calibration and cache options below have no effect.<br><br>Default is <b>exact</b>.","ui":"settings_lora"},