diff --git a/py/noise_injection.py b/py/noise_injection.py index 1d7b042..42ff377 100644 --- a/py/noise_injection.py +++ b/py/noise_injection.py @@ -70,7 +70,7 @@ class LGNoiseInjection: RETURN_TYPES = ("MODEL",) FUNCTION = "apply" - CATEGORY = "🎈LAOGOU/Sampling Utils" + CATEGORY = "advanced/model" DESCRIPTION = "将参考图像的特征（如水珠、纹理等）注入到生成结果中。" def apply(self, model, vae, reference_image, strength, start_percent, end_percent, mask=None): @@ -79,8 +79,8 @@ class LGNoiseInjection: m = model.clone() - # 编码参考图像 - ref_latent = self._encode_reference(vae, reference_image) + # 编码参考图像（使用模型的 latent format） + ref_latent = self._encode_reference(vae, reference_image, model) # 预处理 mask mask_latent = None @@ -113,11 +113,19 @@ class LGNoiseInjection: ref = ref_latent.to(device=cfg_result.device, dtype=cfg_result.dtype) if ref.shape[2:] != cfg_result.shape[2:]: - ref = F.interpolate(ref, size=cfg_result.shape[2:], mode='bilinear', align_corners=False) + # 根据张量维度选择插值模式 + if len(ref.shape) == 5: # 5D张量 [B, C, T, H, W] + ref = F.interpolate(ref, size=cfg_result.shape[2:], mode='trilinear', align_corners=False) + else: # 4D张量 [B, C, H, W] + ref = F.interpolate(ref, size=cfg_result.shape[2:], mode='bilinear', align_corners=False) if ref.shape[0] != cfg_result.shape[0]: if ref.shape[0] == 1: - ref = ref.expand(cfg_result.shape[0], -1, -1, -1) + # 根据张量维度调整expand + if len(ref.shape) == 5: + ref = ref.expand(cfg_result.shape[0], -1, -1, -1, -1) + else: + ref = ref.expand(cfg_result.shape[0], -1, -1, -1) else: ref = ref[:cfg_result.shape[0]] @@ -127,11 +135,19 @@ class LGNoiseInjection: current_mask = mask_latent.to(device=cfg_result.device, dtype=cfg_result.dtype) # 调整 mask 尺寸到 latent 空间 if current_mask.shape[2:] != cfg_result.shape[2:]: - current_mask = F.interpolate(current_mask, size=cfg_result.shape[2:], mode='bilinear', align_corners=False) + # 根据张量维度选择插值模式 + if len(current_mask.shape) == 5: # 5D张量 [B, C, T, H, W] + current_mask = F.interpolate(current_mask, size=cfg_result.shape[2:], mode='trilinear', align_corners=False) + else: # 4D张量 [B, C, H, W] + current_mask = F.interpolate(current_mask, size=cfg_result.shape[2:], mode='bilinear', align_corners=False) # 调整 batch size if current_mask.shape[0] != cfg_result.shape[0]: if current_mask.shape[0] == 1: - current_mask = current_mask.expand(cfg_result.shape[0], -1, -1, -1) + # 根据张量维度调整expand + if len(current_mask.shape) == 5: + current_mask = current_mask.expand(cfg_result.shape[0], -1, -1, -1, -1) + else: + current_mask = current_mask.expand(cfg_result.shape[0], -1, -1, -1) else: current_mask = current_mask[:cfg_result.shape[0]] @@ -170,11 +186,13 @@ class LGNoiseInjection: return (m,) - def _encode_reference(self, vae, reference_image): + def _encode_reference(self, vae, reference_image, model): """编码参考图像""" loaded_models = comfy.model_management.loaded_models(only_currently_used=True) latent = vae.encode(reference_image) - latent = comfy.latent_formats.Flux().process_in(latent) + # 使用模型自己的 latent format，而不是硬编码 Flux + latent_format = model.get_model_object("latent_format") + latent = latent_format.process_in(latent) comfy.model_management.load_models_gpu(loaded_models) logging.warning(f"[FeatureInj] Reference encoded: shape={latent.shape}") return latent @@ -242,7 +260,7 @@ class LGNoiseInjectionLatent: RETURN_TYPES = ("MODEL",) FUNCTION = "apply" - CATEGORY = "🎈LAOGOU/Sampling Utils" + CATEGORY = "advanced/model" DESCRIPTION = "直接输入 latent 进行特征注入，自动使用 latent 的 noise_mask 作为遮罩。" def apply(self, model, reference_latent, strength, start_percent, end_percent): @@ -288,11 +306,19 @@ class LGNoiseInjectionLatent: ref = ref_latent.to(device=cfg_result.device, dtype=cfg_result.dtype) if ref.shape[2:] != cfg_result.shape[2:]: - ref = F.interpolate(ref, size=cfg_result.shape[2:], mode='bilinear', align_corners=False) + # 根据张量维度选择插值模式 + if len(ref.shape) == 5: # 5D张量 [B, C, T, H, W] + ref = F.interpolate(ref, size=cfg_result.shape[2:], mode='trilinear', align_corners=False) + else: # 4D张量 [B, C, H, W] + ref = F.interpolate(ref, size=cfg_result.shape[2:], mode='bilinear', align_corners=False) if ref.shape[0] != cfg_result.shape[0]: if ref.shape[0] == 1: - ref = ref.expand(cfg_result.shape[0], -1, -1, -1) + # 根据张量维度调整expand + if len(ref.shape) == 5: + ref = ref.expand(cfg_result.shape[0], -1, -1, -1, -1) + else: + ref = ref.expand(cfg_result.shape[0], -1, -1, -1) else: ref = ref[:cfg_result.shape[0]] @@ -302,11 +328,19 @@ class LGNoiseInjectionLatent: current_mask = mask_latent.to(device=cfg_result.device, dtype=cfg_result.dtype) # 调整 mask 尺寸到 latent 空间 if current_mask.shape[2:] != cfg_result.shape[2:]: - current_mask = F.interpolate(current_mask, size=cfg_result.shape[2:], mode='bilinear', align_corners=False) + # 根据张量维度选择插值模式 + if len(current_mask.shape) == 5: # 5D张量 [B, C, T, H, W] + current_mask = F.interpolate(current_mask, size=cfg_result.shape[2:], mode='trilinear', align_corners=False) + else: # 4D张量 [B, C, H, W] + current_mask = F.interpolate(current_mask, size=cfg_result.shape[2:], mode='bilinear', align_corners=False) # 调整 batch size if current_mask.shape[0] != cfg_result.shape[0]: if current_mask.shape[0] == 1: - current_mask = current_mask.expand(cfg_result.shape[0], -1, -1, -1) + # 根据张量维度调整expand + if len(current_mask.shape) == 5: + current_mask = current_mask.expand(cfg_result.shape[0], -1, -1, -1, -1) + else: + current_mask = current_mask.expand(cfg_result.shape[0], -1, -1, -1) else: current_mask = current_mask[:cfg_result.shape[0]] diff --git a/py/sigmas_editor.py b/py/sigmas_editor.py index 8daf3a1..c3dd27e 100644 --- a/py/sigmas_editor.py +++ b/py/sigmas_editor.py @@ -73,18 +73,20 @@ class SigmasEditor: # Send sigmas data to frontend via PromptServer (只有输入sigmas改变时才发送) if unique_id is not None: # 只根据输入的sigmas创建缓存键（不包括adjustments） - current_sigmas_key = tuple(sigmas_np.tolist()) + # 使用 hash(sigmas_np.tobytes()) 避免浮点数精度比较问题 + current_sigmas_key = hash(sigmas_np.tobytes()) # 检查是否与上次输入的sigmas相同 last_sigmas_key = self._last_sent_data.get(unique_id) # 只有输入sigmas改变时才发送数据到前端 if last_sigmas_key != current_sigmas_key: + # 当输入sigmas改变时，不发送旧的调整值，让前端重置为原始值 PromptServer.instance.send_sync("sigmas_editor_update", { "node_id": unique_id, "sigmas_data": { "original": sigmas_np.tolist(), - "adjusted": adjusted_sigmas.tolist(), + "adjusted": None, # 重置为None，让前端使用新的原始值 } }) diff --git a/pyproject.toml b/pyproject.toml index 2fa6d4b..ddb0524 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -1,7 +1,7 @@ [project] name = "comfyui_lg_samplingutils" description = "This is a toolset designed for ComfyUI by LAOGOU-666, providing a series of practical sampling nodes, making our operation more intuitive and convenient" -version = "1.0.1" +version = "1.0.2" license = { text = "LICENSE.txt" } dependencies = [""]