Class LTXVGemmaCLIPModelLoader

gemma_encoder.py:217–275

@comfy_node(name="LTXVGemmaCLIPModelLoader", description="Gemma 3 Model Loader")
class LTXVGemmaCLIPModelLoader:
    @classmethod
    def INPUT_TYPES(s):
        return {
            "required": {
                "gemma_path": (
                    folder_paths.get_filename_list("text_encoders"),
                    {"tooltip": "The name of the text encoder model to load."},
                ),
                "ltxv_path": (
                    folder_paths.get_filename_list("checkpoints"),
                    {"tooltip": "The name of the ltxv model to load."},
                ),
                "max_length": (
                    "INT",
                    {"default": 1024, "min": 16, "max": 131072, "step": 8},
                ),
            }
        }

    RETURN_TYPES = ("CLIP",)
    RETURN_NAMES = ("clip",)
    FUNCTION = "load_model"
    CATEGORY = "lightricks/LTXV"
    TITLE = "LTXV Gemma CLIP Loader"
    OUTPUT_NODE = False

    def load_model(self, gemma_path: str, ltxv_path: str, max_length: int):
        path = Path(folder_paths.get_full_path("text_encoders", gemma_path))
        model_root = path.parents[1]
        tokenizer_path = Path(find_matching_dir(model_root, "tokenizer.model"))
        gemma_model_path = Path(find_matching_dir(model_root, "model*.safetensors"))
        processor_path = Path(find_matching_dir(model_root, "preprocessor_config.json"))
        tokenizer_class = ltxv_gemma_tokenizer(tokenizer_path, max_length=max_length)

        processor = None
        try:
            image_processor = AutoImageProcessor.from_pretrained(
                str(processor_path),
                local_files_only=True,
            )
            processor = Gemma3Processor(
                image_processor=image_processor,
                tokenizer=tokenizer_class().tokenizer,
            )
            logger.info(f"Loaded processor from {model_root} - enhancement enabled")
        except Exception as e:
            logger.warning(f"Could not load processor from {model_root}: {e}")

        clip_dtype = torch.bfloat16
        ltxv_full_path = folder_paths.get_full_path("checkpoints", ltxv_path)
        clip_target = comfy.supported_models_base.ClipTarget(
            tokenizer=tokenizer_class,
            clip=ltxv_gemma_clip(
                gemma_model_path, ltxv_full_path, processor=processor, dtype=clip_dtype
            ),
        )