"""Shared DENO node metadata for ComfyUI object_info and node help UI.""" from __future__ import annotations import copy from pathlib import Path try: import tomllib except ModuleNotFoundError: # pragma: no cover - Python < 3.11 tomllib = None PACKAGE_NAME = "deno-custom-nodes" def get_current_version() -> str: pyproject_path = Path(__file__).with_name("pyproject.toml") if tomllib is None or not pyproject_path.exists(): return "unknown" try: data = tomllib.loads(pyproject_path.read_text(encoding="utf-8")) return str(data.get("project", {}).get("version") or "unknown") except Exception: return "unknown" def _version_description_prefix() -> str: current = get_current_version() return ( f"DENO Custom Nodes v{current}.\n" "The DENO info button checks Comfy Registry and marks this node when an update is available." ) def _with_version_prefix(description: str) -> str: description = str(description or "").strip() prefix = _version_description_prefix() if description.startswith("DENO Custom Nodes v"): return description return f"{prefix}\n\n{description}" if description else prefix NODE_INPUT_TOOLTIPS = { "DenoResolutionSetup": { "mode": "Choose how the output size is decided: preset ratio, manual width/height, or the input image ratio.", "ratio_preset": "Target aspect ratio used in Preset Ratio mode.", "megapixels": "Target total image size in megapixels for automatic sizing modes.", "width": "Manual or fallback output width in pixels.", "height": "Manual or fallback output height in pixels.", "divisible_by": "Round the final size to a model-safe multiple such as 16 or 32.", "resize_method": "Choose center crop, draggable crop-position, or fit/letterbox behavior when resizing an input image.", "interpolation": "Resize filter. Lanczos is the default quality choice.", "crop_x": "Saved horizontal crop position. Drag inside the Resize Box preview to change it.", "crop_y": "Saved vertical crop position. Drag inside the Resize Box preview to change it.", "crop_zoom": "Saved crop zoom. Drag any crop corner while keeping the output aspect ratio locked.", "image": "Optional source image. If empty, the node creates a blank image at the selected size.", }, "DenoMultiImageLoader": { "image_paths": "Selected image file list managed by the node UI.", "mode": "Choose how each image is resized before the batch output is built.", "ratio_preset": "Target aspect ratio used in Preset Ratio mode.", "megapixels": "Target total image size in megapixels for automatic sizing modes.", "width": "Manual or fallback output width in pixels.", "height": "Manual or fallback output height in pixels.", "divisible_by": "Round the final batch size to a model-safe multiple.", "interpolation": "Resize filter. Lanczos is the default quality choice.", "resize_method": "Choose crop-fill or fit/letterbox behavior for loaded images.", }, "DenoMiniMaxH3ReferenceImageLoader": { "image_paths": "Ordered MiniMax H3 reference-image list managed by the node UI.", }, "DenoMiniMaxH3ReferenceToVideo": { # These complete the legacy object_info metadata contract. The V3 # define_schema path keeps upstream H3 tooltips/layout unchanged. "clip": "MiniMax H3 text/image encoder used to tokenize the prompt and references.", "vae": "MiniMax H3 video VAE used to encode reference images and videos.", "audio_vae": "MiniMax H3 audio VAE used to encode reference audio.", "prompt": "Prompt for the generated video. Refer to loaded images as , , and so on.", "width": "Generated video width in pixels.", "height": "Generated video height in pixels.", "length": "Generated frame count at 24 fps.", "ref_image_size": "Keep stock H3 reference sizing: match the generation area or use the larger max reference pipeline.", "ref_images": "Ordered image bundle from the DENO MiniMax H3 Reference Image Loader.", "ref_videos": "Stock MiniMax H3 reference-video slots, in prompt tag order.", "ref_video_audios": "Stock MiniMax H3 soundtrack slots paired by number with reference videos.", "ref_audios": "Stock MiniMax H3 standalone reference-audio slots, in prompt tag order.", }, "DenoMiniMaxH3AccLoader": { "model": "Matching native ComfyUI MiniMax H3 diffusion model to accelerate; full and curve-pruned models are supported.", "acc_lora": "Official Alibaba MiniMax-H3 Acc-LoRA. Match FL2VA with FL2VA/T2VA and Ref2VA with Ref2VA.", }, "DenoAudioTranscript": { "audio": "Source mono or stereo audio to transcribe locally. The node downmixes it to mono and resamples it to 16 kHz for Whisper.", "model": "Local Whisper model size. large-v3 prioritizes lyric and difficult-speech accuracy; large-v3-turbo is the faster default. The official checkpoint downloads on first use to ComfyUI/models/stt/whisper.", "language": "Choose a known language or let Whisper detect it automatically.", "model_after_run": "CUDA Smart Swap unloads ComfyUI models before transcription. Unload also releases Whisper afterward; Keep loaded is an advanced repeated-run option.", "manual_transcript": "Optional exact lyrics or dialogue. When non-empty, this wording is authoritative while Whisper still runs to provide approximate segment timing.", }, "DenoAudioAnalysisFinalize": { "analysis": "Raw Gemma 4 audio analysis from Text Generate. Reasoning chatter is removed and only the supported analysis fields are kept.", "clip": "The same Gemma 4 CLIP value used by Text Generate. It lets the node release only that audio-analysis model after the text is ready.", "model_after_run": "Unload releases only the connected audio-analysis CLIP model. Keep loaded is an advanced option for repeated runs.", }, "DenoTextEncoderUnload": { "positive_conditioning": "Positive conditioning to pass through unchanged after the connected text encoder is unloaded.", "negative_conditioning": "Optional negative conditioning. Accepts an encoded negative prompt or Conditioning Zero Out.", "text_encoder": "The exact CLIP/text encoder to unload. Connect the same CLIP used by every upstream encoding branch.", }, "DenoAdvancedImageSourceLoader": { "image_paths": "Files, folders, absolute paths, or web URLs selected by the advanced source UI.", "mode": "Choose how images are resized for the batch output.", "ratio_preset": "Target aspect ratio used in Preset Ratio mode.", "megapixels": "Target total image size in megapixels for automatic sizing modes.", "width": "Manual or fallback output width in pixels.", "height": "Manual or fallback output height in pixels.", "divisible_by": "Round the final batch size to a model-safe multiple.", "interpolation": "Resize filter. Lanczos is the default quality choice.", "resize_method": "Choose crop-fill, fit, or advanced size handling for loaded images.", "recursive_folders": "Include image files found inside subfolders.", "list_output_mode": "Choose whether the image_list output keeps original sizes or matches the batch size.", "disabled_image_paths": "Saved skipped sources. Managed by the UI when you disable items.", "images": "Optional incoming images to merge with the selected file/path sources.", }, "DenoLTXSequencer": { "positive": "Base positive LTX conditioning before guide images are inserted.", "negative": "Base negative LTX conditioning before guide images are inserted.", "vae": "VAE used to encode the guide/reference images into the LTX latent space.", "latent": "LTX latent timeline that receives the guide image keyframes.", "multi_input": "Batch of guide/reference images to insert into the LTX sequence.", "num_images": "How many images from the batch to use.", "insert_mode": "Choose whether insertion positions are entered as frame numbers or seconds.", "frame_rate": "Frame rate used to convert seconds into frame positions.", "strength_sync": "Keep guide strengths synced across sequencer nodes when the frontend sync tool is used.", "bypass": "Pass positive, negative, and latent through unchanged.", }, "DenoLTX23PresetLoader": { "pipeline_mode": "Choose Checkpoint Style, KJ Style, or GGUF Style loading.", "checkpoint_name": "Full checkpoint file used by Checkpoint Style.", "diffusion_model_name": "LTX diffusion/transformer model used by KJ Style.", "gguf_unet_name": "GGUF LTX model used by GGUF Style.", "video_vae_name": "Video VAE for LTX video latents.", "audio_vae_name": "Audio VAE for LTX audio latents.", "text_encoder_name": "Gemma/text encoder file used to build the CLIP output.", "text_projection_name": "Text projection file used with KJ/GGUF style loading.", "clip_device": "Device choice for the text encoder. Default is recommended for most users.", "weight_dtype": "Weight dtype for KJ diffusion loading. Default is recommended for most users.", }, "DenoLTXModelDownloader": { "model_root": "ComfyUI models root where the helper checks target files.", "presets_json": "Saved preset/package list for the helper panel. Advanced users can edit it manually.", }, "DenoMultiLoraLoader": { "model": "Base diffusion model that enabled LoRAs are applied to.", "clip": "Optional base CLIP model. Connect it when your LoRAs also affect CLIP.", "active_loras": "Number of LoRA slots shown by the frontend UI.", }, "DenoLTXMultiLoraLoader": { "model": "Base LTX model that enabled LoRAs are applied to.", "clip": "Optional CLIP model to pass through or patch when a LoRA affects CLIP.", "active_loras": "Number of LTX LoRA slots shown by the frontend UI.", }, "DenoLTXPromptGuide": { "clip": "LTX text encoder/CLIP used to encode the prompts.", "positive_prompt": "Main editable LTX prompt text.", "language": "Language used only for the dialogue-duration estimate, not translation.", "frame_rate": "Frame-rate value added to the LTX conditioning and returned for downstream timing.", "show_negative_prompt": "Saved Show/Hide state for the negative prompt area.", "negative_prompt": "Editable negative prompt text.", }, "DenoLTXTiledSpatialUpscaler": { "samples": "Video-only LTX latent to upscale. AV combined latents should be separated before this node.", "upscale_model": "LTX latent spatial upscaler model.", "vae": "LTX VAE used for the latent channel statistics around the upscaler.", "horizontal_tiles": "Frame width split count. 2 means left and right tiles.", "vertical_tiles": "Frame height split count. 3 means top, middle, and bottom tiles.", "overlap": "Overlap in input latent tokens. Larger values blend more context but use more time.", "blend_mode": "Overlap weighting curve. Hann is the recommended starting point.", "aggressive_memory_cleanup": "Run extra cleanup between tiles. This can be much slower, but may help fragmented VRAM or OOM cases.", "debug": "Print tile plan and shape diagnostics to the ComfyUI console.", }, "DenoLTXAVStepFusedTiledSampler": { "noise": "Noise source used once for the global AV sampler trajectory.", "guider": "BasicGuider or CFGGuider for the LTX AV second pass, including LTX guide metadata from upstream Sequencer/Add Guide nodes.", "sampler": "ComfyUI sampler object. The node keeps the sampler update global.", "sigmas": "Sigma schedule for the low-denoise AV second-pass refinement.", "latent_image": "LTX AV nested latent containing video and audio. If guides are present, run LTXVSeparateAVLatent after this sampler, then apply LTXVCropGuides to the video latent before decode.", "horizontal_tiles": "Frame width split count. 2 means left and right tiles.", "vertical_tiles": "Frame height split count. 3 means top, middle, and bottom tiles.", "overlap": "Overlap in latent video tokens for each model-prediction tile, including guide-bearing video frames.", "audio_mode": "Freeze keeps audio unchanged while still using it as context for video denoising.", "blend_mode": "Overlap weighting curve. Hann is the recommended starting point.", "aggressive_memory_cleanup": "Run extra cleanup between AV tile predictions. This can be much slower, but may help fragmented VRAM or OOM cases.", "debug": "Print AV hook calls, sigma labels, and tile diagnostics to the ComfyUI console.", }, "DenoBerniniPromptGuide": { "clip": "Wan/Bernini text encoder or CLIP used to encode the prompts.", "task_type": "System Prompt mode for the Bernini/KJ-style instruction prefix.", "positive_prompt": "Editable instruction prompt. Write it like a clear chatbot instruction.", "reference_prompt_helper": "Adds a short image0/image1/image2 reference naming hint for reference modes.", "negative_preset": "Preset used to fill the visible negative prompt box.", "show_negative_prompt": "Saved Show/Hide state for the negative prompt area.", "negative_prompt": "Editable negative prompt text that is actually encoded.", }, "DenoIdeogramDirector": { "width": "Output image width passed downstream.", "height": "Output image height passed downstream.", "seed": "Seed value passed downstream. The Director Fixed/Random control manages it.", "high_level_description": "Short whole-scene summary used in the Ideogram JSON prompt.", "background": "Background or setting description used in the Ideogram JSON prompt.", "style_mode": "Choose no style block, photo style, or art style.", "aesthetics": "General visual mood or aesthetic notes for the style block.", "lighting": "Lighting notes for the style block.", "medium": "Medium/material notes for art-style prompts.", "photo": "Photography notes used when Photo style is active.", "art_style": "Art style notes used when Art style is active.", "backdrop": "Shown on the board as a layout backdrop to trace over. It is not used for generation.", "import_mode": "How connected JSON should handle an existing board.", "caption_data": "Internal saved board state managed by the editor. Do not edit it by hand.", "seed_lock": "Saved Fixed/Random seed state used by the Director UI.", "auto_save": "Save the latest generated result automatically when enabled.", "save_prefix": "Filename prefix used by Save Image / Auto-save.", "aspect_ratio": "Ratio label derived from the selected output size.", "include_aspect_ratio": "Advanced compatibility toggle. Normally off because width/height carry the real shape.", "translate_output": "Final prompt language mode. Board text can stay in your language while output becomes English.", "view_language": "Language used to read and edit the board UI descriptions.", "translation_engine": "Translation engine used when refreshing board language or final English output.", "libretranslate_url": "Custom LibreTranslate server URL, used only when the custom engine is selected.", "import_json": "Optional connected JSON prompt from an upstream LLM or Prompt Text node.", "input_width": "Optional connected output width. Snaps to a multiple of 16 and wins when width and height are both connected.", "input_height": "Optional connected output height. Snaps to a multiple of 16 and wins when width and height are both connected.", "input_megapixels": "Optional connected megapixel budget. Used with the current aspect ratio when width and height are not both connected.", }, "DenoLocalLLMRefiner": { "provider": "Choose the local provider to call: Ollama, LM Studio, llama.cpp, vLLM, Custom, llama-swap, or Unsloth. Select Unsloth only for an Unsloth Studio server; select LM Studio when an Unsloth GGUF is running inside LM Studio.", "ollama_model": "Ollama model name detected from your local Ollama server.", "lm_studio_model": "LM Studio model name detected from your local LM Studio server.", "custom_server_url": "Local OpenAI-compatible server URL used by llama.cpp, vLLM, Custom, llama-swap, or Unsloth Studio. Unsloth defaults to http://127.0.0.1:8888/v1 and requires DENO_LOCAL_LLM_UNSLOTH_API_KEY; the key is never stored in the workflow or PNG metadata.", "custom_model": "Model name to call on the selected local OpenAI-compatible server.", "system_prompt": "Optional instructions that steer how the local LLM rewrites or reviews the prompt.", "thinking": "Ask for model thinking/reasoning when the selected local model supports it. If LM Studio rejects its optional reasoning-control field before output starts, DENO retries once without that field; the server then decides its default reasoning behavior.", "seed": "Seed used for local LLM generation.", "seed_mode": "How the visible seed changes after each queued run.", "model_memory": "Choose whether the local model unloads after the run or stays warm.", "keep_minutes": "Requested keep duration where the provider supports a TTL. For Unsloth it tracks warm-state/swap decisions but does not schedule a timed unload; use Unload after run or Unload LLM. llama-swap uses its own server-side unload timeout.", "comfy_vram_policy": "Choose whether ComfyUI models should be unloaded before local LLM work.", "prompt": "Main prompt text. The in-node textarea and STRING socket feed this same backend input.", "image": "Optional image sent to a vision-capable local model.", "video_seconds": "Optional video length. A positive FLOAT adds an English duration sentence to every user prompt sent to the local LLM.", "audio_context": "Optional source-audio context appended without replacing the user prompt. Only explicitly labeled user-supplied wording is authoritative; automatic transcript and acoustic analysis remain untrusted evidence.", }, "DenoAIReviewGate": { "review": "Text or JSON verdict from a reviewer LLM or text node.", "review_mode": "Review checks the verdict text. Pass bypasses the review gate.", "approve_once": "One-shot approval state controlled by the node button.", "image": "Optional image to pass through only when the review approves it.", "audio": "Optional audio to pass through together with the approved image/media result.", "reviewer_state": "Internal reviewer snapshot used by the frontend buttons.", }, "DenoPromptText": { "text": "Multiline prompt, system prompt, template, or JSON text to connect into other nodes.", }, "DenoRTXVFXEasyUpscale": { "images": "Input image or video frame batch to enhance with NVIDIA RTX VFX.", "mode": "RTX VFX operation: upscale, high bitrate, denoise, or deblur.", "resize_type": "How the output size is chosen for resize-capable modes.", "scale": "Scale multiplier used by scale-based resizing.", "megapixels": "Target total output size in megapixels for ratio-based resizing.", "width": "Manual output width in pixels.", "height": "Manual output height in pixels.", "divisible_by": "Round output dimensions to a VFX/model-safe multiple.", "device": "GPU device index for NVIDIA VFX.", "ratio_preset": "Target aspect ratio used in preset-ratio resize mode.", "resize_method": "Choose crop-fill or fit/letterbox behavior when changing aspect ratio.", }, "DenoRTXVFXVideoFinisher": { "images": "Input video frame batch to finish with one or two RTX VFX passes.", "first_pass": "Optional same-size first pass, usually denoise or deblur.", "first_quality": "Quality/strength for the first pass.", "upscale_pass": "Second pass operation, usually RTX Video Super Resolution.", "upscale_quality": "Quality/strength for the upscale pass.", "resize_type": "How the final output size is chosen.", "scale": "Scale multiplier used by scale-based resizing.", "megapixels": "Target total output size in megapixels for ratio-based resizing.", "width": "Manual output width in pixels.", "height": "Manual output height in pixels.", "divisible_by": "Round output dimensions to a VFX/model-safe multiple.", "ratio_preset": "Target aspect ratio used in preset-ratio resize mode.", "resize_method": "Choose crop-fill or fit/letterbox behavior when changing aspect ratio.", }, "DenoImageCompare": { "mode": "Comparison view: slider, side-by-side, difference, toggle, or swap-oriented view.", "split_position": "Slider split position between image A and image B.", "toggle_image": "Which image is shown first in Toggle mode.", "swap": "Swap A and B before previewing.", "image_a": "First image batch for comparison.", "image_b": "Second image batch for comparison.", }, "DenoVideoCompare": { "mode": "Comparison view for two video frame batches.", "split_position": "Slider split position between video A and video B.", "toggle_image": "Which video is shown first in Toggle mode.", "swap": "Swap A and B before previewing or outputting.", "fps": "Preview frame rate for playback.", "burn_labels": "Burn A/B labels into the output frames.", "video_a": "First video frame batch.", "video_b": "Second video frame batch.", "audio_a": "Optional audio for video A preview sync.", "audio_b": "Optional audio for video B preview sync.", }, "DenoVideoPreview": { "images": "Image frame batch to preview as a video.", "frame_rate": "Playback frame rate for the preview video.", "audio": "Optional audio attached to the temporary preview video.", }, } NODE_OUTPUT_TOOLTIPS = { "DenoResolutionSetup": ( "Resized image, or a blank image if no input image was connected.", "Final output width in pixels.", "Final output height in pixels.", ), "DenoMultiImageLoader": ( "Loaded images resized into one same-size IMAGE batch.", "Final batch width in pixels.", "Final batch height in pixels.", ), "DenoMiniMaxH3ReferenceImageLoader": ( "Ordered MiniMax H3 reference-image bundle with each image's decoded size and aspect ratio preserved.", "The same ordered sources as a mixed-size IMAGE list for nodes such as DENO Local LLM Loader.", ), "DenoMiniMaxH3ReferenceToVideo": ( "Positive MiniMax H3 conditioning containing the ordered image, video, and audio references.", "Empty MiniMax H3 audio/video latent for sampling.", ), "DenoMiniMaxH3AccLoader": ( "MiniMax H3 model patched with the selected Alibaba Acc-LoRA and dynamically scheduled PDD output heads.", ), "DenoAudioTranscript": ( "Structured transcript data with requested/detected language, timestamped segments, and a heuristic confidence band for an LLM prompt builder.", "Effective plain transcript: the exact user-supplied lyrics or dialogue when provided, otherwise the Whisper transcript.", "The original source AUDIO unchanged, for a guaranteed Whisper-before-Gemma Smart Swap chain.", ), "DenoAudioAnalysisFinalize": ( "Canonical-order supported audio-analysis fields with Gemma reasoning and unrelated chatter removed.", ), "DenoTextEncoderUnload": ( "The original positive conditioning unchanged, after the connected CLIP/text encoder and its clones are unloaded.", "The original negative conditioning unchanged when the optional negative input is connected.", ), "DenoAdvancedImageSourceLoader": ( "Loaded sources resized into one same-size IMAGE batch.", "Loaded sources as an IMAGE list for workflows that accept different-sized images.", "Final batch width in pixels.", "Final batch height in pixels.", "Number of images loaded from all active sources.", ), "DenoLTXSequencer": ( "Positive conditioning with guide-image timing inserted.", "Negative conditioning matching the sequenced positive conditioning.", "Latent timeline with guide frames appended.", ), "DenoLTX23PresetLoader": ( "Loaded LTX model.", "Loaded text encoder / CLIP stack.", "Video VAE for LTX video latents.", "Audio VAE for LTX audio latents.", ), "DenoMultiLoraLoader": ( "Model after all enabled LoRA slots are applied.", "CLIP after enabled CLIP LoRA changes, or the original/empty CLIP passthrough.", ), "DenoLTXMultiLoraLoader": ( "LTX model after all enabled LoRA slots are applied.", "CLIP after enabled LoRA changes, or the original/empty CLIP passthrough.", ), "DenoLTXPromptGuide": ( "Encoded positive conditioning with LTX frame-rate metadata.", "Encoded negative conditioning with LTX frame-rate metadata.", "Frame-rate integer passthrough for downstream timing nodes.", ), "DenoLTXTiledSpatialUpscaler": ( "Upscaled video latent reconstructed from overlapping spatial tiles.", ), "DenoLTXAVStepFusedTiledSampler": ( "Refined AV latent. Video is tiled with guide metadata preserved and audio is preserved from the input.", "Denoised AV latent from the callback x0. Video is x0 with guide metadata preserved and audio is preserved from the input.", ), "DenoBerniniPromptGuide": ( "Encoded positive conditioning for the Bernini/KJ workflow.", "Encoded negative conditioning for the Bernini/KJ workflow.", ), "DenoIdeogramDirector": ( "Minified Ideogram JSON prompt to connect into positive CLIPTextEncode.", "Selected output width in pixels.", "Selected output height in pixels.", "Selected seed value.", "Drawn board boxes as downstream BBOX data.", ), "DenoLocalLLMRefiner": ( "Final Result returned by the Local LLM node. When the node executes, its latest Result is embedded in saved workflow metadata; Thinking/reasoning is not persisted. Batched prompts return a STRING list.", ), "DenoAIReviewGate": ( "Original image passed through only when the review approves it.", "Original audio passed through only when the review approves it.", ), "DenoPromptText": ( "The same text as a connectable STRING output.", ), "DenoRTXVFXEasyUpscale": ( "RTX VFX enhanced frame batch.", ), "DenoRTXVFXVideoFinisher": ( "Final RTX VFX enhanced frame batch.", ), "DenoVideoCompare": ( "Composited comparison frame batch.", ), "DenoVideoPreview": ( "Unchanged input frames passed through after preview creation.", ), } def _pattern_tooltip(node_id: str, input_name: str) -> str: if node_id == "DenoLTXSequencer": if input_name.startswith("insert_frame_"): index = input_name.rsplit("_", 1)[-1] return f"Frame position for guide image {index} when Insert Mode is frames." if input_name.startswith("insert_second_"): index = input_name.rsplit("_", 1)[-1] return f"Second position for guide image {index} when Insert Mode is seconds." if input_name.startswith("strength_"): index = input_name.rsplit("_", 1)[-1] return f"Guide strength for image {index}." if node_id == "DenoMultiLoraLoader": return _multi_lora_slot_tooltip(input_name, ltx=False) if node_id == "DenoLTXMultiLoraLoader": return _multi_lora_slot_tooltip(input_name, ltx=True) return "" def _multi_lora_slot_tooltip(input_name: str, ltx: bool) -> str: if "_" not in input_name: return "" prefix, _, raw_index = input_name.rpartition("_") if not raw_index.isdigit(): return "" index = raw_index if prefix == "enabled": return f"Enable or skip LoRA slot {index}." if prefix == "lora": return f"LoRA file used by slot {index}." if prefix == "trigger": return f"Trigger words saved for slot {index}; useful for copying into prompts." if prefix == "description": return f"User notes saved for LoRA slot {index}." if ltx: if prefix == "strength": return f"Overall LoRA strength for LTX slot {index}." if prefix == "audio": return f"Audio-module scale for LTX LoRA slot {index}." if prefix == "video": return f"Video-module scale for LTX LoRA slot {index}." else: if prefix == "model_strength": return f"Model strength for LoRA slot {index}." if prefix == "clip_strength": return f"CLIP strength for LoRA slot {index}." return "" def _tooltip_for(node_id: str, input_name: str) -> str: return NODE_INPUT_TOOLTIPS.get(node_id, {}).get(input_name) or _pattern_tooltip(node_id, input_name) def _inject_input_tooltips(node_id: str, input_types: dict) -> dict: cloned = copy.deepcopy(input_types) for section in ("required", "optional"): entries = cloned.get(section) if not isinstance(entries, dict): continue for input_name, spec in list(entries.items()): tooltip = _tooltip_for(node_id, input_name) if not tooltip: continue if isinstance(spec, tuple): parts = list(spec) if len(parts) >= 2 and isinstance(parts[1], dict): parts[1] = {**parts[1], "tooltip": tooltip} else: parts.append({"tooltip": tooltip}) entries[input_name] = tuple(parts) elif isinstance(spec, list): parts = list(spec) if len(parts) >= 2 and isinstance(parts[1], dict): parts[1] = {**parts[1], "tooltip": tooltip} else: parts.append({"tooltip": tooltip}) entries[input_name] = parts return cloned def decorate_node_classes(node_class_mappings: dict) -> None: for node_id, node_cls in node_class_mappings.items(): node_cls.DESCRIPTION = _with_version_prefix(getattr(node_cls, "DESCRIPTION", "")) output_tooltips = NODE_OUTPUT_TOOLTIPS.get(node_id) if output_tooltips: node_cls.OUTPUT_TOOLTIPS = tuple(output_tooltips) if getattr(node_cls, "_deno_metadata_wrapped", False): continue original_input_types = node_cls.INPUT_TYPES @classmethod def _input_types_with_deno_metadata(cls, _node_id=node_id, _original=original_input_types): return _inject_input_tooltips(_node_id, _original()) node_cls.INPUT_TYPES = _input_types_with_deno_metadata node_cls._deno_metadata_wrapped = True