Detection / Image to Prompt — v2.2

Detection / Image to Prompt — v2.2 is for turning single images, image folders, or video frames into image prompts, tags, object detections, segmentation masks, and bounding-box results, with additional tools for timeline prompts, prompt variations, song lyrics, and interactive چ

Choose single-image mode or folder/video mode, then optionally use Image Selector to choose the images or frames to process.

The selected image is shared with Detection, Image to Prompt, and Timeline Prompts through Set/Get named channels, so groups can be muted without breaking the visible wiring.

Detection accepts a text query such as “face, hair” and can show isolated regions, binary masks, bounding-box text, and OCR output.

Image to Prompt can return detailed descriptions and comma-separated tags; Timeline Prompts can return structured timelines for Wan2.1 and LTX-Video.

Text-only modes can expand a seed prompt into three variations, compose song lyrics or poetry from a theme, and return direct chat responses.

Configure Hugging Face, GGUF, or Ollama credentials in config.json when using Docker-based backends; local GGUF and Transformers backends use models in models/llms/ or models/text_encoders/.

Suggestion · non vérifié

Requires the ComfyUI_Eclipse node pack, which provides the smart loaders, Smart Detection, image loading and selection, Set/Get routing, and preview nodes used here.

Smart LM Loader offers Transformers, GGUF, Ollama, llama.cpp, vLLM (Docker), vLLM (Native), SGLang (Docker), and WD14 Tagger backends.

Add custom model entries under the transformers, gguf, or ollama keys in registry/user_models.json; names beginning with an underscore are ignored.

Suggestion · non vérifié

For batch tagging, set Folder Mode, enter the folder path, and either bypass Image Selector for automatic processing or leave it active to choose files.

Suggestion · non vérifié

The guide says this version was updated for Eclipse 4.1.15 because of a breaking change in Smart LM Loader; verify your installed Eclipse version before use.

Suggestion · non vérifié

The two parallel Image-to-Prompt loader nodes are described as a showcase setup; the guide says captioning and tag generation can instead run sequentially in one loader’s multi-task system.

Suggestion · non vérifié