Detection / Image to Prompt — v2.2 (Nodes 2.0)

This workflow is for analyzing images and creating text: it accepts a single image, a folder of images, video frames, or text prompts and themes, and can produce captions, tags, object detections and masks, timeline prompts, prompt variations, lyrics, and chat responses.

Load one image directly or choose a folder or video; for folder or video input, Image Selector lets you choose files or frames before the active image is routed to Detection, Image to Prompt, and Timeline Prompts.

Named Set/Get channels share the image with those groups, while Prompt Variations, Song Lyrics, and Direct Chat use text prompts, themes, or selected system prompts.

Image to Prompt displays image descriptions and comma-separated tags in Show Text; Detection shows textual results, isolated regions, and binary masks.

Timeline Prompts produces time-coded descriptions for video models such as Wan2.1 and LTX-Video; Prompt Variations expands a seed prompt into three variations.

Song Lyrics creates lyrics or poetry from user-supplied themes, and Direct Chat displays responses from the selected language or vision-language model.

Eclipse 4.1.30 is listed as a requirement for the Nodes 2.0 layout.

The workflow requires the ComfyUI_Eclipse node pack.

For batch image tagging, switch to Folder Mode and enter a folder path; bypass Image Selector to process files sequentially, or leave it active to choose files when prompted.

Suggestion · non vérifié

Choose a vision-language model in Smart LM Loader for image descriptions, and enter a query such as “face, hair” in detection_prompt for object detection.

Suggestion · non vérifié

For Docker-based backends, configure Hugging Face, GGUF, or Ollama credentials in config.json; the guide places local GGUF and Transformers models in models/llms/ or models/text_encoders/.

Suggestion · non vérifié

The two parallel Image to Prompt loaders are a demonstration; captioning and tag generation can instead run sequentially in one loader.

No license is specified for this workflow.