Hey comfyui world! I’ve crafted a ComfyUI workflow that batch-processes images to generate captions using BLIP Analyze Image and Florence2Run, then saves each image with its own .txt caption file using ImageBatchSaver—perfect for AI training datasets. Check it out on my site: makuta.io/comfyui-batch-captioning.

Workflow Breakdown:

LoadImagesFromFolderKJ (from comfyui-kjnodes): Loads up to 6 images from a folder (e.g., C:\Users\clement\ComfyUI\output\Character02).

BLIP Analyze Image (from was-ns): Generates basic captions (48-96 chars, CPU-friendly).

Florence2Run (from comfyui-florence2): Adds detailed captions (task: "more_detailed_caption", fp16/sdpa).

PreviewImage (comfy-core): Visualizes images for quick checks.

ImageBatchSaver (from the ComfyUI-Batch-Process pack [by Zar4X]—it saves images with companion .txt files per image (e.g., Alix_0001.png + Alix_0001.txt).

Critical Path-Saving Tip:

Connecting LoadImagesFromFolderKJ’s image_path output to ImageBatchSaver’s output_path input auto-derives filenames from input images and saves to ComfyUI’s default output folder (e.g., C:\Users\[user]\ComfyUI\output\Character02). Great for quick tests!

For a custom folder (e.g., a dedicated training dir), skip the connection and manually enter the full filepath in ImageBatchSaver’s output_path widget (e.g., D:\MyDatasets\Character02). This ensures precise control.

Screenshots: [Embed 3-6 images: workflow canvas, LoadImagesFromFolderKJ-to-ImageBatchSaver connection, sample outputs]

Download the JSON and full setup guide at makuta.io/comfyui-batch-captioning.

Install nodes via ComfyUI Manager: search for comfyui-kjnodes, was-ns, comfyui-florence2, batch-process, comfyui-easy-use (for easy showAnything previews).

What do you think? Ideas for LoRA or video batching? Join the discussion on my site or share your remixes! makuta.io/comfyui-batch-captioning

#ComfyUI #StableDiffusion #AICaptions