JoyCaptionAlpha2Online(NYJY)

JoyCaption Alpha Two is the sequel to the captioner the training community crowned the best in its class - and where Alpha One gave you three caption styles, Alpha Two goes nine, including some you'd never expect from a caption model: "MidJourney", "Product Listing", "Social Media Post", "Art Critic". This node gives you all of that through the online route: image in, caption out, no model downloads, no VRAM cost. If you're building a dataset and want to sample how the same image reads under different caption personas, this is the fastest way to do it.

How it works

Same architecture as the Alpha One online node: it speaks the Gradio protocol to the fancyfeast-joy-caption-alpha-two.hf.space Space - upload the image, join the queue, poll until done, return the text. It's free because it's a shared public Space, and it's subject to the same quotas and queues that all shared Spaces are. If you've used the Alpha One variant, everything about the mechanics is identical except the Space and the richer options.

The inputs that matter

• image - the IMAGE to caption.
• caption_type - the headline feature. Nine choices: Descriptive (default), Descriptive (Informal), Training Prompt, MidJourney (writes like a Midjourney prompt), Booru tag list / Booru-like tag list (tag-soup output for the Danbooru-prompting crowd), Art Critic, Product Listing, and Social Media Post. Each is a genuinely different caption persona - great for A/B-ing what your training captions should sound like.
• caption_length - same control as Alpha One: any through very long, plus exact word counts like 20–70 and beyond.
• extra_options (optional) - an input of the pack's JoyCaption2ExtraOption type. It's the advanced-options object for the Space; for normal use leave it disconnected and you get the Space's defaults.

Output: captions, a STRING.

Installing

Standard pack install: ComfyUI Manager → Install via Git URL → https://github.com/aidenli/ComfyUI_NYJY, restart. No downloads, no config.

Common issues & honest caveats

The shared-Space economics apply again: daily quotas (the author's docs mention roughly 20-ish requests/day for the online JoyCaption family), possible queue waits, and your image leaving your machine for Hugging Face's servers. If you hit the quota mid-dataset, the local JoyCaption node in this pack is the no-quota path - same captioning line, heavier footprint.

The honest take: Alpha Two's nine caption types make it the best exploration tool in the pack - it answers "how would a social-media prompt differ from a training caption for this image?" in one run. But for brute-force captioning of a thousand-image training set, you'll want a local setup with the type you chose already locked in. Use this one to find your caption voice, then scale where it's cheap.