VNCCS VL analyzer

VL is Vision-Language: hand this node a picture and it looks at what's being worn and writes you a description of it. That's the plumbing behind two buttons the README talks up - ANALYZE CAPTIONS on the Character Cloner workflow, and the CLONE CLOTHES tab on Clothes Designer, where you upload a photo of an outfit (or a character wearing one) instead of describing every garment by hand. Rather than you typing "school uniform, red ribbon, knee socks" from scratch, VL Analyzer looks at the reference and does a first pass for you.

Where it fits in the pipeline

VNCCS 3.0 generates the base character on Illustrious or Anima, then hands pose and clothing work to Qwen-Image-Edit - the author's own words, asked what runs under the hood: "it's almost pure Qwen Image Edit 2511 with loras." An edit model like that needs a text description of what it's meant to reproduce on your character, and that description is exactly what VL Analyzer produces. It's the step between "here's a photo of an outfit" and "here's the text prompt that clones it."

Worth knowing if you're VRAM-constrained: the description this node writes is comparatively cheap. It's the Qwen-Image-Edit stage downstream that's the heavy part of VNCCS 3.0's pipeline - people running it report around 8GB VRAM as roughly the floor. VL Analyzer isn't where that cost lives; what happens with its output afterward is.

The inputs and output that matter

• image (IMAGE, required) - the reference photo: an outfit on its own, or a character wearing it.
• clothing_tags (STRING, multiline, optional to leave blank) - whatever you already know about the outfit going in. This isn't decorative; it's the anchor the analysis works from rather than starting cold. If you already know it's "sailor uniform, pleated skirt," giving it that head start beats leaving the field empty and hoping it infers everything correctly on its own.
• description (STRING) - the one output. Wire it into whatever prompt or description slot the clothes-cloning step downstream expects.

Notice the naming: clothing_tags in, description out. This isn't a general-purpose image captioner bolted onto the pack - it's scoped specifically to reading clothing off a reference, which is all VNCCS actually needs it for.

One more thing worth flagging: the schema exposes no model picker and no API key field, unlike the Q4/Q5/Q8 quality tier you choose at Control Center or the Illustrious-vs-Anima choice at Character Creator. Whatever vision-language model is doing the looking is fixed by the pack - this isn't a node where you tune the backend. It also doesn't take Control Center's pipe the way every generator node does, so it isn't drawing on that downloaded model set at all.

How to install it

Ships with the main pack:

• ComfyUI Manager - search VNCCS - Visual Novel Character Creation Suite, install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/AHEKOT/ComfyUI_VNCCS.git, then cd ComfyUI_VNCCS_Utils && pip install -r requirements.txt, restart, and let Comfy Manager grab anything else the graph references.

Common issues & troubleshooting

The description comes back vague or generic. Garbage in, garbage out applies directly here - the README's own advice for Character Cloner reference images is the same advice for this node: make sure the picture is good quality and the subject is full-body, "otherwise the model will invent everything that is not visible in the picture." A cropped, blurry, or heavily obscured shot gives VL Analyzer less to work with, and the description will be as thin as what it can actually see.

You typed nothing in clothing_tags and the result feels off. It's an optional field, but it's not pointless - pre-filling it with whatever you already know about the outfit steers the analysis instead of leaving it to guess everything cold. Worth trying before assuming the node itself is the problem.

Nothing happens if you're not cloning from a reference. This node only has something to work with when you're feeding it an actual photo. If you're building an outfit from scratch rather than off a picture, this isn't the tool - that's the Clothes Wizard and the manual tag fields on Clothes Designer.