VNCCS Character Selector QWEN

VNCCS 3.0 runs "almost pure Qwen Image Edit 2511 with loras," per the author's own description of the rebuild, plus Anima for portrait generation. Both of those read natural-language prompts far better than the tag-and-token style Illustrious-era workflows use - and this node exists because that shift changes what a "costume override" field should even look like. CharacterAssetSelectorQWEN is the same lookup as the plain Character Selector, rebuilt with multiline text fields for a pipeline that wants sentences, not tags.

What it does differently from the non-QWEN version

Same job: pick a character and a costume, get back the prompt text, seed, paths and character sheet you need downstream. The difference is entirely in the override fields. Where the original Character Selector's face/head/top/bottom/shoes are plain single-line strings suited to swapping a Danbooru-style tag or two, this version's equivalents are multiline, built for writing an actual descriptive sentence - the difference between typing glasses and writing "round wire-frame glasses, slightly tinted." That's a direct match to how Anima's natural-language prompting and Qwen-Image-Edit's instruction-style editing actually respond: they reward specific, written-out description more than compressed tag lists do.

This node also has one output the plain version doesn't: costume_prompt, a separate string specifically describing the outfit, kept apart from the general positive_prompt. That separation matters for an edit-model pipeline where you might want to hand the outfit description to a clothing-specific step independent of the rest of the character prompt.

Inputs and outputs that matter

• character and costume (enum, from your library) - identical lookup keys to the non-QWEN version.
• Optional face, head, top, bottom, shoes (STRING, multiline) - write full descriptions here rather than short tags. Same head/face-for-accessories convention from the README applies: accessories like glasses or a hat belong there specifically so they don't get dropped during later stages like emotion generation.
• Optional extra_negative_prompt (STRING, multiline) and new_costume_name (STRING).
• Outputs: face_details, face_path, sheet_path, positive_prompt, seed, negative_prompt, character_sheet (IMAGE), and the QWEN-specific costume_prompt.

Installing it

Ships with the base pack. ComfyUI Manager: search "VNCCS - Visual Novel Character Creation Suite", install latest, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/AHEKOT/ComfyUI_VNCCS.git, restart.

Common issues & troubleshooting

Not sure whether to use this node or the plain Character Selector. Match it to your generation pipeline, not personal taste - if you're on VNCCS 3.0's Qwen/Anima path, this is the one built for it; if you're still running an Illustrious-based pipeline with tag-style prompting, the plain CharacterAssetSelector fits that better.

Sentence-style overrides get partially ignored, or the model latches onto one word instead of the full description. Natural-language prompting on Anima specifically is unforgiving of internal contradictions - it tries to render everything you write literally, including combinations that don't make physical sense, rather than gracefully dropping the impossible part the way an SDXL-style model tends to. Keep multiline descriptions internally consistent rather than stacking conflicting details.

costume_prompt and positive_prompt seem redundant. They serve different consumers downstream - if your workflow only reads positive_prompt, you can ignore costume_prompt entirely; it exists for pipelines that specifically need the outfit description isolated from the rest of the character prompt.