凤希AI - 多视图提示词增强

FxAiMultiViewGenerator is a prompt builder, not a prompt loader. Feed it your character description and it writes a complete, structured multi-view character-sheet instruction in Chinese - the kind of thing that makes Qwen-Image-Edit produce a single canvas with a close-up plus front/side/back views of the same person, all in matching outfit and style. If you've ever typed that by hand into an edit model and watched it improvise extra views or change the costume, this node exists so you don't have to.

How it works is simple template assembly. You give it 提示词 (your outfit/hairstyle/appearance description) and then five booleans that decide which views go on the sheet: 人脸特写 (face close-up), 正面 (front), 右侧面 (right profile), 背面 (back), 左侧面 (left profile) - all default to on. It builds a Chinese prompt that (1) states the character's outfit/hairstyle spec, (2) forces a pure-white background, (3) fixes the canvas layout - close-up first, then full-body views left to right - and (4) forbids extra views, background junk, and split images. It also emits a matching 系统提示词 (system prompt) that hard-rules the edit model to replace the outfit and hairstyle per your spec rather than inherit the reference image's, and to keep every view of the same person identical. That pairing is the key detail: output 1 is the user instruction, output 2 is the system frame around it, and both feed the Qwen edit encode node in the pack's 三视图/四视图 example workflows.

The two outputs are both STRING: 提示词 and 系统提示词. Wire them into the user-prompt and system-prompt inputs of a Qwen-Image-Edit encode node (FxAiQwenEditEnhancedV2 has a 系统提示词 port that takes exactly this). This is squarely the 2026 approach to character sheets - reference image in, plain-language instruction, no ControlNet - as covered in the KB's character-consistency.md and the Qwen-Image-Edit panel.

The caveat is the same one that applies to every character-sheet workflow on Qwen-Edit: the model re-emits the whole frame, so faces drift and unedited pixels move (the 2511 revision mitigates but doesn't kill this). A prompt node can't fix that - it just makes sure the layout instructions are airtight so you're not also fighting the model over how many views to draw. Keep your outfit description specific, because the system prompt tells the model to obey it completely and discard the reference's clothing.

It's part of fxai-toolkit by 凤希AI (fxai.site), under the 凤希AI/图片 menu:

cd ComfyUI/custom_nodes
git clone https://github.com/fxai666/fxai-toolkit

or via ComfyUI Manager (search "fxai-toolkit"), then restart. If you're building multi-view character sheets with Qwen, this is the quickest way to get the prompt half right.