萌宝AI·电商设置

Generating a product page isn't one prompt, it's a brief: who the customer is, what the copy says, how many images, what ratio, whether a model appears, how the type should look, whether to slip in a few "reverse" back-of-pack shots. Type that as prose every time and you'll drift - image three will have a model in it when you didn't want one, and half the set will invent a certification badge that doesn't exist.

MengBaoEcommerceSettings (萌宝AI·电商设置) is a form that compiles to a prompt. Fill in the fields once, and it emits a structured instruction set plus machine-readable settings for the rest of the graph to consume. It's the least glamorous node in the pack and arguably the most useful for anyone producing listing images weekly.

What you fill in

Four free-text fields - product_name, copy_information, special_requirements, product_size - plus seven dropdowns: language (Chinese, English, German, French, Japanese, Korean, Portuguese), quantity (1–25), aspect_ratio (1:1 through 9:16), usage (main image, main + detail, detail page, poster, seeding image, other), page_content (concise / medium / rich), font_style (twelve options from minimal sans-serif to vintage art), and reverse_pages and model_setting / model_appearance_count.

The compiler does real work with the last three, and the rules are visible in the source:

• Reverse pages (the back-of-pack / lifestyle shots) are only allowed when usage includes 详情页 - detail page. Otherwise you get zero regardless of what you asked for. The ceiling is min(3, quantity // 4), so a 4-image set can't have one and a 16-image set caps at three. On auto, it recommends zero under five images, one under ten, two under sixteen, three above that.
• Model appearance clamps to the quantity and is nulled entirely when model_setting is "no model".
• page_content carries an explicit density directive - concise means one hero subject per image, rich allows two to four support groups but forbids inventing data to fill space.

The anti-hallucination clause

This is the bit worth stealing even if you don't install the pack. When a field is blank, the generated prompt says so and forbids the model from filling the gap: missing copy becomes "only use verifiable product facts, do not write unverified claims or specs", missing size becomes "do not infer real product dimensions from photo pixels or whitespace", and it ends with a line stating that all selling points, parameters, certifications, test data, materials, sales volume and usage effects must come from user-supplied information or verifiable fact.

Anyone who has shipped a listing image with an invented "clinically proven" badge will understand why each of those clauses exists.

Inputs and outputs

Everything is a widget; nothing to wire in. Four outputs:

• user_prompt - the compiled STRING, the one you route into MengBaoImageAPI's prompt
• aspect_ratio - a STRING echoing the selection (9:16 and friends)
• quantity - an INT, wire it straight into batch_size on the generation node
• settings_json - the resolved settings as indented JSON, including the internally computed reverse_page_mode, reverse_page_count and model_appearance_count

user_prompt → prompt and quantity → batch_size is the honest wiring, and it's type-clean. The generation node's own enum widgets (aspect ratio, resolution, quality) still control the actual output geometry - the aspect_ratio output is there for logging or for nodes that take a string, not for magically setting the model. Set the ratio on the generator to match.

Installing it

Part of ComfyUI-MengBaoAI (Corkery520, 17 nodes, MIT, Registry id mengbaoai). Manager: search MengBaoAI, mengbaoai, 萌宝AI. CLI: comfy node install mengbaoai. Git while the Registry listing is pending:

cd ComfyUI/custom_nodes
git clone https://github.com/Corkery520/ComfyUI-MengBaoAI.git
cd ComfyUI-MengBaoAI
python -m pip install -r requirements.txt

numpy, Pillow, requests - this node is pure Python string assembly, so the install cost is the pack's, not this node's. Restart ComfyUI, then search 电商设置, MengBaoEcommerceSettings, or 萌宝. Note the field labels are Chinese-first and follow ComfyUI's locale setting; the values stay Chinese in the compiled prompt (the output language directive handles what language appears in the images).

Things that bite

• It generates a brief, not images. Nothing here calls an API. Pair it with a generation node - and one that can actually honour "8 consistent images", which means a cloud model or a very disciplined local workflow with a fixed seed and reference image.
• quantity of 25 is a big bill. It's an INT wired to batch_size, and batch_size is a loop of single billable API calls. Start at three.
• Consistency is still on you. The prompt demands stable product appearance, structure and packaging text across the set. The generator does its best; check the results, and use the replica chain if you need a specific product to survive across pages.
• The blank-field clauses are protective, not magic. The model still occasionally invents. If a claim is legally sensitive, put it in special_requirements and verify the output before publishing.