excel_qwen_font

Generating good "3D text" or stylized typography renders needs surprisingly specific vocabulary - the difference between a prompt that gets you a clean embossed logo versus a mushy blob often comes down to naming the right texture, font family, and structural treatment. excel_qwen_font is a curated-vocabulary picker for exactly that: instead of hand-typing (and remembering) precise texture and structure terms, you select from three genuinely large dropdowns and it assembles a prompt for you.

It's marked deprecated in the pack now, and despite the "excel" in its name, it doesn't touch a spreadsheet at all - it's a pure prompt-assembly node that appears to have been filed under the excel naming family and later moved to Deprecated without a direct successor visible among this pack's current nodes. If you relied on it, expect to either hand-write the descriptive vocabulary yourself going forward, or check the pack's active /prompt category for whatever replaced it.

How it works

Three dropdown fields do the heavy lifting, and their sheer size tells you what kind of tool this is:

• text_effect - 310 choices covering surface textures: marble, oak or peach wood grain, brushed metal, smooth or embossed leather, linen or denim cloth, concrete, brick, and hundreds more.
• text_font - 190 choices spanning both Chinese and Latin typefaces: Microsoft YaHei, Heiti, Arial, Calibri, SimSun, various bold and light weights, and many others.
• text_array - 117 choices for 3D structural treatment: extrusion, beveled edges, multi-layer stacking, boolean-operation forms, shell structures, oblique cuts, twisted forms, hollow perforation, embossed relief, spherical wrapping, and more.

Around those, prefix_text and suffix_text let you bookend the assembled description, and text_medium/text_show carry additional text content, with a separate neg field for negative-prompt text. Everything combines into two outputs: pos and neg.

The inputs and outputs that matter

• prefix_text (multiline STRING) and suffix_text - text that wraps around the assembled descriptors.
• text_effect, text_font, text_array - the three large curated dropdowns described above; all three field labels and their option lists display in Chinese in the node UI.
• text_medium, text_show - additional string fields for text content.
• neg (STRING) - negative-prompt text you supply directly.
• Outputs: pos and neg (STRING) - the assembled positive and negative prompts.

How to install it

Bundled with the full pack. Via ComfyUI Manager: search "ComfyUI-Apt_Preset" → Install → restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git

Windows: run install.bat from the cloned folder through ComfyUI's embedded Python. Linux/Mac: activate your ComfyUI venv and install requirements from that folder by hand. Restart ComfyUI. No heavy dependencies specific to this node - it's string assembly, not model inference.

Common issues & troubleshooting

Dropdowns are enormous and entirely in Chinese. With 300+, 190+, and 117 options respectively and no English labels visible in the schema, finding the right entry means scrolling and reading Chinese, or working from a translated reference. There's no search-as-you-type behavior documented at the schema level - treat it as a browse, not a lookup.

Node is deprecated and you're not sure what to use instead. No direct successor is visible among this pack's other prompt nodes - excel_Qwen_camera covers camera framing, not typography or texture, so it isn't a substitute. If you need this functionality going forward, either keep using this node as-is (it still runs) or write the descriptive tags into your prompt by hand.

Prompt output feels overloaded or contradictory. With three independently-chosen large vocabularies plus prefix/suffix/medium/show text all concatenating together, it's easy to end up with a prompt describing textures and structures that don't visually make sense together (say, a "hollow perforated" structure combined with a "solid embossed" texture) - the node won't catch that for you, so sanity-check the combination before you commit GPU time to it.