AC_FUN_SUPER_DESIGN_LARGE
An all-in-one sampler with a design-prompt library baked in
- MODEL
- LATENT
- Positive
- Negative
- width
- height
- VAE
AC_FUN_SUPER_DESIGN_LARGE is the architecture-and-landscape variant of this pack's all-in-one sampler. It's the same mega-node idea - checkpoint, up to three LoRAs, text encode, latent, KSampler all in one box - with one extra trick that makes it interesting: it ships with a bundled prompt library aimed at design visualization, and a switch that lets the node read prompts straight out of those files instead of a text box.
The one thing that makes it different
While AC_FUN_SUPER_LARGE takes plain prompt strings, this one adds three widgets: Switch (Manual operation / Auto), PositiveList, and NegativeList. Set Switch to Auto, pick an entry from PositiveList and one from NegativeList, and the node pulls the prompt text from files that live inside the pack folder - roughly 60 positive files under Positive/Architecture, Positive/Factory, Positive/InDoor_AC, and Positive/Landscape, plus 54 negative-prompt files. So you get one-click prompts like "non-linear Zaha Hadid architecture" or "urban planning aerial view" without typing a word.
The fun detail: the filenames are Chinese but the prompt content is English ("A Complex curve-shaped scientific and technological buildings, close to river, ... by Zaha Hadid"). The author clearly curated these for Chinese-speaking designers, but the prompts themselves work on any standard SDXL or SD 1.5 model. You can also drop your own .txt files into those folders - they show up in the dropdowns after a restart.
The rest of the inputs
The remaining widgets are the same load-bearing set as the plain AC_FUN_SUPER_LARGE: boolean Single/Double/Triple for the three LoRA slots, select_model toggling Prompt_to_Image vs Image_to_Image (the latter VAE-encodes the picked image), dispatch Custom vs All_ready (the latter reads the 37-entry resolution preset list), plus seed, steps, cfg, sampler_name, scheduler, and denoise. Outputs are MODEL, LATENT, Positive, Negative, width, height, and VAE - wire LATENT + VAE into a decoder or the pack's save/preview nodes to see anything.
Installing it
Same pack, same install as every other AC_Super node - ComfyUI Manager (search "AC_FUN") or:
cd ComfyUI/custom_nodes
git clone https://github.com/A719689614/ComfyUI_AC_FUNV8Beta1
Restart and you're done. No model downloads; the prompt files come bundled. The README is in Chinese, and it's from Cc啊程, a ComfyUI instructor on Bilibili.
Where people get burned
Two things. First, the same LCM-default trap as its sibling: 10 steps at cfg 1.4 assumes an LCM LoRA. Run it on a plain SDXL model and you'll want 25+ steps and cfg around 5–7. Second, the Auto mode is a black box by design - you get whatever prompt is in the file, and if you don't read Chinese filenames, you're choosing prompts by vibes. That's fine for iteration, but when you land on something you like, switch to Manual to see and control the exact text. And remember this whole family samples inside the node: no inserting a ControlNet between encode and sample. This is the node you reach for when you want to crank out design variations fast, not when you need surgical control.
Inputs (29)
| Name | Type | Default | Description |
|---|---|---|---|
| ckpt_name | COMBO | 0 options: | |
| lora_name_1 | COMBO | 0 options: | |
| strength_model_1 | FLOAT | 1.00-10–10 | — |
| strength_clip_1 | FLOAT | 1.00-10–10 | — |
| lora_name_2 | COMBO | 0 options: | |
| strength_model_2 | FLOAT | 1.00-10–10 | — |
| strength_clip_2 | FLOAT | 1.00-10–10 | — |
| lora_name_3 | COMBO | 0 options: | |
| strength_model_3 | FLOAT | 1.00-10–10 | — |
| strength_clip_3 | FLOAT | 1.00-10–10 | — |
| boolean | COMBO | 3 options: Single, Double, Triple | |
| positive | STRING | (best quality), ((masterpiece)), (highres), illustration, original, extremely detailed,1girl, solo, kashima \(kancolle\), gloves, breasts, hat, epaulettes, grey hair, large breasts, white gloves, smile, military uniform, military, uniform, twintails, tsurime, white background, beret, simple background, looking at viewer, wavy hair, buttons, blush, long hair, frilled sleeves, purple eyes, upper body, frills, long sleeves, neckerchief, red neckerchief | — |
| negative | STRING | (worst quality, low quality, blurry, bad eye, ),(wrong hand, bad anatomy, wrong anatomy, ),(cgi, illustration, cartoon, poorly drawn, watermark),head out of frame, | — |
| resolution | COMBO | 37 options: 512*512, 576*576, 640*640, 640*480, 480*640, 512*768, +31 | |
| dispatch | COMBO | 2 options: Custom, All_ready | |
| width | INT | 51264–5277 | — |
| height | INT | 76864–5277 | — |
| batch_size | INT | 11–64 | — |
| seed | INT | 00–18446744073709550000 | — |
| steps | INT | 101–10000 | — |
| cfg | FLOAT | 1.400–100 | — |
| sampler_name | COMBO | lcm | 44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38 |
| scheduler | COMBO | 9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3 | |
| Switch | COMBO | 2 options: Manual operation, Auto | |
| PositiveList | COMBO | 60 options: 工业设计3D提示词-B1.txt, 建筑设计常规住宅-004.txt, 建筑设计常规住宅-001.txt, 城市规划设计鸟瞰图.txt, 非线性建筑雪地场景.txt, 非线性建筑黄昏场景.txt, +54 | |
| NegativeList | COMBO | 54 options: 摄影系列反向提示词-003.txt, 景观反向提示词-005.txt, 国风反向提示词-001.txt, 景观反向提示词-004.txt, 通用反向提示词002.txt, 3D卡通反向提示词-001.txt, +48 | |
| denoise | FLOAT | 1.000–1 | — |
| select_model | COMBO | 2 options: Prompt_to_Image, Image_to_Image | |
| image | COMBO | 1 options: example.png |
Outputs (7)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| LATENT | LATENT | — |
| Positive | CONDITIONING | — |
| Negative | CONDITIONING | — |
| width | INT | — |
| height | INT | — |
| VAE | VAE | — |