Sign In

PromptEnhancer Selector

Updated: Apr 25, 2026

tool

Download

1 variant available

Archive Other

2.26 KB

Verified:

Type

Other

Stats

0

Reviews

No reviews yet

Published

Apr 24, 2026

Base Model

Ernie

Hash

AutoV2
CF9530336F

This is a custom node designed to maximize the potential of Ernie-Image's prompt enhancer.

The prompt describes the scene you want to depict in any language. Ernie's prompt enhancer is a tool that structurally rewrites the given prompt using LLM, and it can also translate, making it usable in many languages. This part can be rewritten to suit various purposes. In effect, you can think of it as achieving the same results as LoRA in preprocessing.

Finally, a Simplified Chinese prompt is generated, so connect it to your existing workflow as prompt text for Ernie-Image, Qwen-Image, Z-Image, etc. It should definitely work with Chinese-made TextEncoders like Qwen. Other countries' LLMs should also work, as long as they are reasonably functional. Since Ernie uses it in Mistral, there's no reason why Flux.2 shouldn't work.

The custom node created here processes prompts in the following modes (though not everything will work as intended):

*The model used in this sample is a prototype created using a similar dataset to the one used here, but (as you can see) it's not very good, so there are no plans to release it.

*The translation of the additional prompts was done by ChatGPT. Please let me know if there are any strange Chinese translations.

Usage

Download the zip file and place the resulting folder into the ComfyUI\custom_nodes folder. Then restart ComfyUI. In Add Node, you'll find it under Prompt/ERNIE in the menu; connect it to the text portion of Clip Text Encode. You can also zoom in on the sample image on CIVITAI, drag and drop the image into the ComfyUI window, and a sample workflow will appear.

default (Base)

1. Faithful conversion mode (runaway prevention)

「严格保留用户原意,不添加未明确指定的重要元素,不改变主体关系,不过度扩写。」

The original meaning of the highest level of reservation, no addition of unclear specified important elements, no change of subject matter, no change of degree of duplication.

effect:

-Reduced arbitrary background additions

-Reduced misunderstanding of subject matter

-For LoRA verification

2. Composition improvement mode

「若用户未指定构图,请使用更自然、更有层次感的构图,避免主体始终位于正中央。」

A collection of unspecified materials for young use, the use of natural materials, and a collection of additional layers, and the main body of the evacuation began in the center of the United States.

effect:

-Central fixation avoidance

-Improved monotony

3. Human figure naturalization mode

「人物请使用自然姿势、真实重心、合理肢体动作,并加入符合场景的细微表情。」

The figure uses natural posture, real center of gravity, rational body movements, and detailed scene expressions.

effect:

-Prevent standing up

-Improving expressionless expression

4. Cinematic mode

「可加入电影感镜头语言、景深、光影层次与动态氛围,但保持主题清晰。」

Addition of additional electric shocks to the sound of the camera, deep language, light and sound effects, but the main feature is clear.

effect:

-Movie-like

-Lighting enhancement

5. Product advertising mode

「突出主体商品,背景简洁干净,强调材质、质感与商业广告级灯光。」

Prominent main products, simple background, high quality materials, quality commercial advertising lights.

effect:

-Product image

-For advertising

6. Anime illustration mode

「使用高质量二次元插画风格,线条清晰,色彩鲜明,角色表现力强。」

The use of high-quality two-dimensional painting style, lines and lines are clear, colors are vivid, and the color representation is strong.

effect:

-2D alignment

7. Photo realism

「以真实摄影风格描述,注意镜头焦段、自然肤色、真实材质与光线。」

It is a real film style description, attention to detail, natural color, real material quality and light line.

effect:

-AI less picturesque

8. Background enhancement mode

「若背景未指定,请补充丰富且合理的环境细节,使场景具有空间感。」

If background unspecified, rich and reasonable environment, mission scenery and space feeling.

effect:

-Preventing background spacing

9. Minimum fill mode (high speed/obedient)

「仅补充必要视觉细节,避免冗长描述,保持提示词简洁高效。」

In addition, it is necessary to review the details, avoid redundancy, and maintain the high efficiency of the explanation.

effect:

-long culture suppression

-inference stability

10. Manga panel style mode

「强调情绪表达、动作瞬间与画面张力,如漫画关键分镜。」

Powerful information display, movement momentary screen intensity, and cartoon key division.

effect:

-Manga look

(Bypass)