Hunyuan (混元) text-to-image prompt builder: Tencent Cloud's preset styles and resolutions, a separate negative prompt and a ready API request.
Hunyuan Image Prompt Builder
Write a Tencent Hunyuan (混元) image prompt the way Tencent documents it: style first (or one of its preset styles), a separate negative prompt, one of the eight resolutions the API accepts, and the request parameters. Runs in your browser; no model is called.
What Hunyuan’s image API accepts
From the SubmitHunyuanImageJob parameters in Tencent Cloud’s SDK and Tencent’s Hunyuan style list, read October 2026.
| Value | Ratio |
|---|---|
1024:1024 | 1:1 |
768:768 | 1:1 |
768:1024 | 3:4 |
1024:768 | 4:3 |
720:1280 | 9:16 |
1280:720 | 16:9 |
768:1280 | 3:5 |
1280:768 | 5:3 |
| Style | ID |
|---|---|
| 日漫动画 | riman |
| 水墨画 | shuimo |
| 莫奈 | monai |
| 扁平插画 | bianping |
| 像素插画 | xiangsu |
| 儿童绘本 | ertonghuiben |
| 3D 渲染 | 3dxuanran |
| 漫画 | manhua |
| 黑白漫画 | heibaimanhua |
| 写实 | xieshi |
| 动漫 | dongman |
| 毕加索 | bijiasuo |
| 赛博朋克 | saibopengke |
| 油画 | youhua |
| 马赛克 | masaike |
| 青花瓷 | qinghuaci |
| 新年剪纸画 | xinnianjianzhi |
| 新年花艺 | xinnianhuayi |
- Prompt must not be empty; Chinese is recommended; up to 1,024 characters. NegativePrompt also takes up to 1,024.
- Only the eight resolutions in the table are accepted; the default is 1024:1024. Clarity can upscale x2 or x4 — 1024:1024 at x2 becomes 2048:2048.
- Revise (prompt rewriting) is on unless you send 0; it rewrites your prompt before generating.
- LogoAdd defaults to 1, which puts a visible AI-generated mark on the image; send 0 to leave it off.
- A preset style is single-choice; with none, write “… style” at the start of the prompt.
Sources: Tencent Cloud SDK · SubmitHunyuanImageJob · 混元生图风格列表
Three steps
Settle the style first
Pick a preset (sent as the Style parameter), or type style words such as “ink-wash style” or “Dunhuang mural style” — Tencent says to put style words at the head of the prompt.
Describe the subject and scene
Say who or what, doing what; the scene sets the place and the light. Framing, mood and detail are optional.
Copy each part, or the request
The prompt goes in Prompt, the negative in NegativePrompt, and the resolution must be one of the eight below.
Why style goes first
Tencent’s Hunyuan style list gives two ways to set a style: choose a preset, sent as the Style parameter, or leave it empty and write style words at the start of the prompt — its example is “black-and-white sketch, a cute puppy”. This builder does both: choose a preset and the style-words box steps aside; leave it empty and your style words lead the prompt.
The negative prompt is its own NegativePrompt field in Hunyuan’s API. List what to keep out — “extra fingers, deformed, text” — and a leading “no” or “avoid” is removed for you, so the field does not end up as a double negative.
Questions
- No. It only assembles text in your browser, with no network and no model. You paste the prompt and parameters into Hunyuan or your API request yourself.
- Hunyuan’s API has its own NegativePrompt field. Putting “extra fingers, watermark” in the Prompt asks the model for them.
- LogoAdd defaults to 1, which adds a visible AI-generated mark. Tencent recommends keeping it so viewers know the image is AI-made; send 0 to turn it off.
- By default, yes: Revise is on unless you send 0, and the result returns the rewritten prompt. With a reference image, rewriting stays on regardless.
- No. The Resolution parameter decides it and takes only eight values — 1280:720 for 16:9.
- The API recommends Chinese. The open-source HunyuanImage 2.1 says it supports Chinese and English prompts natively.
Method & sources
How it computes
Pure string assembly, style words first as Tencent's style list advises, or a preset style sent in the Style parameter instead. Prompt and NegativePrompt are separate and each checked against Tencent's 1,024-character limit. Resolution is one of the eight values SubmitHunyuanImageJob accepts; Clarity (x2/x4), Revise and LogoAdd are shown with their documented defaults. No model is called.
What this tool implements
- Corrected 2026-10-26 (rebuild): the negative terms were pasted into the prompt, asking Hunyuan FOR 'extra fingers, watermark'; NegativePrompt is now its own field. Resolution was free text; it is now one of the eight documented values. The page never said that Revise (prompt rewriting) and LogoAdd (a visible 'AI-generated' mark) are on by default; both are now visible switches. Eighteen preset styles use Tencent's own style IDs.
- Parameters read on 2026-10-26 in Tencent Cloud's SDK request model and its style-list page (last updated 2026-04-27). Assembly is in image-prompt-kernel.js, tested.
Sources
- Tencent Cloud SDK (Python), hunyuan v20230901 models.py, SubmitHunyuanImageJobRequest: Prompt 「最多可传1024个 utf-8 字符」; Resolution 「768:768（1:1）、768:1024（3:4）、1024:768（4:3）、1024:1024（1:1）、720:1280（9:16）、1280:720（16:9）、768:1280（3:5）、1280:768（5:3），不传默认使用1024:1024」; Clarity 「x2：2倍超分 x4：4倍超分」; Revise 「不传默认开启」; LogoAdd 「默认为1」. https://github.com/TencentCloud/tencentcloud-sdk-python/blob/mast…
- Tencent Cloud. 混元生图风格列表: 「请将下表中的 风格编号 传入接口入参中的 Style」, or leave Style empty and write 「xx 风格 + 其他提示词」 in the prompt. https://cloud.tencent.com/document/product/1729/105846
Pick up where you left off
Stored only in this browser — never sent to our servers.