docs: add prompt garden image workflow tutorial
|
After Width: | Height: | Size: 2.4 MiB |
|
After Width: | Height: | Size: 3.1 MiB |
|
After Width: | Height: | Size: 1.6 MiB |
|
After Width: | Height: | Size: 229 KiB |
|
After Width: | Height: | Size: 193 KiB |
|
After Width: | Height: | Size: 125 KiB |
|
After Width: | Height: | Size: 235 KiB |
|
After Width: | Height: | Size: 245 KiB |
92
mkdocs/docs/en/examples/image-prompt-garden-nb151.md
Normal file
@@ -0,0 +1,92 @@
|
||||
# Import an Image Prompt from the Prompt Library and Compare Models
|
||||
|
||||
This example walks through a complete image prompt workflow: choose a reusable template from the prompt library, import it into Optimizer, edit a few core variables, use smart variable fill for the remaining fields, and compare outputs from two image models.
|
||||
|
||||
The example uses the [vintage commemorative postage stamp poster](https://garden.always200.com/prompts/vintage-commemorative-postage-stamp-poster) prompt from the prompt library. Its import code is `ZH-NB-151`.
|
||||
|
||||
## 1. Choose a Template in the Prompt Library
|
||||
|
||||
Open the prompt detail page and review the prompt description, sample images, and import code. This template is designed to turn solar terms, cities, character birthdays, product launches, or commemorative events into vertical collectible postage stamp posters.
|
||||
|
||||
There are two entry points on the prompt page:
|
||||
|
||||
- Click "Open in Optimizer" to jump directly to the Optimizer web interface.
|
||||
- Copy the import code `ZH-NB-151` and import it manually in Optimizer later.
|
||||
|
||||

|
||||
|
||||
## 2. Import It into Optimizer
|
||||
|
||||
Open Optimizer's text-to-image workspace. If you did not jump directly from the prompt library page, click the flower icon in the right toolbar and choose the Prompt Garden import action.
|
||||
|
||||

|
||||
|
||||
Paste the import code `ZH-NB-151` into the dialog and click Import.
|
||||
|
||||

|
||||
|
||||
After import, Optimizer keeps the template body and detects 7 variables: `title`, `commemoration`, `date`, `subject`, `palette`, `denomination`, and `tagline`. The sample values from the template are also imported, so you can edit them directly.
|
||||
|
||||

|
||||
|
||||
## 3. Edit the Core Variables
|
||||
|
||||
You can edit variables one by one. In this example, the stamp poster structure stays the same, while the core theme changes to Beginning of Summer and Yoimiya from Genshin Impact:
|
||||
|
||||
| Variable | Example value |
|
||||
| --- | --- |
|
||||
| `title` | 立夏 |
|
||||
| `commemoration` | 二十四节气·立夏纪念 |
|
||||
| `date` | 2026 / 05 / 05 |
|
||||
| `subject` | 原神的宵宫 |
|
||||
|
||||
The remaining variables can be filled manually, or left empty and completed with smart variable fill.
|
||||
|
||||

|
||||
|
||||
## 4. Complete Details with Smart Fill
|
||||
|
||||
Click Smart Variable Fill. Optimizer generates values for the remaining fields from the existing variables. The confirmation dialog shows the generated values, reasons, and confidence scores, so you can apply only the suggestions you want.
|
||||
|
||||
In this run, smart fill completed the palette, denomination, and tagline so they matched the Beginning of Summer and Yoimiya theme.
|
||||
|
||||

|
||||
|
||||
After applying the suggestions, you can still adjust the values. The final generation used:
|
||||
|
||||
| Variable | Example value |
|
||||
| --- | --- |
|
||||
| `palette` | 暖金、翠竹绿、浅米纸色 |
|
||||
| `denomination` | 120分 纪念邮资 编号:2026-4 |
|
||||
| `tagline` | 夏木已成阴,烟花映长空 |
|
||||
|
||||
## 5. Select Models and Generate a Comparison
|
||||
|
||||
After confirming the variables, choose the image models in the test area. In this example:
|
||||
|
||||
- Column A: `gpt-image-2`
|
||||
- Column B: `Doubao Seedream 5.0 Lite`
|
||||
|
||||
Generate both columns to compare how the models interpret the same prompt. Column A is closer to a complete vintage postage stamp poster, with more stable perforations, postmark, denomination, date block, and title hierarchy. Column B looks more like a bright character illustration card: lighter and cleaner, but less like a collectible poster.
|
||||
|
||||

|
||||
|
||||
## 6. Review the Final Result
|
||||
|
||||
If one result is closer to the target, download it or continue refining the variables around that direction. Below is the final `gpt-image-2` result from this run.
|
||||
|
||||

|
||||
|
||||
## What This Example Validates
|
||||
|
||||
- Whether a prompt library template imports smoothly into the text-to-image workspace.
|
||||
- Whether the variable structure is preserved and can be edited directly or completed with smart fill.
|
||||
- How different image models handle layout, text, subject identity, and detail from the same prompt.
|
||||
- Whether the image prompt keeps the output aligned with a vintage commemorative postage stamp poster.
|
||||
|
||||
## Related Pages
|
||||
|
||||
- [Text-to-Image Workspace](../image/text2image-workspace.md)
|
||||
- [Smart Variable Fill](../auxiliary/smart-fill.md)
|
||||
- [Prompt Library](../basic/prompt-garden.md)
|
||||
- [Model Testing Strategy](../user/model-testing-strategy.md)
|
||||
92
mkdocs/docs/zh/examples/image-prompt-garden-nb151.md
Normal file
@@ -0,0 +1,92 @@
|
||||
# 从提示词库导入图像提示词并对比模型
|
||||
|
||||
这个示例演示一条完整的图像提示词工作流:从提示词库选择一个可复用模板,导入到 Optimizer,修改少量核心变量,用智能填充补齐剩余变量,最后用两个图像模型对比生成效果。
|
||||
|
||||
示例使用提示词库中的[复古纪念邮票海报](https://garden.always200.com/prompts/vintage-commemorative-postage-stamp-poster),导入码为 `ZH-NB-151`。
|
||||
|
||||
## 1. 在提示词库选择模板
|
||||
|
||||
打开提示词详情页后,可以先查看提示词说明、示例图和导入码。这个模板适合把节气、城市、角色生日、产品发布或纪念事件做成竖版收藏级邮票海报。
|
||||
|
||||
在页面右侧有两种入口:
|
||||
|
||||
- 点击“在优化工具中打开”,直接跳转到 Optimizer Web 界面。
|
||||
- 复制导入码 `ZH-NB-151`,稍后在 Optimizer 中手动导入。
|
||||
|
||||

|
||||
|
||||
## 2. 在 Optimizer 中导入
|
||||
|
||||
进入 Optimizer 的文生图工作区。如果没有从提示词库页面直接跳转,也可以点击右侧工具栏的花朵标志,选择“导入花园提示词”。
|
||||
|
||||

|
||||
|
||||
在弹窗中粘贴导入码 `ZH-NB-151`,点击“导入”。
|
||||
|
||||

|
||||
|
||||
导入完成后,Optimizer 会保留模板正文,并识别出 `title`、`commemoration`、`date`、`subject`、`palette`、`denomination`、`tagline` 这 7 个变量。模板自带的示例值也会一起带入,方便直接改写。
|
||||
|
||||

|
||||
|
||||
## 3. 修改核心变量
|
||||
|
||||
变量可以直接逐项修改。这个例子保留邮票海报的结构,只把核心主题改成“立夏 + 原神的宵宫”:
|
||||
|
||||
| 变量 | 示例值 |
|
||||
| --- | --- |
|
||||
| `title` | 立夏 |
|
||||
| `commemoration` | 二十四节气·立夏纪念 |
|
||||
| `date` | 2026 / 05 / 05 |
|
||||
| `subject` | 原神的宵宫 |
|
||||
|
||||
其余变量可以手动填写,也可以先留空,交给智能填充补齐。
|
||||
|
||||

|
||||
|
||||
## 4. 用智能填充补齐细节
|
||||
|
||||
点击“智能填充变量值”后,Optimizer 会根据已有变量生成剩余字段。确认弹窗会展示生成值、生成理由和置信度,可以选择性应用。
|
||||
|
||||
本例中,智能填充补全了配色、面值编号和纪念短句,让它们与“立夏”和“宵宫”的视觉主题保持一致。
|
||||
|
||||

|
||||
|
||||
应用后可以继续微调变量。例如最终生成使用了:
|
||||
|
||||
| 变量 | 示例值 |
|
||||
| --- | --- |
|
||||
| `palette` | 暖金、翠竹绿、浅米纸色 |
|
||||
| `denomination` | 120分 纪念邮资 编号:2026-4 |
|
||||
| `tagline` | 夏木已成阴,烟花映长空 |
|
||||
|
||||
## 5. 选择模型并生成对比
|
||||
|
||||
变量确认后,在测试区选择要对比的图像模型。本例中:
|
||||
|
||||
- A 列:`gpt-image-2`
|
||||
- B 列:`Doubao Seedream 5.0 Lite`
|
||||
|
||||
点击生成后,可以直接并排比较两个模型对同一提示词的理解。A 列更接近完整的复古邮票海报,齿孔、邮戳、面值、日期块和标题层级更稳定;B 列更像明亮的角色插画邮票卡片,画面轻盈但整体收藏海报感弱一些。
|
||||
|
||||

|
||||
|
||||
## 6. 查看最终效果
|
||||
|
||||
如果某一列结果更接近目标,可以下载或继续围绕它微调变量。下面是本次 `gpt-image-2` 生成的最终效果图。
|
||||
|
||||

|
||||
|
||||
## 这个示例适合验证什么
|
||||
|
||||
- 提示词库模板能否顺畅导入到文生图工作区。
|
||||
- 变量结构是否能保留,并支持直接改写或智能填充。
|
||||
- 同一提示词在不同图像模型上的版式、文字、主体识别和细节差异。
|
||||
- 图像提示词是否能把输出稳定约束在“复古纪念邮票海报”这个目标上。
|
||||
|
||||
## 相关页面
|
||||
|
||||
- [文生图工作区](../image/text2image-workspace.md)
|
||||
- [变量智能填充](../auxiliary/smart-fill.md)
|
||||
- [提示词库](../basic/prompt-garden.md)
|
||||
- [模型选择与测试策略](../user/model-testing-strategy.md)
|
||||
@@ -113,6 +113,8 @@ plugins:
|
||||
- 创意写作: examples/creative-writing.md
|
||||
- 商务沟通: examples/business-communication.md
|
||||
- 教学培训: examples/educational-training.md
|
||||
- 图像提示词工作区:
|
||||
- 从提示词库导入并对比模型: examples/image-prompt-garden-nb151.md
|
||||
- 开发指南:
|
||||
- 媒体示例: guide/media.md
|
||||
- locale: en
|
||||
@@ -176,6 +178,8 @@ plugins:
|
||||
- Creative Writing: examples/creative-writing.md
|
||||
- Business Communication: examples/business-communication.md
|
||||
- Educational Training: examples/educational-training.md
|
||||
- Image Prompt Workspace:
|
||||
- Import from Prompt Library and Compare Models: examples/image-prompt-garden-nb151.md
|
||||
- Developer Guide:
|
||||
- Media Examples: guide/media.md
|
||||
# 多版本:由 mike 管理(无需在此启用插件)
|
||||
|
||||