[Update] I tried out the new feature in Amazon QuickSight that allows automatic generation of BI screens from images and prompts

This is Ishikawa from the Cloud Business Division. Generate Sheet, which generates just one sheet into an existing analysis, and a feature that creates an analysis from a dashboard image have been added to Amazon Quick, so I tried both. It's a practical and incredibly convenient update.

https://aws.amazon.com/jp/about-aws/whats-new/2026/09/generate-sheet-and-generate-analysis-from-an-image/

Two Features: "Generate Sheet" and "Generate Analysis from an Image"

Amazon Quick already has Generate Analysis, which generates multi-sheet analyses all at once from natural language prompts. This update provides two methods for this Generate Analysis.

The first, Generate Sheet, is a feature that generates just one sheet within an open analysis. You can add a sheet to an existing analysis without rebuilding the entire analysis.

The second is a feature that generates an analysis by attaching a dashboard image to the prompt. Dashboard images from BI tools other than Amazon Quick are also supported, and it rebuilds them as editable analyses within the scope that Amazon Quick supports.

Both are for Author Pro users with an Enterprise subscription at the time of announcement. Authors will also be provided with promotional access as part of Amazon Quick Enterprise until December 2026, unless the organization has restricted access.

https://docs.aws.amazon.com/quick/latest/userguide/generating-an-analysis.html

The two features differ in where they are launched and the maximum prompt length that can be entered.

According to the official documentation, Generate Sheet is described as adding a sheet with "visuals selected to match the data, filter controls, and calculated fields such as year-over-year and month-over-month comparisons." The verification for this time will be confirming how far this actually goes.

For generation from an image, it is described as Amazon Quick reading the image and building sheets and visuals supported by Amazon Quick.

Trying It Out

Prerequisites

• Amazon Quick Enterprise Edition and at least one Author Pro user
• At least one dataset
• Verification environment: ap-northeast-1 (Tokyo) and us-east-1 (Northern Virginia)
• Dataset used for verification: Superstore order data

At the time of writing, these two features only have a console UI, and all verification was done through console operations.

The Generate Button Does Not Appear in the Tokyo Region

First, in the Tokyo region, I opened an existing analysis and pressed the + on the sheet tab. According to the documentation steps, the new sheet dialog that opens here should have Interactive sheet and Generate.

However, only two options were displayed: [Cancel] and [Add], with no Generate. So I switched regions to check. I switched the region to Northern Virginia from the user dropdown and repeated the same operation with the Japanese UI.

The left is the Tokyo region, and the right is Northern Virginia. Both are in the Japanese UI, with the same new sheet dialog open. The Tokyo region shows only [Cancel] and [Add], while the Virginia region shows [Cancel] and [Add] plus [Generate].

The What's New states "generally available in all AWS Regions where Amazon Quick is available," but at the time of writing, the entry point for Generate Sheet could not be confirmed in the Tokyo region. The subsequent verification will proceed in Northern Virginia.

Note that the [Add image] button was also displayed on the Generate analysis screen in the Tokyo region. It appears that even within the same update, the situation differs by feature.

Trying Generate Sheet with a Japanese Prompt

Selecting Generate opens the Create a sheet screen.

The heading reads "Describe the sheet you want to visualize," and the input field contains the placeholder "Example: Show monthly sales trends by region and highlight the best-selling products."

The "280 characters remaining" below the input field is the prompt limit for this dialog. Compared to the 1,400 characters for Generate analysis described later, this is quite short, making it necessary to narrow down and write your requirements.

The datasets listed in the Dataset section below are only those included in the open analysis. Here, there is only one: superstore_orders. Other datasets in the account cannot be selected. The Generate button in the lower right cannot be pressed until both the prompt and dataset fields are filled in.

This time, I entered the following prompt.

A sheet analyzing sales and profit by category and sub-category. Include KPIs for sales and profit, monthly sales trend, sales mix by category, and sub-category profit margin ranking. Also add a year-over-year calculated field and filter controls for region and segment.

After entering the prompt and pressing [Generate], a [Generating sheet] tab appears next to the sheet tabs, and you can still operate other sheets in the analysis while generation is in progress. Generation took a few minutes.

The completed sheet was named "カテゴリ・サブカテゴリ売上利益分析" (Category/Sub-Category Sales Profit Analysis), and the sheet name was also given in Japanese from the Japanese prompt. There are 5 visuals.

• Total Sales and Profit Overview (KPI)
• Monthly Total Sales and Profit Trends (line)
• Total Sales by Product Category (donut)
• Average Profit by Sub-Category (horizontal bar)
• Total Sales and Profit by Sub-Category (table)

The sheet name was in Japanese, but the titles and descriptions of each visual were in English. The description text directly below the sheet was also a mix of Japanese and English: "カテゴリ別・サブカテゴリ別の売上と利益を分析するシート。KPI、月次推移、カテゴリ構成、サブカテゴリ利益率ランキングを含む。 providing comprehensive data insights and analysis for informed decision making."

Pay attention to the 529.43% KPI at the top of the image. The visual's description reads "Revenue and profitability with year-over-year comparison," claiming a year-over-year comparison, but as you can see from the sales $23.14M on the left and profit $3.68M on the right side by side, the actual value is a comparison between sales and profit. Opening the left pane revealed that profit was set as the target value.

The lower-left Average Profit by Sub-Category also has a Y-axis title of Profit Margin, but the plotted values like $14.65K are average profit as a monetary amount. The center Monthly Total Sales and Profit Trends has "monthly" in the title, but dates are lined up on the X-axis.

Here, I checked the two items specified in the prompt: filter controls and calculated fields. I opened the filter panel to check, but there were 0 sheet-level filters. I also checked the field list in the data panel, but the columns were exactly the same as the original dataset columns, with no calculated fields added.

Trying Generate Sheet with an English Prompt

To isolate whether the Japanese prompt was the cause, I generated again with an English prompt against the same analysis. The filter controls and calculated fields are this time explicitly specified.

Sales and profit by category and sub-category. Include KPIs for sales and profit, monthly sales trend, category mix, and sub-category profit margin ranking. Add a year-over-year growth calculated field and filter controls for region and segment.

The generated sheet was "Sales & Profit by Category and Sub Category," and the number of visuals increased to 6.

• Total Sales with Yearly Trend (KPI)
• Total Profit with Yearly Trend (KPI)
• Total Sales Trend Over Time (line)
• Total Sales by Category (donut)
• Total Profit by Sub-Category (horizontal bar)
• Total Sales by Sub-Category (vertical bar)

The KPI construction was more reasonable than the Japanese version. The top two show 3.58% and 13.28%, displaying a comparison between time periods, with dates of 2015/5/25 and 2016/8/7, and 2015/6/5 and 2015/6/7 respectively. Opening the left pane revealed that order_date was set in the trend group. It is not configured to put profit into the target value as in the Japanese version.

However, filter controls and calculated fields were also 0 with the English prompt. Despite the description text directly below the title saying "Filterable by region and segment," as shown in the image, no filter controls were placed on the sheet. The point where the X-axis title in the lower-left Total Profit by Sub-Category is Profit Margin while the content is monetary amounts was the same as the Japanese version.

Changing the language of the prompt did not change the fact that filter controls and calculated fields were not created.

Generating an Analysis from a Dashboard Image

Next, I tried generating an analysis from an image. For the input image, I built an HTML mock of a dashboard from a BI tool other than Amazon Quick and prepared a 1600 x 1000 pixel legacy_bi_dashboard.png (shown below).

The elements included in the image are as follows.

• 4 KPIs (Total Sales, Total Profit, Orders, Profit Ratio, all with year-over-year)
• Monthly Sales Trend (2 lines for current year and previous year)
• Sales by Segment (donut)
• Sales by Sub-Category (horizontal bar, top 10)
• Sales by Region (vertical bar)
• Top Products by Sales (table)
• 5 filter panels on the right (Order Date, Region, Segment, Category, Ship Mode)
• 3 tabs at the top right (Overview, Product Detail, Customers)

A title, "Superstore Sales Overview," is placed in the upper left. The highlight will be how this title is handled later.

Selecting Generate analysis from the analysis list opens the Visualize your data with AI screen.

[+ Add data] and [Add image] are lined up below the input field. Selecting the image from [Add image] attaches a chip labeled legacy_bi_dashboard.png to the right of it.

At the bottom of the screen, it reads: "Generate Analysis uses AI to create multi-sheet analyses with calculated fields, filter controls, and multi-dataset support. Generated analyses should be reviewed before publishing." Creating calculated fields and filter controls, as well as reviewing before publishing, are explicitly stated on this screen.

I selected the Superstore order data as the dataset, and kept the prompt minimal since I wanted to see the result of reading the image.

Recreate the attached dashboard image as closely as possible using this dataset. Keep the same KPIs, charts and filter controls as the image.

During generation, progress is displayed in stages. The 9 steps I was able to observe are as follows. There are dedicated steps for Generating calculated fields and Generating filters. In Generate Sheet, there was no display of processing equivalent to these.

"Superstore Sales Overview" is exactly the title from the upper left of the original image. The title is being read from the image. Note that I selected Preview analysis outline instead of Generate analysis, but within the polling range I observed, the screen for editing the plan was not displayed, and generation proceeded all the way to the end.

The generated analysis was named "Superstore Sales Overview" and had 9 visuals. The correspondence with the original image is as follows.

In the field list in the left pane, a field called profit_ratio has been added. This was not in the original 20 columns.

Let me zoom in on the top of the sheet.

In the Controls row, 5 controls are lined up: order_date equals All, region equals All, segment equals All, category equals All, and ship_mode equals All. These roughly match all 5 of the Order Date, Region, Segment, Category, and Ship Mode filter panels placed in the original image.

The title is "Superstore Sales Overview," taken directly from the original image's title. The description text below reads: "Executive sales analytics dashboard with 4 KPIs (Total Sales, Total Profit, Orders, Profit Ratio) with YoY comparisons, Monthly Sales Trend, Sales by Segment donut, Sales b..." and the names of the 4 KPIs from the original image are listed directly.

I opened [Edit] from the menu for the calculated field profit_ratio mentioned earlier and confirmed its definition.

/*
Profit margin as a percentage of total sales.
*/
ifelse(sum(sales) = 0, NULL, sum(profit) / sum(sales) * 100)

This means it took the PROFIT RATIO KPI from the original image and created a calculated field for a metric that does not exist in the dataset. The implementation is pleasing, with a branch to avoid division by zero and a comment explaining the content.

On the other hand, there were also aspects that were not reproduced. The 3 tabs in the original image were consolidated into 1 sheet, and no tab structure was created. The Monthly Sales Trend was drawn with 2 lines for the current year and previous year, but what was generated was a single line.

Caution was also needed with the numbers themselves. While the donut shows $23.14M as the total sales for the entire period, the Total Sales Revenue KPI in the upper left displays $560.44K. The KPI is configured to compare periods segmented by a trend group, which differs in meaning from the total for the entire period that the KPI in the original image was showing.

Analysis

Testing the two features against the same dataset revealed differences in the generated analyses (visuals).

The official documentation states for Generate Sheet that it adds a sheet containing filter controls and calculated fields such as year-over-year comparisons. In this case, there were 0 filter controls and calculated fields with both Japanese and English prompts, and even when explicitly specified. Since one generation is not enough to draw a conclusion, it seems advisable to check these two points when using it.

In contrast, for generation from an image, dedicated steps for Generating calculated fields and Generating filters ran, and 5 controls and 1 calculated field were actually created. It appears that even within the same generative BI, the processing flow itself differs depending on whether the input is an image or text.

Let me also summarize what could and could not be reproduced in the image-based generation.

What was reproduced

• Dashboard title
• Metric selection for 4 KPIs
• Visual types (donut, horizontal bar, vertical bar, line, table)
• 5 filter target columns
• Creating a calculated field for a metric that does not exist in the dataset

What was not reproduced

• Tab structure (3 tabs consolidated into 1 sheet)
• 2-line year-over-year comparison (generated as 1 line)
• KPI aggregation scope (period comparison rather than total for entire period)

While "what you want to see" as depicted in the image was read, the part corresponding to "how to segment and display it" was missed. For the use case of creating a starting point when migrating from another BI tool, the work of manually arranging visuals can be skipped. However, since the aggregation definitions cannot be used as-is, a process of reviewing the settings of each visual after generation is still necessary.

A note on prompt language as well. Sheets were generated with Japanese prompts, and sheet names were also given in Japanese. However, within the scope of this verification, the English prompt produced more reasonable KPI construction. The Japanese version wrote "year-over-year comparison" in the description while putting profit into the target value, meaning the intent of the visual and its settings are misaligned.

Regarding regions, at the time of writing, the Generate Sheet button was not present in the Tokyo region. This differs from what the What's New states, but it is likely due to the pace of deployment. If you try in the Tokyo region and cannot find Generate, switching to Northern Virginia can help isolate the issue.

Points I hope to see improved in the future are as follows.

• Generation of filter controls and calculated fields in Generate Sheet
• Multi-sheet generation from tab structures within images
• Reading of series configurations like year-over-year comparisons within images
• Rollout of Generate Sheet to the Tokyo region
• Japanese language support for visual titles and descriptions

In Closing

I tried Amazon Quick's Generate Sheet and Generate Analysis from an Image.

Generate Sheet is practical in that you can add a sheet to an existing analysis without rebuilding it. The prompt limit is 280 characters and you can only select from datasets included in the analysis. Since filter controls and calculated fields were not generated this time, it is realistic to treat it as a tool for creating a starting draft of visuals.

The image-based analysis generation produced more in-depth results than expected. It reproduced the dashboard image's title, KPIs, visual types, and filter target columns, and created calculated fields for metrics that do not exist in the dataset. For the scenario of rebuilding a dashboard created in another BI tool in Amazon Quick, it can skip the first step.

Neither result can be published as-is, and a process of checking aggregation definitions and visual settings is necessary. Amazon Quick itself encourages a review before publishing on the generation screen.

If you have an Enterprise subscription and at least one Author Pro user in your environment, you can try it without any additional configuration. Authors are also eligible for promotional access until December 2026. If you cannot find Generate in the Tokyo region, try switching regions to check.