One prompt, many images: why I’m building Berry AI Studio
The reason I started Berry AI Studio was straightforward: AI generates a lot of material. One run can output several images, and the same prompt can accumulate many different variations.
The image managers I was using did not handle this workflow well. The files were there, but reviewing different results from the same prompt was not as convenient as I wanted.
That is the starting point of the tool previously called Berry AIGC Toolbox, now Berry AI Studio.
What I want to manage is a set of related attempts
In a folder, each image is a separate file. To the person creating them, those files can be attempts connected by the same idea and the same prompt.
I need to move between those results: which version is closer to what I wanted, which ones should I keep, and where do I want to continue? As the collection grows, simply being able to store and open the files does less to support that work.
I do not want the tool to decide which image is best. I want it to make my own selection easier.
The same prompt does not mean the same result
Images made from the same prompt can still turn out differently. In this workflow, those differences are variations worth reviewing, not duplicate files to remove.
That is why the library needs to consider relationships between images as well as individual files. A set of attempts should not become a wall of unrelated thumbnails when I return to it.
Why reading the prompt is only a beginning
Extracting prompts and metadata is not, by itself, the reason I wanted to build the software.
Those details provide connections and context: search for related results by prompt, inspect an image’s settings, and use that information while deciding which version to keep. The workflow I want to improve is finding the set, looking through its variations, and recording a choice.
A complete metadata display is useful, but it is not enough if revisiting the related images still feels cumbersome.
What the current app connects
- Prompt-content filtering: narrow the collection without relying only on filenames.
- Grid or table browsing: look through a set of results before choosing one to inspect closely.
- Quick Look and arrow-key navigation: review variations consecutively without opening every file again.
- The inspector: keep the selected image’s prompt, generation settings, and workflow available.
- Favorites, ratings, and batch organization: leave a selection you can return to.
There is an important detail here: the current prompt filter uses partial content matching. It can retrieve related results, but that is not the same as automatically grouping images by an identical full prompt or building a version history.
These are foundations for the workflow, not a claim that managing same-prompt variations is a finished problem.
The implementation should support reviewing, not just storing
The current desktop app combines Tauri, Rust, Vue, and SQLite. Indexed prompt information gives search something to work with beyond filenames. Virtualized browsing and Quick Look support two different scales of attention: looking through a collection and inspecting one result.
Local indexing does not mean every feature is offline; model information can come from Civitai. The core collection workflow, though, stays centered on your images and their generation context.
The next visit is the test I care about
Can I return to the results from a prompt, find the versions I was considering, see which one I kept, and carry on with less back-and-forth?
That is the direction I want to keep improving. Generating more images should not mean accepting more tedious work afterward.
Read the Berry AI Studio product story, or explore the project overview, prompt-filter definition, and Quick Look implementation. Technical details checked on September 6, 2026.