Asked and answered.

Describe the character in detail once, then reuse that exact description in every prompt and change only the scene and camera. Small wording changes to the face, hair or outfit give you a different person, so copy the description word for word. Character Prompt Builder gives you this part as a separate identity block for exactly that reason. If your image tool accepts a reference image, adding one of your earlier results helps too.

You can build 7 types: a person, a real animal, an anthro (an animal that stands upright and wears clothes), an animated character, a robot, an original creature or a fantasy race such as an elf or orc. Each type gets its own questions, and answers that don't apply never end up in the prompt.

Yes. There are 99, covering all 7 types and all 19 art styles, from photorealistic portraits and wildlife shots to anime, claymation, pixel art, watercolour and character sheets. Filter them by type and style, pick one and every question is answered for you. Change any answer before you copy the prompt.

Any of them. The default is plain sentences, which Flux, Nano Banana, GPT Image, Midjourney v7, Firefly, Veo and Sora all read well. Two more styles are there if you need them: Midjourney, with the --ar and --v 7 flags, and Stable Diffusion tags for SDXL and SD 1.5.

No. The prompt is assembled in your browser from your answers using a local template, so it is instant and costs nothing. You paste the result into whichever image or video tool you already use.

Yes. Save characters to My characters, kept in your browser, and load them back for new scenes. You can also download a character as a file to back it up or share it.

Yes. Once you have opened it, your browser keeps a copy of its files, so it opens again and keeps working with no connection. Your characters stay in that browser too.

Start from what the chart should say: compare items, show change over time, show parts of a whole, show a spread, show a relationship, or compare with a target. Then look at your data: how many categories, how many points in time, how many measures. Bars compare, lines show time, a donut shows two to five parts, a histogram shows a spread, a scatter shows a relationship, and a bullet chart compares with a target. Chart Compass does this for you and draws the result.

When there are more than about five slices, when the slices are close in size, or when you want to compare two pies. People compare angles poorly, so small differences disappear. A sorted bar chart shows the same shares exactly.

Each chart Chart Compass recommends names its Power BI visual (for example Line and clustered column chart, or Scatter chart with a measure in Size) and the fields that go into its wells. Charts Power BI does not have built in, like the box plot or the Sankey diagram, point to an AppSource visual.

Yes. The charts are the same in every tool; Chart Compass gives the Tableau Show Me type and the Excel Insert menu for each one alongside Power BI.

A table with a header row, copied from Excel or Google Sheets, or a CSV or TSV file, up to 5,000 rows. It understands dates like 2026-03, Mar 2026 and Q1 2026, and numbers like 1,234.5, 1.234,5, $1,200, 12% and 3.4K.

No. It is read and drawn in your browser. It is kept in your browser's storage so it is still there next time; pasting new data replaces it, and Paste your own clears it.

It shows every file Claude Code reads, what each setting in them does, and where each file lives. You change what you want, and it shows you the exact file that would be written, before anything exists.

The more specific one. A managed policy set by an administrator beats everything, then command-line flags, then .claude/settings.local.json (you, in this project), then .claude/settings.json (the team, in this project), then ~/.claude/settings.json (you, everywhere). The studio shows this order for every surface.

In settings.json files at three levels. ~/.claude/settings.json applies to you everywhere, .claude/settings.json in a project is shared with your team, and .claude/settings.local.json in a project is only for you and is not committed.

No. It runs in your browser and cannot reach your disk. It gives you the file or an install script to copy, and you decide whether to use it.

Yes. Before it writes anything, the script copies every file it is about to change into a dated folder under ~/.benibix-backups and notes it in an undo.sh there; it prints that folder when it finishes. Running undo.sh copies your old files back, removes the ones that were not there before, and takes out the folders the script made once they are empty.

Claude Code 2.1.296. Every fact carries a badge saying where it came from: read from the installed program, seen while running it, taken from the documentation with a date, or reasoned and not verified.

In ~/.codex/config.toml for you, and .codex/config.toml in a project, which Codex reads only when you trust the project. Set CODEX_HOME to move the whole ~/.codex folder.

Yes. Read in a setup takes your .codex folder, a repository, or files such as config.toml, hooks.json, AGENTS.md and .rules files, read in your browser, and opens them in the editors. Anything it cannot edit is listed and stays in your files: the install script merges into config.toml rather than replacing it.

Requirements set by an administrator limit everything. Below them a -c override or flag wins, then the project's .codex/config.toml (the one nearest your folder first), then a profile chosen with --profile, then your ~/.codex/config.toml, then Codex's defaults.

Codex retired it. Use on-request with a read-only sandbox, or mark the project untrusted with trust_level = "untrusted" under [projects], which also skips the project's own .codex/ files. The studio offers on-request, never and granular.

Codex draws it from its own items, set in [tui] status_line in config.toml or picked with /statusline. There is no script to install. The studio lets you pick and order the 31 items and the terminal title, and shows the result.

No. It runs in your browser and cannot reach your disk. It gives you the files or an install script to copy, and the script keeps a copy of every file it changes, with an undo.sh that puts them back.

Codex 0.162.1, and the Codex documentation as read on 11 October 2026. Every fact carries a badge saying where it came from: the documentation with its date, the source code, or the installed program's help.

Yes. Lucide is under the ISC licence, Material Symbols under Apache 2.0, and Tabler, Phosphor, Fluent, Heroicons, Bootstrap Icons and Iconoir under the MIT licence. All three allow commercial use, changes and redistribution without paying or asking. The one condition is to keep the copyright and licence notice with what you ship; Icon Studio links each set's licence file, and you can include it with your project.

A visible credit on your website or in your app is not required. The MIT, ISC and Apache 2.0 licences ask that the copyright and licence notice stay with copies of the icons, which usually means a line in your project's licence or third-party notices file.

Copy the icon as a Power BI measure, create a new measure with it, and set its data category to Image URL (Modeling, Data category). Add the measure to a table or matrix and set the image size under Format visual. You can wrap the SVG in a SWITCH to show a different icon or colour per row.

Lucide has 2,100 clean outline icons, popular in web apps. Tabler has over 5,800 in outline and filled. Phosphor has over 1,400 icons in six weights, from thin to bold and duotone. Fluent is Microsoft's own set, the look of Windows and Office. Material Symbols is Google's, with almost 4,000 icons in outlined, rounded and sharp. Heroicons comes from the makers of Tailwind CSS, Bootstrap Icons from Bootstrap, and Iconoir is a large set of simple, even outlines.

The motion is part of the SVG, as CSS, so it plays wherever SVG plays: web pages, Figma prototypes and many design tools. A PNG is a single frame, so it shows the icon still.

No. The icons come from this site, and styling and exporting happen in your browser. Your favourites and presets are kept in your browser only.

A key performance indicator is a measurable value that shows how well a team or business is reaching a goal. Good KPIs have a clear formula, a direction (higher or lower is better), a target, and a set rhythm for checking them, such as weekly or monthly.

Usually five to nine on one page. Start from the starter dashboard of your area in KPI Atlas, keep the ones you will act on, and move the rest to detail pages.

It depends on the question. A trend over time is a line chart, actual against target is a bullet chart or a gauge, a ranking is a sorted bar chart, and a change explained step by step is a waterfall. KPI Atlas ranks the best charts for each KPI and says why.

Yes. Every KPI has a DAX measure. The table and column names are examples, so rename them to match your model. Star several KPIs and copy all their measures at once from My KPI set.

They are typical ranges as they are often reported, given only where such a range holds broadly, and a starting point rather than a standard. Your own history is usually the better target: compare with your last periods first.

Yes. Everything runs in your browser and nothing you choose is uploaded. Your starred KPIs are kept in your browser; a shared set travels inside its link.

No. Every motion is plain HTML and CSS: keyframes, transitions and a few CSS custom properties. Click motions, like the toggle switch, need three lines of script to switch a class, and pointer motions, like the 3D tilt card or the magnetic button, a short function that follows the pointer. Motion Studio writes those for you, in plain JavaScript or as a React effect, a Vue component or a Svelte action.

The CSS of every motion ends with a prefers-reduced-motion rule that turns its animation off and leaves it in its resting state: the spinner is drawn, the text is shown, the progress bar is full. That respects the setting in Windows, macOS, iOS and Android.

Open the motion, copy the HTML where it should appear and the CSS into your stylesheet. For React, Vue or Svelte, copy the component instead. The class names start with the motion's name, so they will not clash with your own styles.

It sets how the speed changes during a motion. Ease out starts fast and lands softly, which suits things arriving; ease in starts slowly, which suits things leaving; a spring overshoots and settles. In CSS it is a cubic-bezier() with two handles, or a linear() list of points for springs and bounces.

Yes. The Export tab records the motion frame by frame in your browser and saves a GIF, or an MP4 or WebM video where your browser can record one, at the size and frame rate you pick. A loop is recorded for one, two or three cycles so it repeats without a jump.

Yes. The motions are free to use, change and ship in personal and commercial work, with no credit needed.

Create a public repository with exactly the same name as your GitHub username and add a README.md file to it. GitHub then shows that README at the top of your profile page. For example, the user octocat would create the repository octocat/octocat. Profile README Studio writes the README.md for you and links to the right GitHub page to create or edit it.

Most stats cards are images drawn by free community servers, and some of those servers are paused or rate limited because they share GitHub API limits. The studio draws its cards itself: a GitHub Actions workflow in your own profile repository redraws them every day with that repository's token, so no public server can break them. Import turns the cards of an old README into these.

No. The studio draws every picture in your browser and loads nothing from other sites, so no widget service sees you. It talks to GitHub only when you enter a username, fill blocks from GitHub, import, check your pictures, run the Profile Doctor or publish. Two optional blocks, the visitor counter and Spotify now playing, show pictures from their own services: the studio shows an example of them, and on your profile GitHub loads them through its image proxy.

Very close, because it uses GitHub's markdown styles and the same list of allowed HTML. Style attributes are removed just like on GitHub, and picture sources switch with the light or dark preview. Small differences can remain, such as code colours and emoji images.

Yes. Import fetches your current README from GitHub or takes pasted markdown, turns known widgets into blocks and keeps the rest as markdown. READMEs made in the studio come back exactly, block for block.

It scores a profile README out of 100 using 9 rules. These cover whether it says who you are, broken widget hosts, whether its pictures load from GitHub, dark mode, badge walls, contact details, showing your work, length and alt text. It can then open a fixed version next to the original, with the widgets drawn by the studio.

Yes. There are organization layouts, a contributors wall and a hiring banner. An organization profile README lives in the organization's public .github repository, in the file profile/README.md.

No account is needed. Your layout autosaves in your browser, and the address bar always holds a link to it that you can share. Publishing with a GitHub token is optional, and the token is only sent to GitHub.

Describe the subject first, then the setting, the light, the style and the framing, in plain words. Be specific about what you want to see ("a red fox asleep in fresh snow, soft overcast light, watercolour") rather than stacking quality words like "8K" or "masterpiece". Add parameters such as --ar for the aspect ratio at the end, and use --no to list things to leave out. Starting from a prompt whose result you have already seen, like the ones in this gallery, saves a lot of trial and error.

Yes. There is no account, no paywall and no sign-up. The gallery runs in your browser as a static page.

Type a model name in the search to see the prompts written for it. The most common are Nano Banana, Midjourney, Flux, Imagen, Ideogram and GPT Image, and some prompts also name Seedream, Firefly, Stable Diffusion, Recraft or Flux Kontext. Most prompts work across several models.

Copy the prompt, open your image tool, attach your photo and paste. Prompts that mention "the subject from the reference photo" are written for exactly this, so the tool keeps your face or product and changes the rest.

Yes. You can use the prompts and the example images for personal and commercial work, without credit. Only republishing the prompt texts themselves, for example in another prompt collection, needs credit (CC BY 4.0), as the terms of use say. The example images are AI-generated with Google Gemini, so label them as AI-generated where the law requires it.

Yes. Open a prompt and use Customize to fill in its placeholders, add details, choose an aspect ratio or apply your brand kit (name, colours, heading font and mood). The copied prompt then includes your changes, and the brand kit is saved in your browser only.

A negative prompt lists what you do not want in the image, such as extra fingers, watermarks or blurry text. Every prompt in the gallery comes with one. Some tools have a separate box for it, while in Midjourney you add it after --no.

A Power BI theme is a JSON file that sets the default colours, fonts and visual formatting for an entire report. You import it once and every visual on every page takes on the look, instead of formatting each chart by hand.

In Power BI Desktop, go to View, then Themes, then Browse for themes, and select the exported JSON file. The report restyles straight away, and visuals you add afterwards take on the theme too.

Those visuals were coloured by hand in Power BI, and a theme only sets defaults, so it cannot override a manual colour. Select the visual, open the Format pane, find the colour setting and choose Reset to default, then apply the theme again.

Yes. Every colour is checked against the 3:1 contrast floor for chart marks (WCAG 1.4.11), with APCA shown alongside, and a matrix flags pairs of colours that are hard to tell apart. A colour blindness preview applies the simulation to the live report, and a one-click fix raises any colour that fails.

Not from the studio. A Power BI theme JSON can embed one background image, written into the file as text, but it is then the same on every page and makes the theme file large. The studio downloads your background as a PNG instead, which you set per page in Power BI Desktop under Format page, Canvas background.

Design the page in Layout, then download it as a PNG at the page size of your report. In Power BI Desktop, select the page, open Format page, Canvas background, choose the image, set Image fit to Fit and transparency to 0%. Then place your visuals on the tiles. Put page navigation buttons on the navigation bar.

The page layout does: pick the Tableau or Looker Studio page size and use the PNG as a background image (in Tableau, a floating Image object behind the other objects; in Looker Studio, an image sent to the back of the page). The theme JSON is a Power BI file; the colours and the accessibility checks apply to any BI tool.

Yes. The SVG and Copy for Figma keep every zone as a named layer (Navigation, Title bar, Section 1, Tile 1.1), and the PowerPoint export is one slide with the background as a picture and every zone as a named rounded rectangle you can move and recolour.

No. Everything runs in your browser. The images you extract a palette from, the themes you import and the files you export never leave your device, and there is no account.

Yes. Save a design as a preset in your browser, or copy a share link that holds the whole design (colours, font, mode, page layout and the look of every visual) in the address itself, so nothing is stored on a server.

It is the line under the prompt in Claude Code that shows live information: the model, the folder, the git branch, how full the context window is, what the session costs and more. Claude Code runs a script you choose and prints its output there. Statusline Studio writes that script for you.

Open the builder, pick a template, then add, remove and style segments while the preview updates. Click Copy install command and paste it into a terminal: it saves the script to ~/.claude/ and sets statusLine in ~/.claude/settings.json. Restart Claude Code to see it.

Yes. There are segments for context used and remaining, tokens, cache reads, session cost, burn rate, and the 5-hour and 7-day limits of Pro and Max plans with their reset times.

No. Text symbols and emoji work in every terminal. Nerd Font icons and the powerline look need a Nerd Font installed and chosen in your terminal; the builder shows which templates need one.

The Bash version needs jq (and awk for some segments). The Python and Node.js versions need only their interpreter. Git segments call git. The builder lists the exact dependencies of your design as you build.

Most often the script is not where settings.json points, Claude Code has not been restarted, or the folder is not trusted yet. Open Generated code and click Not working?: its one-paste check looks at the file, its permission, the tools it needs and settings.json, then runs the script with sample data and shows any error.

No. Your design is kept in your browser, and a share link carries the whole design inside the link itself. Nothing you build is sent to a server.

In ~/.claude/projects on macOS and Linux and %USERPROFILE%\.claude\projects on Windows. There is one folder per project and one .jsonl file per session, holding every prompt, response, token count and tool call.

No. Usage Insights for Claude Code has no server: it is plain files your browser runs. Your transcripts are read on your device, the numbers are worked out in your browser tab, and nothing is sent to me or anyone else. Prompt and response text is dropped as the files are read, and only your settings (view, range, baselines, seats and the ids you export with) are remembered in your browser.

Claude Code records each session's total cost in its transcript, and the dashboard adds those up. Events from other sources carry their own cost_usd. On a flat-rate plan the figure is what the usage would have cost at API prices, not what you paid.

Each person opens their own folder here and uses Export to save their events with a pseudonymous id and a team id. One person then loads all the files together. Any gateway or proxy that writes the catalog's unified event schema works the same way.

Every session that ended in an output, a commit or an accepted edit counts as a task, simple or agentic. Each is multiplied by a baseline in minutes that you set, and the result is shown as a range of plus or minus 30 percent. It is an estimate, and it says so.

Groups smaller than five people are merged into Other teams, or hidden when even that is under five, so nobody can be singled out from an aggregate. The same rule applies to retention cohorts.

The Catalog view lists the OpenTelemetry metric names Claude Code exports, for building the same widgets in Grafana or Datadog. The dashboard itself reads the transcripts and the unified event schema rather than an OpenTelemetry collector.

In ~/.codex/sessions on macOS and Linux and %USERPROFILE%\.codex\sessions on Windows, in folders by year, month and day, one rollout-*.jsonl file per session. Codex compresses older files to .jsonl.zst; this tool reads both. If you set CODEX_HOME, the folder is there instead.

No. Usage Insights for Codex has no server: it is plain files your browser runs. Your session files are read on your device, the numbers are worked out in your browser tab, and nothing is sent to me or anyone else. Prompt, response and command text is dropped as the files are read, auth.json and config files are never opened, even inside a dropped folder, and only your settings (view, range, baselines, seats and the ids you export with) are remembered in your browser.

Yes, when you use Codex with a ChatGPT sign-in. Codex stores how much of each window you had used with every token count, and the dashboard shows the highest share per day, your peaks and how often a limit was reached. With an API key there are no such windows to show.

Not from the session files: they hold tokens and usage limits, not prices. The dashboard shows input, cached, output and reasoning tokens per model instead. Events from a gateway that carry cost_usd are added up when you load them.

Everyone opens their own ~/.codex/sessions folder here and exports their events, under a pseudonymous id and a team id. One person loads all those files at once and gets the team. A gateway or proxy that writes the catalog's unified event schema can feed it the same way.

A skill you pick with $name or /skills is counted as yours; a skill Codex picks itself is counted when it reads that skill's SKILL.md. A subagent counts once per spawn_agent call, with its time and tool calls taken from its own session file.

A team of fewer than five people is folded into Other teams, and hidden if even that has fewer than five, so no one can be picked out of a total. Retention cohorts follow the same floor of five.