ToRun, the complete guide: every feature explained in 16 minutes

Images, video, music, and a voice for your words. You describe what you want, you see the cost, and it gets made. Here's how each one works. Open Image generation and describe what you want to see. Auto picks the model we recommend for the job: good results at the best price, and you can choose another one yourself. The estimated cost shows before you press Generate. To change a photo, switch to Edit. Upload it and write what should change, like the sky. The rest of the picture stays as it was.

Video generation works the same way. Write what should happen, choose the length and the resolution, and start from a picture if you like. The price depends on both, and the estimate updates as you change them. Not sure how to put it? Enhance prompt turns a short idea into a clearer, more detailed prompt before you run it. For music, describe the song and pick a style. Add your own lyrics, or switch vocals off for an instrumental. The finished track goes to your library. Text-to-Speech reads any text aloud.

Paste it, pick a voice and a speed, and download the audio file. Want something personal? Moments templates turn your photos into a short video, with a song written for the occasion. Everything you make lands in your library, and every call is recorded with its own cost in your billing history.

In ToRun chat, you just ask. The right model answers, searches the web when it needs to, researches in depth, and finds places on a map. Leave the model on Auto. For each message it works out what the turn needs, keeps the models that can do it, and picks the best value. Attach a photo, and that turn moves to a model that can see it. Prefer a particular model? Pick it yourself. An explicit choice is always honoured, and nothing cheaper is swapped in. When a question needs current information, the assistant searches the web on its own, in the middle of its answer, and attaches the sources so you can check them.

It can use tools too: make an image, write a note, add to a list, set a reminder or start a workflow. What it did shows in the reply. For bigger questions, switch on Deep Research. It runs many searches, reads what it finds, and writes one answer with its sources. It usually takes a few minutes, and the exact cost shows in your billing history. Ask for a place, like a pharmacy open now near you, and you get a map with every result, nearest first, and whether each one is open.

Your location is used only for that search.

Projects keep your work together, your own files turn into answers, and memory means you don't have to repeat yourself. Here's how. A project groups related chats, files and memory. Give it a name and a system prompt, and every new chat in it follows that prompt. Under Knowledge, add PDFs, documents or text files. The assistant searches them in this project's chats. So when you ask about your prices, the assistant looks them up in your own price list. Memory works across all your chats.

Mention something worth keeping, like a diet or the units you prefer, and it remembers, so you don't have to say it again. Everything it remembers is listed on your memory page. Edit any item, or tell it to forget, and it's deleted for good. Things you stop using fade over time.

Talk with someone in another language, get your words written down as you speak, or have a video transcribed, translated and subtitled. Here's how. Open Live Translation, choose the language you speak and the one to translate into, and start talking. Your voice is translated as you go. It works one direction at a time. Tap swap, and the other person answers in their own language. Both sides appear in the transcript, and you can save the whole conversation as a note. Live Transcription writes down what you say in real time.

The language is detected on its own, or you can set it. Have a recording already? Upload the audio file in Speech-to-Text, and get the text back. It works with videos too. Pick a video and a language, and you get the transcript, the translation and subtitle files in both languages, saved to your library.

When something's ready, share it with a link. See what every call cost. And take all of it with you on your phone. In a chat, press Share. Choose when the link expires and whether your uploaded files go with it, then create it. Anyone with the link sees a frozen copy of the chat, without signing in. You can see how often it was viewed, and revoke it at any moment. The link stops working right away. Workflow results, canvases and media from your library are shared the same way, and a pasted link shows a preview card with the title and the first image.

Your billing history lists every AI call on its own line: when, what kind, which model and what it cost. Filter by kind or date, and open any line to see how its price was worked out. And it all comes with you in the mobile app, on iPhone, iPad and Android, and on Apple silicon Macs. Chat, create, talk to the voice assistant, and see the same costs as on the web.

You fill in one form, and a finished video comes out. In the next two minutes, you'll see exactly how. This is a workflow. Think of it as a recipe: a few AI steps, already put together, that turn your answers into a finished result. There are workflows for images, videos, songs, voiceovers and written work: bedtime stories, birthday slideshows, product shots, jingles, read-alouds and more. Pick one, and you'll see the steps it runs, before you commit to anything. Then fill in the form.

Every field starts with a sensible default, and the dropdowns come with ready-made choices. There's no prompt to write. Before anything runs, you see the estimated cost. Not sure yet? Run a cheap preview first, to check the direction. The preview is a quick, low-cost draft, not the final result. When you're happy with it, run it for real. Now every step runs on its own. First, the story is written. Then the narration and the illustrations are made at the same time. Finally, everything comes together in one video.

Your video is ready. It's saved in your library, where you can watch it or share it publicly. Your inputs are never shown on the public page. Every run is kept in your workflow history, with its status and cost. Open any run to see what happened, or re-run it in the editor and make it your own.

What actually happens when you press Run? Let's open one workflow and look inside. A workflow is a graph. Each box is a step, and the arrows set the order. When two steps don't depend on each other, they run side by side. Each step has one job: writing, speaking, drawing or editing video. For every job, ToRun picks a model that can do it. Your answers from the form flow into the steps. The story step writes for Ada, about a brave little fox. Then its story is passed on, to the narration and to the illustrations.

While it runs, the run page shows every step: its status, how long it took, and what it cost. Some workflows check their own work. If a step falls short of the bar, it runs again automatically, and every attempt stays visible. Some steps wait for you. A song workflow, for example, shows you the lyrics first. You can approve them, edit them, or ask for new ones, before the music is made. Every AI call in a run is recorded on its own, with its own cost: the story, the narration, every single illustration.

So the total is never a mystery. When the last step finishes, the result lands in your library, ready to share. Your inputs always stay private.

Every workflow, from a bedtime story to a research report, is built from the same few blocks. Let's go through them. The main block is the AI step. It sends a prompt to an AI model and gets back text, an image, a voice, music or a video. Every call is billed on its own. Knowledge blocks bring in facts: a web search, a search through your own files, and a citation step that formats the sources an answer used. Quality blocks check the work. A quality gate scores a step's output and sends it back for another try if it falls short.

A human approval pauses the run until you've had a look. Logic blocks shape the flow. A branch follows the arrow whose condition matches. A loop repeats its steps for every item in a list. And output blocks finish the job: a media step joins images, voice and music into one video, and a save step stores the result. Every template is made of these blocks. Open any of them in the editor, and you'll see exactly how it's built.

Found a template that's almost right? Open it in the editor, and make it your own. Here's how. Every template has an Open in editor button. It opens the same steps on the editor's canvas. Each box is a step, and each arrow passes a result along. Click a step, and it opens in the inspector. Here, the story step asks for a gentle story. Make it funny, and in rhyme. The name and the topic from the form still flow into the same places. Need a check? Drag a quality gate in from the palette, drop it after the illustrations, and connect it with arrows.

Then save it to your workspace. The template itself doesn't change; what you saved is your own workflow. Press Run this, fill in its form as before, and your version runs, step by step.

One question in, a cited research report out. Let's follow one through the multi-source research template. You type the question, then choose which sources to search, how deep to go, and a citation style. First, an AI step splits the question into a few smaller sub-questions. Then three searches run side by side: the web, academic sources and your own files. Each one runs only if you picked that source; the condition sits on its arrow. Next, an AI step reads everything that came back and checks the sources against each other.

A citation step formats the sources in the style you chose, and a last AI step writes the report around your sub-questions.

Write a story, pick a channel, and get back a finished episode. Here's what happens in between. You write the story with your characters' names, and pick a channel. Its characters, voices and branding come with it. First, a writer step turns the story into shots and lines of dialogue. Then two steps run side by side: every line is voiced in its character's own voice, and music is made for the episode. Next, each shot is made with the characters speaking, their lips in sync with their lines.

Captions are written, and a media step cuts everything together: the shots, the voices, the music, captions, titles and your branding.

Before a workflow runs, its form shows the estimated cost for your exact choices. You decide, then you run. Every template card shows an estimated cost per run, and once people have used it, the typical cost of its recent runs. A preview is a low-cost draft, and it tells you exactly what it shortened: the text, the narration, the number of images, the length of the song. While the run goes, every step shows its status, how long it took, and what it cost. Each AI call is recorded on its own, with its own cost, so the total always adds up.

And in the editor, any step can have a budget cap. If its estimated cost is over the cap, the step is skipped instead of spending more.

A visitor scans a QR code, asks a question, and gets an answer in seconds, in their own language. There's no app to install and nothing to sign up for. The code opens the venue's own page, with its most common questions ready to tap. Visitors ask in whatever language they speak, and the answer comes back in that same language. Every answer comes only from the venue's own material: the documents, floor plans and instructions it uploads. Nothing comes from the open web. It gives directions too, step by step, starting from the point where the code was scanned.

And if the answer isn't in the material, it says so in one honest sentence, instead of guessing. It's anonymous by design. It never asks for a name, a phone number or an email. The venue's team sees what visitors ask, which answers were missing, and any problems they reported, so the material keeps getting better.

Here is how a venue sets up its own AI assistant, from the documents it already has to the first answer a visitor gets. The setup page lists what's still missing before the assistant can answer, and ticks each step off as you finish it. Start with the venue's details: its name, time zone and language. Then give the assistant a name, add your logo and pick your colour. The visitor page follows your brand. Next, upload the documents you already have: PDFs, Word and Excel files, presentations, plain text, even a photo of a price list.

Each one is read and split into passages the assistant can search. From then on, every answer comes only from these documents. If something isn't in them, the assistant says so instead of guessing, and it never asks visitors for personal data. Upload a plan for each floor. Mark the places people ask about, like the restaurant, the toilets, lifts and exits. Then put a QR point wherever you'll hang a code. That's how the assistant gives directions from where the visitor is standing.

Download all the codes at once, print them, and hang each one at its point. Before anyone else does, open a visitor page and ask a question yourself, just as a visitor would. Once visitors start asking, the demand report shows what they ask most, and what they looked for but couldn't find. So you know exactly what to add next. AI assistants for venues, on ToRun.