Keep up to date on the latest products, workflows, apps and models so that you can excel at your work. Curated by Duet.

Stay ahead with the most recent breakthroughs—here’s what’s new and making waves in AI-powered productivity:

AutoDoc - Better Notes, Ready to Share

Stop taking notes and start paying attention. AutoDoc is a free AI meeting assistant from the makers of Duet Display. It records your meetings, including the slides and demos a transcript would miss, and turns them into organized notes without sending anything to the cloud. The new 1.2 release is its biggest update yet. Notes have been redesigned with a clear summary up top and the conversation organized by topic. You can now export them as PDF, Word, or Markdown to share with your team, and AutoDoc now runs much faster on computers with less than 16 GB of RAM. AutoDoc is free, open source, and available for Mac and Windows.

Apolosign Digital Calendar 2.0

Apolosign released a major 2.0 software update for its smart home digital calendar, bringing consumer focused AI tools to family organization. The new update features AI recipe tools that scan photos or web links to extract ingredients and cooking steps automatically, turning raw web content into structured meal plans. It also adds AI background removal for custom visual stickers, helping families coordinate schedules and tasks effortlessly on a central home display.

AskNicely Claude Connector

AskNicely introduced its Claude Connector, bringing live customer feedback and survey metrics directly into Anthropic Claude. Small business teams and marketers can query live customer sentiment, analyze feedback trends, and surface service issues using natural conversation without exporting spreadsheets. The tool streamlines customer experience management and helps teams route customer reviews to key marketing platforms automatically.

Venuelocity vProposal 

Venuelocity by Tambourine officially released vProposal, an AI enabled digital proposal system created for hospitality businesses, venues, and event planners. The platform automatically generates high end digital proposals complete with interactive floor plans, customized AI written copy, and embedded 360 degree virtual venue tours. By syncing real time event data directly with property management systems, small sales teams can close client deals significantly faster.

Accor ALL Concierge

Accor rolled out ALL Concierge, an AI powered conversational travel companion designed for everyday travelers as highlighted. The consumer app offers personalized destination recommendations, compares hotel amenities based on user preferences, and manages bookings in smooth conversational exchanges. Travelers gain access to an on demand virtual assistant that synthesizes guest reviews and provides instant local guidance throughout their stay.

Accor ALL Concierge Streamlines Travel Planning with Agentic Intelligence

Accor recently expanded its digital hospitality platform with the launch of ALL Concierge, an advanced conversational travel assistant integrated directly into the ALL loyalty application. Designed to simplify trip planning and hotel stays, the tool acts as an around the clock personal travel companion. It uses generative artificial intelligence to understand natural language requests, allowing users to discover destinations, compare properties, and organize itinerary details through simple conversational exchanges.

The standout feature of ALL Concierge is its real time itinerary synthesis and property match engine. Rather than forcing travelers to filter through hundreds of search results or read endless guest reviews, the assistant instantly analyzes guest preferences alongside verified hotel data. It compares amenities, room categories, dining options, and local attractions across thousands of global properties in seconds. The assistant can even generate customized day by day schedules tailored to specific user interests such as family activities, fine dining, or quiet wellness retreats.

For frequent travelers and small business owners managing business trips, this feature delivers massive time savings and removes the hassle of travel management. Travelers no longer need to jump between multiple browser tabs to compare hotel amenities, check local transportation options, or read conflicting online reviews. By delivering precise recommendations backed by real time availability and personalized preferences, ALL Concierge empowers users to make confident booking decisions in minutes rather than spending hours on manual research.

Beyond pre-trip planning, the concierge continues providing value throughout the entire stay. Once checked in, guests can use the assistant to request on property services, reserve dining tables, or get live recommendations for nearby cultural events. The system interfaces directly with hotel management systems to ensure guest requests reach property staff immediately. By blending intuitive conversational AI with direct hotel operations, Accor offers a seamless digital guest experience from initial search to final checkout.

How to Ask for a Picture

Here’s a five-minute test that tells you almost everything about where AI image tools have landed.

Open ChatGPT or Gemini and ask for a birthday card that says “Happy Birthday, Margaret” across the top.

Two years ago, that request was a running joke. You’d get something card-shaped and pretty, with lettering that looked confident from a distance and fell apart up close — letters that were almost words, a doubled letter here, a missing one there, sometimes a second garbled line of text nobody asked for. Every guide to AI images had a section on this, and the advice was always the same: don’t ask for words.

Now it works. Clean script, correct spelling, typically on the first try. Same for a flyer with a headline, a date, a time, and three lines of detail.

That single change is worth understanding, because the reason text used to fail — and why it got so much better — explains what these tools actually are. Once you know what they are, you stop using them badly.

It was never spelling

The older image models didn’t write letters. They drew shapes that looked like letters, the same way they drew shapes that looked like hands or shadows or fur.

Which is why “Margaret” came out as a plausible-looking arrangement of letter-ish marks. The model had no real grasp of a word as a fixed sequence of characters. It had a very good sense of what confident lettering on a greeting card looks like.

What changed is that newer models got much better at connecting language to what they draw — including the exact letters in a word, where those letters go, and what they should look like.

That’s the thing to hold onto. You are not searching a vast library for a picture that already exists. You’re describing a picture to something that now understands your description remarkably well — which means the quality of what you get back depends largely on how well you’ve described it.

Which brings us to the habit worth breaking.

Stop typing incantations

If you’ve read anything about AI images, you’ve seen prompts that look like this:

portrait of a woman, 8k, ultra detailed, masterpiece, award winning photography, trending on artstation, hyperrealistic, octane render

That style comes from 2022, when these models had shallow language understanding and people discovered that certain magic words nudged the output toward better-looking results. It was folklore, and it worked well enough to spread.

It’s over. OpenAI’s own prompting guidance now says plainly that “short prompts, descriptive paragraphs, JSON-like structures, instructions, and tags can all express the same intent,” and tells you to pick whichever is easiest to read and update “rather than relying on special syntax.” There is no incantation. There is no secret word.

Google’s guidance puts the replacement principle in one line: “Describe the scene, don’t just list keywords.”

The practical version: spend your words on concrete visual details — things a person could actually see in the finished image. “Masterpiece” describes nothing. “Late afternoon light coming through a dusty window” describes something.

Here’s the twist: paste that keyword pile into a modern tool and you’ll get something perfectly polished — a clean, professional studio portrait. The models are good enough now that the incantation doesn’t hurt, and a couple of those words — “hyperrealistic,” say — do nudge the style. But none of them say anything about your picture, so you get the model’s default idea of “portrait of a woman,” not yours. Swap the empty words for descriptive ones and you stop getting a stock photo and start getting the picture you had in mind.

And don’t overcorrect into an essay. OpenAI’s own beginner guidance says “a good image prompt does not need to be long. In most cases, 1–3 clear sentences are enough.”

The things beginners leave out

Here’s the shift that matters. A beginner types a noun phrase — a cat in a hat — and accepts whatever comes back. Someone good at this specifies the decisions they’d otherwise be leaving to a machine.

Google’s consumer guide organizes it as five questions, and they’re a genuinely useful checklist:

• Subject — who or what is in it, what do they look like, what are they wearing, how are they posed

• Action — what’s happening

• Setting — where, and when

• Style — a photograph, an illustration, a watercolor, a pencil sketch

• Composition — close-up or wide, portrait or landscape, eye level or low angle

Then there’s the tier above that, which is where results start looking deliberate instead of generic: light and lens. Google’s advice for realistic images is to think like a photographer, and its own template runs roughly: a photorealistic [shot type] of [subject], [doing something], in [setting], lit by [lighting], creating a [mood] atmosphere, captured with [camera or lens], emphasizing [textures], in [aspect ratio].

You don’t need to know photography to use that. “Soft light from a window on the left” is a real instruction. So is “shot from low down, looking up.” So is “85mm portrait lens, blurry background.” One honest caveat from OpenAI’s docs: treat camera specifications “as cues for appearance, not a guarantee of exact physical simulation.” Saying f/1.8 isn’t operating a lens. It’s requesting a look.

The single most underused instruction is telling it what the picture is for. Google’s guidance makes this point directly: “Create a logo for a high-end, minimalist skincare brand” beats “Create a logo,” because purpose carries an enormous amount of information about style, restraint, and color. Almost nobody thinks to do this.

And decide your shape at the start, not by cropping later. Aspect ratio changes what the model composes — ask for a vertical poster and it will build a vertical composition, often leaving usable empty space. Cropping afterward rarely gets you there — and if you need room for words later, ask for the empty space outright.

Putting words in the picture

Back to Margaret. Text is no longer the thing to avoid — it’s one of the most useful things these tools do. Birthday cards, yard-sale flyers, a sign for the bake sale, a menu for a dinner party. The current tools handle all of it well.

What separates a good result from a merely correct one is the same thing as everywhere else: direction. The vendors’ advice is specific and worth following:

• Put the exact words in quotation marks. Google: enclose your desired words in quotes, then describe the typography — “a bold sans-serif font,” “neon cursive signage.”

• Give it the hierarchy. On a flyer, say what’s the headline, what’s the date and time, and what’s the small print. The model will size and arrange them accordingly instead of guessing.

• Say where it goes. “Centered across the top.” “In the lower right corner.” Placement is part of the design.

• Say “no other text.” Otherwise you may get a helpful extra tagline you didn’t ask for.

And if the text changes — wrong date, new venue — you don’t have to start over. Ask for just that change: change the date to Saturday, October 10, and keep everything else exactly the same.

One habit is still worth keeping, and it’s the same one you’d apply to anything you print: read it over before it goes to the copy shop. It rarely slips now, but a wrong digit in a phone number is expensive once there are a hundred copies of it.

How to stop it looking like AI

There’s a particular look — glossy, over-lit, faintly airbrushed, everything symmetrical and nothing out of place. Readers have learned to spot it and mildly resent it.

The fix is the most counterintuitive thing in this whole subject. In 2022, people added words like “cinematic” and “dramatic lighting” to escape flat, muddy output. Now you subtract them.

Look at how OpenAI writes its own example prompts for photographs meant to read as real. One asks for an image that looks “like a real photograph someone could have taken, not an overly enhanced or cinematic movie-poster image,” and ends by telling the model to “avoid cinematic lighting, dramatic color grading, or stylized composition.” Another asks for visible pores and wrinkles, worn materials, natural color balance, subtle film grain, a candid unposed feel — then explicitly rules out glamorization and heavy retouching.

That’s the move: ask for ordinary light, ordinary framing, and visible imperfection. Say “candid.” Say “unposed.” Say “no retouching.”

One qualification, because it cuts both ways: polish is a choice, not a defect. If you’re making a greeting card or a product shot, soft flattering light is exactly what you want. The AI look is only a problem when you were trying to make something that looks real.

The second move matters more than the first

Beginners treat these tools like a slot machine — prompt, dislike the result, delete, write a new prompt, pull again.

That’s the wrong model entirely. Google’s documentation states that “multi-turn conversation is the recommended way to iterate on images.” You’re meant to keep talking.

The phrasing that unlocks this is some version of: change only the jacket to dark green. Keep everything else exactly the same — same pose, same lighting, same background. Naming what must stay is as important as naming what changes, because otherwise the whole image drifts.

Three rules that make iteration work:

Change one thing at a time. If you adjust four things at once and it gets worse, you’ve learned nothing.

Restate what you’re protecting. Every time. “Same face, same composition.”

Start over when it drifts too far. Google’s own advice: if a character’s features wander after many rounds of editing, open a fresh conversation and describe them properly from the top. Sometimes the fastest fix is a clean sheet.

Same dog, second picture

The thing people most want, and find hardest: their actual dog, their actual kid, a character who stays recognizable across a whole set of images.

Two techniques do most of the work.

Upload reference photos. This is what they’re for. Give it clear pictures of the subject and ask for a new scene. ChatGPT’s image model was upgraded earlier this month specifically to be better at carrying a recognizable subject into new settings, and Gemini will accept several reference images for a single character.

Give it a name. Google’s guide recommends assigning a distinct name to each character or object, then referring to that name in later prompts. “Keep Rufus exactly as established” is a surprisingly effective instruction.

And if you’re generating images of real people: use a reference photo, and make sure you have their permission. That’s OpenAI’s own guidance, and it’s right.

Which tool, honestly

Most articles on this subject hand you eight options. You need about three.

Start with the assistant you already have. ChatGPT and Gemini both generate images on their free tiers, both are among the highest-rated image models available to consumers, and neither requires a new account or a new credit system. ChatGPT ships templates literally labeled “Poster” and “Merch,” plus a sketch feature that lets you draw a rough shape and have it rendered properly — useful for people who can’t put what they want into words. Gemini’s free tier hands you the same model paid users get by default, and it handles text in many languages — it will even translate the words inside an existing image while aiming to leave everything else as it was. Pick whichever you’re already signed into. Neither company publishes how many free images you get; both will simply tell you when you’ve hit a limit.

Use Canva when the job is a layout, not a picture — something you’ll print at an exact size, reuse every month with new details, or hand to someone else to tweak. Generate the artwork anywhere, drop it into Canva, and add your wording as separate text boxes so you can update it anytime. Pro runs $18 a month, and the free tier includes a small monthly allowance of AI generations — though for the layout part you don’t need the AI at all.

Bing Image Creator is the answer if you refuse to pay for anything, and it deserves more credit than it gets. Free with a Microsoft account, a daily allowance of fast generations and slower ones after that, no subscription anywhere in the flow.

And the ones to skip, at least at first:

Midjourney still produces the most distinctive-looking work, and serious hobbyists should eventually go there. But there’s no free trial on its website, it starts at $10 a month, asks you to think about GPU minutes, and — a detail that surprises people — your images are public by default, browsable and remixable by strangers, unless you pay for a plan several tiers up and switch on its Stealth mode. That’s a bad fit for a beginner making a family photo.

Ideogram is the tool everyone still recommends for text in images. It earned that reputation and has since moved upmarket toward developers, and the mainstream assistants have largely caught up. Its free tier won’t even commit to how many generations you get.

Adobe Firefly markets itself on being commercially safe, trained on licensed and public-domain material. That claim about the training data is real. But the legal protection people assume comes with it — indemnification if someone makes a claim against you — depends on your plan, the model you pick, and the agreement you signed. A standard consumer subscription generally doesn’t include it. And Firefly now offers third-party models from a dropdown, which changes the picture again.

Leonardo is a cheap way to reach many models at once. Read the next section before using its free tier.

Who owns the picture

Short answer for the two tools most readers will use: under their terms, you do.

OpenAI’s terms are the most explicit of anyone’s — you “own the Output,” and OpenAI assigns you whatever rights it might have had. Google’s are softer but land in the same place: “Google won’t claim ownership over that content.” In both cases you’re free to use what you make, including commercially, subject to the service’s terms and other people’s rights.

One distinction worth knowing: that’s about your agreement with the company, not copyright. The U.S. Copyright Office has said that prompting alone doesn’t make you the author of an AI image in the copyright sense — though your own creative additions, arrangements, and edits can be protected. So you can use and sell the picture. You may not be able to stop someone else from copying it.

But the free tiers of some smaller tools are a genuine trap, and nobody mentions it:

• Leonardo’s terms state that for free subscriptions, ownership of the intellectual property in your output vests in Leonardo — and free generations are public, which means other people can reuse them. You’re still allowed to use your images commercially; what you don’t get is ownership or privacy. Those start at the paid tier.

• Recraft says outright that images generated on its free plan are owned by Recraft and are not licensed for commercial use.

If you’re making a logo for something real, that matters enormously. Check before you build a business on a free tier.

Two more things worth knowing. Both ChatGPT and Gemini now embed invisible provenance markers in the images they generate — the same underlying watermarking technology, plus metadata recording that it was AI-made — though coverage can vary by product and how the file is exported. You can’t turn the invisible marking off. Gemini’s visible sparkle logo is a setting you can switch off in most countries, but the invisible one stays.

And Google’s terms of service list “misleading others into thinking that generative AI content was created by a human” as a prohibited use. Which is a reasonable line, and a good one to keep on your own initiative.

The five-minute version

If you read none of the above: describe what you’d see rather than how good you’d like it to be. Say what the picture is for. Pick your shape before you start. Put the words you want in quotes and tell it where they go. When something’s close, ask for one change and name everything that should stay.

The tools stopped needing magic words a while ago. What they need now is someone who knows what they want to see — which is a harder skill, and a much more useful one.

Margaret’s card was never the hard part. It just used to look like it.

Partner Spotlight: Periscope VPN by Duet Display

Periscope by Duet Display provides a fast and reliable solution to keep your online activity protected across all your devices. You connect from your remote device to a a computer or AppleTV in your home so that you remote activity appears as if you are at your house.

The software is remarkably easy to set up, allowing you to establish a secure connection in just a few clicks. Periscope safeguards your connection using state of the art end to end encryption, ensuring that sensitive personal information, business communications, and financial data remain completely private. Built on a steadfast commitment to privacy, Periscope enforces an absolute no logging policy, meaning your browsing activity is never tracked, recorded, or saved. Periscope runs on Mac, Windows, iPhone, iPad, and AppleTV. Learn more and get protected today at getPeriscope.com.

Stay productive, stay curious—see you next week with more AI breakthroughs!