Gemini is coming to Makersuite & so are Stubbs

The first ever look at the multimodal Gemini and an extra, secret tool that Google has hidden from everyone, leaked as well.

Bedros Pamboukian
6 min readOct 15, 2023

PaLM 2 is going to be replaced by Gemini. A more powerful, multimodal version of PaLM 2. And Gemini is coming — right to Makersuite, for everyone to try out.

Makersuite has some basic features, supports text to text outputs, but we want more. We want multimodality. Look at Bard — that supports image inputs, and some people have been complaining that they’ll just use unofficial APIs to get image input functionality into their product.

If you’re considering doing the same, don’t, because Gemini, a multimodal AI model, is coming right to Makersuite — and I have proof.
Read on and you’ll see leaked screenshots, too. Note that the leaks are legal, it’s public info, just hidden to the client.

But there’s more to the leaks than just Gemini, though that may be stealing the show. We’ve got Stubbs. What is a Stubb? Well, it’s a feature where you get to build and launch your own AI Generated app, directly from Makersuite. At the time of writing, there has been no external information about this — not even a little teaser from Google, not even a single mention. But I found it — and I’ve got access to it. Leaked screenshots below, as well.

Let’s get the minor stuff out of the way — Translation between languages will be fully supported in Makersuite, and Google will provide a sample prompt for translating between Spanish and English when it’s out, so keep an eye out for that.

Stubbs

A feature where you can create working apps directly in a site with just one prompt, it’s revolutionary, and also never MENTIONED anywhere. Google has only been teasing Gemini, but that seems to have been a distraction from what could be the biggest Google release of the year. This won’t replace app developers, but rather be a massive boost to the industry — from the looks of it, this’ll be like AI Generated Figma prototypes, and won’t create full code, but rather working AI Generated app prototypes.

User interface for Stubbs
Creating a stub? Stubb? Stubbs? Also has a very Playground AI look — I dig it. I’m seeing this dot background everywhere in AI!

Being able to create apps and launch them in just one site, with such a streamlined UI, it’s perfect. You can generate, deploy and even publish Stubbs! You publish it and you can then share the link — Is this new ground for AI? Probably! There’ll be a community gallery, where you can publish your Stubbs for everyone to see — and you can also Remix Stubbs, and have your own twist on an idea.

This leak is so early that the UI isn’t even finalized?

Featuring a Stubbs gallery, where you can view Stubbs other people made, which I find pretty cool — and yes, by default, it won’t publish — you have to explicitly publish it, so don’t worry about the public being able to view everything you create

Interestingly, Stubbs has prompt suggestions, powered by the lower-end text-bison. Why not use the supercharged, next-gen future ones? Speaking of which…

Gemini

Time to talk about the multimodality. Makersuite has a codename - Alkali. So does Gemini.

Welcome home, Jetway.

In the image above you can see a Sample Images section — these are specific to Alkali / Makersuite, so these are cherrypicked samples to showcase what it can do — and it seems like it’ll do a lot. Text recognition, object recognition, captioning, understanding the image — this’ll be fun!

Multimodal Model right under Text Bison for convenience

It also has an Output Type — you can allow it to include images — whether this will be capable of generating images or not is unknown — it’s also possible (and more likely) that it’ll include links to external images — the description says Include images and not Generate images.

In some articles you may see Gemini described as an addition to Bard. I am here to tell you… nuh uh. It’s an integration — it’ll be in Vertex AI and available to developers through Makersuite, and can then be placed anywhere.

Take a look

Home page includes Stubbs, so you’ll spot it right away when it’s added
Hovering over the image shows an expanded preview
Google Drive integration makes it super easy to add images, but it would honestly just be better if I could drag and drop the file directly into the editor — sad. Image 2: it supports copying images and pasting it in though!
It’s not just limited to the Write your prompt field — you can also test your prompt with images in a neat way.
Data prompts also support multimodality
And yes. It is Gemini.

Leaked Features

Google Stubbs — allowing you to create functional app prototypes, deploy and share them, with just a single prompt and optionally an image of the app you want to create / clone.

Google Stubbs Gallery — View and remix other Stubbs with ease in a centralized manner, as well as publish your own Stubbs to it.

Makersuite Autosave — I’ve lost work in prompts because I failed to charge my laptop in time, leaving me with no chance to save. Autosaving is coming though!

Deepmind Gemini / Jetway — Multimodal prompt creation that can take in Images (although the team says it’ll support audio as well) and seemingly output multimodal content including HTML Content.

Makersuite Translation support —Before, it would block the prompt if a certain ratio of English : Non-English text was met.
Padding the prompt with a massive amount of English text and adding a stop sequence was an easy bypass to this, at the expense of additional tokens. However, translation worked flawlessly, giving me the exact same outputs as Google Translate. Now it’ll be implemented without a filter, I suppose due to a more powerful model (Gemini / Jetway)

Limitations

Interestingly, Text and Data prompts support multimodality whereas Chat does not — compare that to Bard, a chatbot which claims to be built on PaLM 2 and supports image input. Maybe another time…

The Stubbs feature will not create full code of the app, but it will deploy a prototype, similar to a Figma prototype fully made by AI.

The image input will not support GIFs in the Makersuite UI.
It’s worth noting you can embed GIFs in there regardless, and it would still get processed — it’s just a client side filter that can easily be bypassed — to use it in the prompt you can just import a CSV with an image tag and the src attribute pointed to a base64 encoded GIF — and it’ll import just fine. (alternatively directly edit the prompt JSON through Google Drive)

Sample <img src=”data:image/jpeg;base64,b64image” class=”input-image”>

Keep in mind that these are not out yet. These are leaked features, first seen by the public right here in this post— it has been revealed that it’ll be out this year though, and I personally can’t wait.

What are your thoughts on this? Are Stubbs going to boost UX Designers, or will it be used in an attempt to replace them?

Regardless, Stubbs FTW, Jetway FTW, overall just Google FTW
Good to see Google working hard at this.

EDIT! Finally out — Technical details on Stubbs.

The leaked images are not from any external source — this (my post) is the source of all these leaks, but the design and concepts are all by the hardworking Makersuite team. As this is a leak, this is all subject to change. Stubbs might not even be out by 2024, but we can all hope!

If you wish to include the images from this article elsewhere, please place a link to this article, or include some sort of credit or acknowledgement.

--

--