The underlying LiteRT-LM framework used in the edge gallery does support the MTP drafters for the smaller models, but according to:

https://developers.google.com/edge/litert-lm/models/gemma-4

> Note: LiteRT-LM supports E2B and E4B models today, with support for larger models coming soon.

So even Google aren't shipping MTP support for the 26B and 31B models yet.

(Though perhaps it'll squeeze in with a small context window? Not sure I understand that aspect yet)

It does seem to use MTP, yes, and it is quite quick — seemingly the underlying LiteRT stuff can do MTP with Gemma 4 and presumably MTP is a big part of the practicality picture here.

The system prompt thing was a surprise when I poked around.

https://developers.google.com/edge/gallery

Anyone with a 16GB Mac — that is quite a lot of journalists, surely — can download that, install a model into it, and play.

Surely journalists have to start asking questions at least about OpenAI's consumer revenue projections now.

I am a major, major AI cynic, but I decided to be an informed cynic so I've been playing with local models for agentic work and a bit of CAD-to-image generation. I really quite like the 26B Gemma model — I've been using it to teach myself some fundamental things and learn OpenCode without developing a cloud dependency. It writes fairly good code and it is helping me learn the things I want to learn at a pace that I prefer.

But if this 12B model is even half as close as they say it is, this casts some doubt on the consumer end of the cloud business model, at least in the short term.

(Not clear if this app is using the MTP drafters; I've still not got them working with Gemma myself, though the Qwen 3.6 built-in MTP support is super in LM Studio)

But what I would observe is that the healing brush does not have a random seed. It will always do the same thing if applied at the same pixel.

(I am actually less sure if content-aware fill randomises; I always got the impression it did not)

This makes it both incredibly powerful and occasionally frustrating.

Because on the one hand, you can learn to apply your judgement to precisely control what it will do, and change the radius or position if you learn it is likely to fail, which becomes instinctive. I absolutely love using it to fix scratches in film scans; it's a quick, precise, controllable tool that can be used in a way that is amazingly convincing, and it ends up quite a "zen gardening" thing as a result. It'll sell you on the cheapest wacom pen once you know how efficient it can be.

On the other hand there are situations where it simply cannot work the way you want because it will always find a pattern you don't want it to.

(You can sometimes use the clone brush tool first, to manually break up the pattern that patchmatch will find)

They are both PatchMatch (well the healing brush certainly is), which is a heroic bit of code. Entirely deterministic statistical algorithm. Not AI by really any definition (including back then)