Frontier AI Models, Asset References in Chat, and Two New Studio Tools

TL;DR:

• Frontier models in chat: Claude Opus 5, Claude Sonnet 5, Kimi K3, and Gemini 3.6 Flash — now at 3 credits on every plan
• Point the AI at exactly the right asset: type @ to drop a real asset into your message, no more guessing at file names
• A new video model, Omni: stronger action and motion for your characters, with cleaner, more reliable transparent backgrounds
• Asset Manager: one place to see, categorize, and clean up every asset already in your game
• Sprite Sizing: get all your sprites correctly proportioned to each other, with a render pivot that decides how each one sits on its spot
• A redesigned Code Studio editor: real file tabs, keyboard navigation, and a quick jump back to recent projects
• Chat that knows how full it really is: one honest heads-up, measured against the real conversation size
• Plus more reliable generated code, faster replies in long chats, and a round of fixes

---

More Muscle in Code Studio Chat

The AI behind Code Studio chat reads your project, writes code, fixes bugs, and wires up game logic. Different jobs call for different horsepower, so back in June we refreshed the lineup and let you pick the model that fits the task right from the chat input.

Now we've refreshed the whole lineup, including our most capable models yet.

The new lineup

The four in bold are new or newly repriced.

• Gemini 3.6 Flash is fast, multimodal reasoning — and it just got a lot more accessible. We re-measured what it actually costs us to run and dropped it from 5 credits to 3, then moved it down to the Free plan. It's also the new default for plan mode, where quick, capable reasoning matters more than raw horsepower.
• Kimi K3 is a big step up from K2.7 for deep, agentic work — the kind of task where the model has to keep a lot of your project in its head, make a plan, and grind through it.
• Claude Sonnet 5 brings frontier-level intelligence to your everyday building, from feature planning to trickier refactors.
• Claude Opus 5 replaces Opus 4.8 at the same 20 credits, and it's the most capable model we offer — built for your hardest problems, the gnarly bug, the big architectural call, the thing nothing else could crack.

Two models have retired. Kimi K2.7 has been replaced by K3, and Claude Opus 4.8 by Opus 5. If either was your saved pick, you don't need to do anything — we move you over automatically. K2.7 users land on Gemini 3.6 Flash, which is a cheaper, faster pick at the same end of the lineup, and Opus 4.8 users land on Opus 5.

As before, the model selector shows every option with its quality and per-message cost. Models above your plan still appear, locked, with an upgrade prompt, so you always know what the next tier unlocks.

---

Point the AI at Exactly the Right Asset

Here's a small frustration with a big footprint. You've generated a knight, a slime, and three tilesets. You want the AI to use the slime — so you type "use the slime sprite" and hope it finds the right one. Sometimes it guesses right. Sometimes it invents a file name that doesn't exist, and you spend the next two messages untangling it.

That guessing game is over.

Type @ to reference a real asset

In Code Studio chat, hit the @ button next to the message box — or just type @ — and a browser of everything in your game pops open. Pick an asset and it drops into your message as a pill: an exact, unambiguous reference to that specific asset.

• Browse by category or search by name — the picker groups your assets the way your game does, and typing filters as you go
• See what you're picking — each asset previews right in the browser, with details like format and duration for animations
• Keyboard-first — arrow through the tree and hit Enter, without leaving the message box
• Works everywhere you chat — the new-chat composer and follow-up messages both support it

The AI can look assets up on its own, too

Behind the scenes, the AI now has direct access to your game's asset catalog. When it needs to know what's available, or the exact details of one sprite — its real key, its dimensions — it looks it up instead of guessing. You'll see those lookups in the chat timeline like any other step it takes.

The result is less back-and-forth: the AI wires up the asset you actually meant, with the name it actually has, the first time.

---

Animation Gets the Omni Video Model

Great character animation lives or dies on the motion. A jump should feel like a jump; an attack should land. So we've added a new video model built specifically for that.

Meet Omni

Omni is our new video model, and it's the stronger pick when movement is the point — attacks, jumps, and locomotion all come out with more convincing action and motion. When you generate an animation, Omni sits right alongside our existing model in the picker, so you choose the right tool for each shot:

• Omni — best for lively, dynamic motion. Renders at 720p, with clip lengths from a quick 3 seconds up to 10.
• Veo 3.1 — still the default, and still the best at holding a character exactly on-model: identity, outfit, and art style, up to 1080p.

The short version: reach for Omni when the motion matters most, and Veo when staying perfectly on-model matters most. Both are one tap away on every animation.

Omni also gives you a choice of how it starts a shot. In Reference mode (the default) it composes its own opening frame directly from your reference art, so there's no separate keyframe step to set up. Prefer the classic approach? First frame mode still generates a keyframe first. You'll see the mode as a simple chip on the animation controls.

Cleaner cutouts, automatically

Sprites need clean edges. A muddy or uneven background is the difference between a crisp cutout and a character with a faint halo around it.

Under the hood, we reworked how generated art paints its background so it lands on a true, uniform key color every time, instead of drifting toward muddy in-between shades. There's nothing to turn on — it just means when we lift the background away, your sprites come out with cleaner, more reliable transparency and sharper edges. Better raw frames in, better game-ready sprites out.

Omni is available on paid plans.

---

Two New Studio Tools

Building a game means juggling a lot of art. Two new tools in Code Studio help you keep all of it organized and looking right together.

Asset Manager: everything in your game, in one place

Once you've added characters, backgrounds, props, and animations to a game, they can be hard to keep track of. The new Asset Manager gives you a single view of every asset that's actually in your project, with live previews — including animated sprites playing right in the grid.

From there you can:

• Sort assets into your own color-coded categories so a big project stays navigable
• Rename anything to a name that makes sense to you
• Spot the stragglers — filters surface assets that aren't used anywhere in your game, or whose art failed to load, so you can clean house with confidence
• Remove what you don't need, with clear prompts when an asset is still referenced so you never delete something out from under your game

Open it from the Asset Manager button in the Code Studio toolbar. Newly generated animations flow in automatically, deduplicated, so what you see is always an accurate picture of your game.

Sprite Sizing: get everything to scale

Here's a familiar headache: your hero was drawn at one resolution, an enemy at another, a pickup at a third. Dropped into the same scene, one of them is comically huge or tiny. Fixing that by hand, in code, is tedious.

Sprite Sizing solves it visually. Pick one sprite as your reference — the yardstick everything else is measured against — then scale the others to match:

• Drag a simple height ratio slider, or hit Match to reference to line a sprite up in one click
• Frame-0 previews show the before-and-after side by side, so you can see the proportions land
• A render pivot sets each sprite's contact point — like a character's feet — so nothing floats or sinks when you rescale it

Hit Apply and your game renders every sprite at the size you chose, from then on. No code editing, no guesswork — just characters and objects that finally look right next to each other.

Render pivots now control drawing, not just resizing

That pivot used to be about resizing only. Now it's the answer to a question every 2D game runs into: when your code draws a sprite at a position, which part of the sprite lands there?

By default that's the top-left corner, which is almost never what you want. A character walking on a platform should be placed by their feet. A projectile should fly from its center. Getting that right by hand means hand-tuning offsets in code for every sprite, and re-tuning them every time the art changes.

Now you set it once, visually. Pick any of nine points on the 3×3 Render pivot grid — or leave it on Auto — and that choice defines how the asset relates to its draw coordinate everywhere: in the Sprite Sizing preview, in animations, in hitboxes, and in the running game. Your existing games keep behaving exactly as they did; Auto preserves the old behavior, so nothing shifts under you.

---

A Redesigned Code Studio Editor

If you spend time in the code editor, it feels noticeably better now. We rebuilt the editor chrome around real, familiar file tabs:

• A tab per open file, the way you'd expect from a desktop editor — with keyboard navigation, middle-click to close, and close buttons on each tab
• Play, Stop, and Rebuild moved up to the toolbar, always in reach instead of buried in the preview
• A recent-projects switcher, so hopping between the games you're actively working on is a single click
• Your open tabs stick around between sessions, so you come back to exactly the files you left open

It's the same editor you know, with a lot less friction getting around it.

---

Chat That Knows How Full It Really Is

Long design-and-build conversations are where a lot of the good work happens — but every AI conversation has a size limit, and hitting it used to mean a send would just fail with a confusing error, no warning it was coming.

Not anymore. Code Studio chat gives you a graduated heads-up:

• As the chat starts to fill up, a gentle "Chat getting full soon" notice appears, while there's still plenty of room to wrap up your current thread
• When you're genuinely near the limit, you'll see a clear "Chat full — clear now" prompt so you can start a fresh conversation on your own terms, instead of typing a message that would just bounce
• Paste one giant message that won't fit? You get a specific "too large" nudge to shorten it, not a dead end

And it's now measuring the real thing

We shipped that warning with a rough estimate of conversation size, and the estimate ran high — enough that dense conversations were being cut off with real room still left. If you hit "chat full" earlier this month and thought surely not yet, you were right.

Chat now sizes your conversation using the AI model's own token counts rather than an approximation, so the warnings track reality. In practice that means noticeably more usable room in a conversation before anything blocks — the limit hasn't changed, but we're no longer measuring you out of space you actually had.

You'll also only ever see one context message now. Previously the orange "getting full" banner, a red limit overlay, and a blocked message box could all appear at once, each disagreeing about the state of things. There's a single, honest notice.

---

Fixes & Improvements

Fixes:

• Keyboard controls no longer crash a fresh game. Generated games could die on the very first frame if the AI wrote a key name a slightly different way than the engine expected — SPACE instead of Space, KeyW instead of W. The engine now accepts every reasonable spelling of a key, so the game just runs
• The AI writes engine code that works the first time. We rebuilt how the AI learns the game engine's API, so it reaches for the real function signatures up front instead of discovering them by hitting an error. Fewer broken first drafts, fewer rounds of "that didn't work"
• The GitHub integration no longer nags you with repeated permission warnings after a connection issue is already resolved, and syncing and pushing are faster
• In image generation, your chosen image count no longer resets when you switch presets, and the Ground Tileset preset now correctly sets a 1:1 aspect ratio
• Baked animations now anchor exactly where they should, so frames line up instead of drifting
• The asset drawer opens reliably, generated animations no longer show up as duplicates, and animations keep their identity when moved between tools
• Reloading a chat no longer brings back the AI's thinking from a message you'd already retried and replaced

Quality of life:

• Replies come back faster in long conversations. We stopped quietly rewriting earlier turns of your chat on every send — which was throwing away the AI provider's cache and making it re-read your whole conversation from scratch each time. Long chats now stay fast instead of getting slower as they grow
• Generation customization is always visible now — the dimension options are right there, with no extra toggle to click
• Hovering an aspect ratio shows a scaled preview rectangle with the estimated pixel size, and the full set of ratios fits without scrolling

---

What's Next

We'll keep pushing on all of it — smarter models, better animation, and tools that make a growing game easier to manage. Your feedback is what points us at the next thing, so keep it coming.

- The Makko Team

📧 Questions? Reach us at support@makko.ai 💬 Join the conversation on Discord