Nimbus 1.1.7 is about making the app feel less fussy when you install models, generate images, or move between iPhone, iPad, and Mac. The theme is simple: fewer mystery states, better default choices, and clearer recovery when a large download is interrupted.
Install once, recover more easily
Large local models are not tiny app assets. They can be several gigabytes, and real life interrupts downloads: the app closes, the network changes, storage gets tight, or iOS decides another process needs attention. In 1.1.7, Nimbus is more careful about deciding when a model is actually ready.
If a Mirage model is missing setup files, Nimbus now says so earlier. If the main model file is already on device, Retry can focus on the missing pieces instead of making you start from zero. Verify Install also has a clearer job: check what is present, repair what can be repaired, and give you a readable explanation when something still needs attention.
Mirage picks safer defaults
Mirage now recommends smaller, device-friendly image models first on iPhone and iPad, including palettized Core ML Stable Diffusion 2.1 and 1.5 options. SDXL remains available when you want it, but it no longer has to be the first thing a new user sees on a phone.
On Mac, where memory and storage are usually more forgiving, Mirage keeps SDXL near the top of the recommendation set. The goal is not to hide power. It is to put the model that is most likely to work well on your hardware in front of you first.
How to use the new recovery flow
- Open Models, then choose the model you were installing.
- If Nimbus shows Retry, run it before deleting the model. Retry can often finish missing files.
- If the model looks installed but Mirage still says it is preparing, choose Verify Install.
- If Verify Install reports a missing file that cannot be repaired, remove the model and start the install again on Wi-Fi.
Those steps are intentionally ordinary. You should not need to know what a scheduler, bundle plan, or runtime manifest is to recover a local model.
Voice and Ashe feel steadier
Overture and Nimbus Quick now show clearer "getting ready" states while speech models load. If your mounted transcription model is not ready yet, Apple Speech can step in as a fallback where available. That means voice input is less likely to feel like it disappeared into a loading screen.
Ashe also gets a lighter work queue and cleaner handoff behavior. You should see less clutter when asking Ashe to track work, run common actions, or move between iPhone and Mac.
What you should notice
The biggest improvement is confidence. Downloads are quieter. Duplicate banners are reduced. Live Activities and progress states are clearer. Model choices explain themselves better. Mirage is less likely to get stuck saying "Preparing model" with no useful next step.
If you use Nimbus daily, 1.1.7 should feel like the same app with less friction around the moments that used to require patience.
Tagged Release · Published Jun 28, 2026 · Read the 1.1.7 release notes · Back to all posts