Releases: avlp12/alis-studio
Alis Studio v0.5.4
A local, on-device image-generation studio for Apple silicon (MLX). Self-contained — double-click to run, no Python or pip to set up.
What's new
- Gallery: every image you generate is now saved and shown in a new Gallery view (top-bar toggle) — browse your past work, reuse a prompt, download, or delete. (History starts from this version.)
- Scroll fix: the window no longer rubber-bands into blank space when scrolling — the header is now fixed and only the content scrolls.
Install
- Download
Alis-Studio-0.5.4.dmg, open it, drag Alis Studio to Applications. - First launch (Gatekeeper): ad-hoc signed, so macOS blocks it once — right-click → Open (confirm), or run
xattr -dr com.apple.quarantine "/Applications/Alis Studio.app". - First image downloads the model weights from Hugging Face.
Requires an Apple-silicon Mac (M1+) with ≥ 24 GB unified memory.
Features
- Self-contained app bundling its own Python + all deps. 4 models: Krea 2 Turbo · Qwen-Image · FLUX.1 schnell / dev.
- Stop button; optional on-device prompt enhancer (Qwen3-4B, translates non-English prompts, memory-gated, off by default).
- Pine brand icon generated by Alis Studio itself.
⚠️ Ad-hoc signed (use the Gatekeeper step). Model weights download on first generation.
Alis Studio v0.5.3
A local, on-device image-generation studio for Apple silicon (MLX). Self-contained — double-click to run, no Python or pip to set up.
What's new
- New brand icon: a pine tree + a generative sparkle + a STUDIO wordmark on the clay squircle — generated by Alis Studio itself. The in-app logo now matches it, sharing one identity with the project's pine-tree profile imagery.
Install
- Download
Alis-Studio-0.5.3.dmg, open it, drag Alis Studio to Applications. - First launch (Gatekeeper): this build is ad-hoc signed, so macOS blocks it once — right-click → Open (then confirm), or run
xattr -dr com.apple.quarantine "/Applications/Alis Studio.app". - The first image downloads the model weights from Hugging Face.
Requires an Apple-silicon Mac (M1+) with ≥ 24 GB unified memory.
Features
- Self-contained app bundling its own Python + all deps. 4 models: Krea 2 Turbo · Qwen-Image · FLUX.1 schnell / dev.
- Stop button; optional on-device prompt enhancer (Qwen3-4B; translates non-English prompts, memory-gated, off by default).
- Includes the v0.5.2 fix for the "There is no Stream(gpu, 0)" generation crash.
⚠️ Ad-hoc signed (use the Gatekeeper step). Model weights download on first generation.
Alis Studio v0.5.2
A local, on-device image-generation studio for Apple silicon (MLX). Self-contained — double-click to run, no Python or pip to set up.
Fixes a critical bug: earlier 0.5.x builds crashed image generation with "There is no Stream(gpu, 0) in current thread" on recent MLX. Please use 0.5.2.
Install
- Download
Alis-Studio-0.5.2.dmgbelow, open it, and drag Alis Studio to Applications. - First launch (Gatekeeper): this build is ad-hoc signed, not notarized, so macOS blocks it the first time. Do one of:
- Right-click the app → Open, then confirm, or
- run once in Terminal:
xattr -dr com.apple.quarantine "/Applications/Alis Studio.app"
- Launch it. The first image downloads the model weights from Hugging Face (several GB).
Requires an Apple-silicon Mac (M1+) with ≥ 24 GB unified memory.
Highlights
- Self-contained app: bundles its own Python + every dependency (mlx, mflux, transformers, the Krea 2 Turbo backend, …).
- 4 models: Krea 2 Turbo (default, pure-MLX) · Qwen-Image · FLUX.1 schnell / dev.
- Stop button to cancel a generation.
- Optional on-device prompt enhancer (Settings → Prompt enhancer): a local LLM (Qwen3-4B) that translates non-English prompts — e.g. Korean — to English and enriches them. Off by default; memory-gated (asks before loading on ≤24 GB Macs).
- App icon generated by Alis Studio itself.
Fixed in 0.5.2
- All MLX work now runs on one dedicated GPU thread, and the prompt enhancer runs in its own process — resolving the "There is no Stream(gpu, 0)" crash that broke generation in 0.5.0–0.5.1.
⚠️ Ad-hoc signed (use the Gatekeeper step above). Model weights are not bundled — they download on first generation.