Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
Commits on main
Studio: bake the blurred chat wallpaper once instead of re-running the blur on every scroll frame (#17253)
81b7720 Daniel Han committed 11 days ago
[Feature] Unsloth Studio / Desktop: let the mic send audio straight to any audio-input model (#16976)
c21ee57 Leo Borcherding committed 11 days ago
Condense comments added since the last cleanup: core, CLI and tests (#16111)
2478e74 Daniel Han committed 11 days ago
Condense comments added since the last cleanup: Studio backend (#16110)
f1a274d Daniel Han committed 11 days ago
Studio: pause Glass blur through bursts of keys and presses under software compositing (#17237)
e36630f Daniel Han committed 11 days ago
Studio video: run LTX-2.5 FP8 on a 24 GB card by streaming the hosted fp8 DiT instead of refusing it (#17238)
1c15fc3 Daniel Han committed 11 days ago
Add configurable voice context to reduce Qwen3-ASR memory (#15810)
9092900 Lee Jackson committed 11 days ago
Studio MiniMax-H3: keep the released bf16 denoiser resident beside a streamed conditioner on ~100 GB cards (#16877)
ed3eec6 Daniel Han committed 11 days ago
Studio update: stop size-checking the managed Windows launcher (#16901)
6c89d20 Debojyoti Banerjee committed 11 days ago
Condense comments added since the last cleanup: installers, scripts and CI (#16112)
54a932c Daniel Han committed 11 days ago
Condense comments added since the last cleanup: Studio frontend and Tauri (#16109)
1641ea9 Daniel Han committed 11 days ago
Studio LTX-2.5: use two CUDA cards, the sidecar text encoder priced on its own card (#17077)
eff80d8 Daniel Han committed 11 days ago
Studio chat: only a note followed by its wrapper starts an attachment (#16915)
696f1c0 Yakshith committed 11 days ago
Studio diffusion: price a streamed Qwen-Image-2.1 edit's reference from its token count, not the flat allowance (A100 8 GiB edit -15%, ahead of the baseline) (#17233)
8efcf85 Daniel Han committed 11 days ago
Studio: fuse the int8 ConvRot rotation into its producers on FLUX.2 / FLUX.3 (ConvRot cost +8.7% to +0.5%) (#17230)
f4d17f4 Daniel Han committed 11 days ago
Studio: keep single-file int8 denoiser weights as int8 on CPU instead of expanding them to float32 (#17178)
1489693 Daniel Han committed 11 days ago
Studio: load INT8 instead of NVFP4 on pre-Blackwell GPUs (#17229)
192e492 Daniel Han committed 11 days ago
Studio LTX-2.5: compact conditioning transport and stage-two connector reuse (#17232)
75d0307 Daniel Han committed 11 days ago
Studio RAG: pause linked-folder indexing while a local chat generates (#17018)
31353df Daniel Han committed 11 days ago
Studio: save Image LoRA training settings as presets (#17019)
763347f Daniel Han committed 11 days ago
Studio: stop the Windows Terminal from creating nul files through Git Bash (#17010)
cdf27f4 Daniel Han committed 11 days ago
Studio: run FP8 on T4 and other Turing GPUs as weight-only fp8 instead of refusing it (#17218)
bf9039d Daniel Han committed 11 days ago
Studio: hold mid-scroll code block work on WebKit, fixing the #16865 scroll regression (#17149)
ae176c8 Daniel Han committed 11 days ago
Studio: open Settings and the shortcuts list with one panel render and one focus move (#17042)
c2a5595 Daniel Han committed 11 days ago
Studio: stop the autoscroll follow flag restyling the whole thread in WebKit (#17035)
78429b3 Daniel Han committed 11 days ago
Studio: drop the live Glass blur while anything moves, on software-composited pages (#17007)
390a309 Daniel Han committed 11 days ago
Studio diffusion: keep Qwen-Image-2.1 scoped LoRA prompt reuse after the no-K/V wrapper, and key callable helpers fully (#17192)
864c246 Daniel Han committed 11 days ago
Studio FLUX.3: pass cuDNN attention the projections' strided Q/K/V instead of contiguous copies (#17180)
dac1edb Daniel Han committed 11 days ago
RoPE: no recompile per sequence length when a compiled backward hands over a seq-major gradient (#17212)
56a96ad Daniel Han committed 11 days ago