CloudDock Universal Usagi 4.0

Overview

Usagi 4.0 is our new “click-and-draw” container for Stable Diffusion. It replaces the Momonga 3.0 stack with a faster boot, a native Training Center, full DreamBooth with Kohya, a modern design, a one-tap Jupyter launcher, and an upgraded App Store.

What’s new vs Momonga 3.0

• Much faster launch: from 4:30 to 1:45 (≈65% quicker). Momonga required ~6 GB first-run dependencies; Usagi needs none.
• CloudDock Training Center: a native, purpose-built GUI for DreamBooth & LoRA—no third-party panels.
• Full DreamBooth with Kohya: first-class support and sensible presets out of the box.
• Modern look: frosted-glass cards, higher contrast, cleaner spacing.
• Jupyter Launcher: one-tap to JupyterLab; compute sessions come online in ~20 s.
• App Store 2.0: more stable and more complete; pulls resources directly from CloudDock servers so updates are painless.

Faster Launch

Usagi 4.0 removes the first-run dependency download and warms up the SD pipeline ahead of time. In practice you land on WebUI in ≈1:45 under typical loads, down from ≈4:30 with Momonga.

CloudDock Training Center

A native training UI designed for artists, not engineers. Create and manage DreamBooth and LoRA jobs, monitor VRAM, and resume runs without juggling scripts.

Training Center — clean panels for datasets, schedules, and checkpoints.

Job details — progress, ETA, logs, and VRAM hints.

DreamBooth with Kohya

Usagi ships with first-class DreamBooth support using Kohya. Sensible defaults help you reach “good-enough” character consistency without mystical flags.

• Dataset helpers: auto-split, caption checks, and quick sanity previews.
• Stable presets for portraits and full-body styles.
• Reproducibility: exports include params.json alongside results.

One-tap Jupyter

Open JupyterLab from the launcher and start coding in ≈20 s. The session inherits your container’s environment and access to models and inputs.

JupyterLab launcher — ≈20 s to ready.

CloudDock App Store 2.0

Install and update models, LoRAs, and tools directly from CloudDock servers. The new frosted-glass UI is resilient and avoids the crash scenarios seen in legacy 3.x panels.

Browse models, LoRAs, and tools with hash-verified provenance.

Progress with ETA and free-space hints. Updates are painless.

Design Refresh

Frosted-glass cards, a deep-blue canvas, better typography, and tighter spacing improve readability while keeping the SD aesthetic.

Quick Checks

nvidia-smi

ps aux | grep jupyter

Troubleshooting

• WebUI slow/blank: lower canvas resolution or browser zoom; try Chrome/Edge.
• Training stalls: check dataset captions and VRAM headroom; resume from last good step.
• App install fails: retry from App Store 2.0; verify free space and network.