Model Management
Downloading, installing, and managing AI models
Last updated: 2026-07-15
Model Management
Overview
360 Hextile downloads AI models on demand from HuggingFace. The built-in model manager handles downloading, installing, updating, and removing models — no manual file management needed.
Official Model Catalog
Diffusion Models (14 available)
| Processor | Models | Size |
|---|---|---|
| SD 1.5 | Standard, specialized variants | 4-6 GB each |
| SD 2.1 | Standard, depth variant | 5-8 GB each |
| SDXL | Base, refiner, turbo | 7-12 GB each |
| SD 3.5 | Turbo (fast), Large (quality) | 12-16 GB each |
| Qwen-Image | GGUF Q4 (13GB), BF16 full (48GB) | 13-48 GB |
| TeleStyle | GGUF Q4 (12GB) | 12 GB |
LoRAs (15 included)
Pre-installed style and enhancement LoRAs. See the LoRA Guide for details.
IP-Adapters (2 available)
Image-guided generation for SD 1.5 and SDXL.
Utility — 360° Depth
The 360° Depth model (DA-2, ~1.4 GB) lives under Utility on the Pipelines page. It is not a diffusion pipeline — it estimates a true equirectangular depth map for any render node so you can reuse it for ControlNet, fog, and export. Install it like any other model, or accept the install prompt the first time you click Generate depth. Requires a GPU and a valid license. See Channels.
Installing Models
First-Run
The first-run wizard (Step 6 — Models) lets you select which models to download: - Recommended models are pre-selected - Choose based on your GPU VRAM - Downloads queue automatically
After First Run
- Open the Pipelines page (main navigation)
- Browse available models per processor
- Click Install to queue a download
- Download progress shown in real-time
- Models are ready immediately after download
Download Queue
- Downloads are serialized (one at a time) to maximize speed
- Priority install moves a model to the front of the queue
- Downloads resume automatically if interrupted
- Exponential backoff retry on failures
Gated Models (HuggingFace Authentication)
Some models require HuggingFace authentication: - SD 3.5 — requires accepting Stability AI license on HuggingFace - Enter your HuggingFace token in Settings → HuggingFace section - Token is stored securely and used for all gated downloads - Get your token at huggingface.co/settings/tokens
Managing Installed Models
- View installed models and disk usage per processor
- Remove unused models to free disk space
- Models are stored locally — no re-download needed unless deleted
- Switch between installed models instantly (within a processor type)
VRAM Note
Only ONE diffusion model loads into GPU memory at a time. When you switch from SDXL to SD 3.5 (for example), the SDXL model is automatically unloaded and SD 3.5 is loaded. This takes a few seconds but ensures your GPU memory is used efficiently.
See also: System Requirements for VRAM requirements | Pipelines for processor details