Models#
The Models tab (5) is the model manager: what is installed, how to add more, and the settings each model carries. It serves every studio; the video model packs have their own page inside it, described in Presets.
Installing models#
The launchpad on the tab offers three manual routes and the starter catalogue:
| Route | Use it for |
|---|---|
| URL or local path | A single file or folder: a .safetensors checkpoint, a LoRA, a diffusers folder, a GGUF. Paste a download URL or a path on this machine. In-place install keeps the file where it is instead of copying it under the models root. |
| HuggingFace | Paste owner/model-name; the tab lists the files in the repository and installs the one you pick. Gated repositories need a HuggingFace token, saved once in the tab. |
| Scan Folder | Point at a folder you already have (another tool's model library); every recognised model is listed with an install button, in place or copied. |
| Starter models | The curated catalogue: quick-start bundles per family (SD 1.5, SDXL, FLUX.1 dev, FLUX.2 Klein, Z-Image Turbo) and every model in the list, with size and license shown before download. |
Installs run in the install queue in the background; the tab shows progress and errors. Maystro reads the file header to identify the architecture and type; when it cannot, the model appears with a warning and you set the base model and type by hand in its panel.
What is in the catalogue#
- Main models for each family, in bf16 and quantised (FP8, NF4, GGUF) variants; the encoders and VAEs they need are installed with them.
- ControlNets and Control LoRAs, IP-Adapters (standard, precise, face), the Z-Image ControlNet Union and Tile.
- Upscale models (RealESRGAN, ESRGAN, SwinIR, AnimeSharp), embeddings, example LoRAs and style models.
- Vision and text LLMs, speech models and voices used by the Prompt Builder's Fill from Image, the voice assistant and the video translation button.
- The video model packs are installed from the video Models page.
Krea 2 is not in the catalogue: bring the weights yourself with the URL or path route. FP8 or int8 variants are the right choice on a 16 GB card.
Managing a model#
Select a model to open its panel: name, description, base model, type, format and path; the default settings (steps, CFG or guidance, scheduler, size, VAE) applied when you select the model on the Generate tab; a cover image; trigger phrases for LoRAs. From here you can convert a checkpoint to diffusers or GGUF, update the path, or delete the model (the file is removed unless it was installed in place).
Sync Models compares the models folder with the database and lists orphaned folders that can be deleted safely.
Where files live#
All copied installs land under the models root, organised by base model and type. In-place installs and scanned folders stay where they are and are referenced by path. Roots, folder layout and the licensing of weights are covered in Models root and your own models.