---
license: apache-2.0
tags:
- moss-image
- mossforge
- mlx
base_model:
- OpenMOSS-Team/MOSS-Image-SFT
- OpenMOSS-Team/MOSS-Image-Base
---

# MossForge model bundle

Everything [MossForge](https://huggingface.co/FiditeNemini) (a native macOS studio for
[MOSS-Image](https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT)) needs, laid out exactly as its
`models/` folder. MossForge's first-run setup reads `mossforge-manifest.json`, downloads the groups
you pick into place and checks every file's SHA-256. You can add more weights and LoRAs to `models/`
later, and MossForge picks them up.

This is a mirror of third-party weights, not new models. Every file keeps its original licence (all Apache-2.0).

| Group | Contents | Size | Default |
|---|---|---|---|
| `essentials` | Essentials | 15.5 GB | yes |
| `sft` | MOSS-Image SFT (bf16) | 10.2 GB | optional |
| `base` | MOSS-Image Base (bf16) | 10.2 GB | optional |
| `turbo-4step` | Turbo LoRA for SFT (akhaliq, 4-step) | 1.2 GB | optional |
| `turbo-8step` | 8-step LoRA for SFT (lodestones lora_latest) | 1.2 GB | optional |
| `turbo-base` | Turbo LoRA for Base (lodestones, 8-step) | 1.2 GB | optional |

## Layout

```
models/
├── diffusion_models/   DiTs (+ .mfmeta.json sidecars)
├── text_encoders/      qwen3.5_9b_int8_convrot.safetensors, qwen3.5-9b-tokenizer/
├── vae/                flux2-vae.safetensors
└── lora/<author>/      LoRAs with their supporting files and sidecars
```

## Provenance and licences

| File(s) | Source | Licence |
|---|---|---|
| `diffusion_models/moss_image_{sft,base}_bf16` | [Comfy-Org/MOSS-Image](https://huggingface.co/Comfy-Org/MOSS-Image) repack of [OpenMOSS-Team/MOSS-Image](https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT) | Apache-2.0 |
| `text_encoders/qwen3.5_9b_int8_convrot` | [Comfy-Org/MOSS-Image](https://huggingface.co/Comfy-Org/MOSS-Image) (Qwen3.5-9B, int8 ConvRot) | Apache-2.0 |
| `text_encoders/qwen3.5-9b-tokenizer/` | [Qwen/Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B) | Apache-2.0 (`LICENSE` included) |
| `vae/flux2-vae` | FLUX.2 VAE (`ae.safetensors`) | Apache-2.0 |
| `lora/lodestones/lodestones-8step-sft-latest` | [lodestones/MOSS-Image-SFT-8-steps-dmd](https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd) `lora_latest.safetensors` | Apache-2.0 |
| `lora/lodestones/lodestones-8step-base*` | [lodestones/MOSS-Image-Base-8-steps-dmd](https://huggingface.co/lodestones/MOSS-Image-Base-8-steps-dmd) | Apache-2.0 |
| `lora/akhaliq/akhaliq-4step-sft` | [akhaliq/MOSS-Image-SFT-4-steps](https://huggingface.co/akhaliq/MOSS-Image-SFT-4-steps) v2.0 | Apache-2.0 |
| `diffusion_models/moss_image_sft_akhaliq-4step-sft_int8_convrot*` | Made by MossForge: the akhaliq 4-step LoRA (v2.0) fused into SFT bf16 in fp32, then int8 ConvRot; runs on stock Euler, 4 steps, CFG 1 | Apache-2.0 |

MOSS-Image is by the OpenMOSS Team. This bundle is unofficial and not affiliated with or endorsed by OpenMOSS.