MOSS-Image | Forge Neo Extension

Generate with MOSS-Image inside Forge Neo, using the checkpoint, preset and Generate button you already know: text-to-image with sharp text in English and Chinese, image-to-image, and layouts placed with boxes in the prompt.

Requires an up-to-date Forge Neo (the neo branch). On an older version it stays disabled and says so in the console, and Forge Neo keeps working as usual.

Features 🚀

🖼️ Native Generation

• Runs inside txt2img and img2img: no separate tab, no extra environment

• The SFT and Base models with their Qwen3.5 9B text encoder, through the VAE / Text Encoder selector you already use

• Uses the same autoencoder as FLUX.2 and FLUX 3, so if you have flux2-vae you already have it

• About a minute per 1024x1024 image at 50 steps on an A40 with the INT8 files

📝 Text and Posters

• Titles, captions and signs come out as written, in English and in Chinese: put the exact words in quotes

• Any size that is a multiple of 16, such as 832x1248 for a poster

📐 Layout by Boxes

• Write the prompt as a JSON layout and tell the model where each thing goes, with a bbox per item, in the format of the MOSS-Image user guide

• Works with the Base and the SFT model

🔁 Image-to-Image

• Put a picture in img2img and describe the result: around 0.6 denoising keeps the scene and follows the details you change

⚡ 8-Step LoRA

• lodestones' 8-step LoRA works with the SFT model: pictures in 8 steps instead of 50, with no negative prompt pass

🧭 Upstream's Guidance

• The negative pass skips block 22 as the official pipeline does, with a setting to change it

🛡️ Safe by Design

• Changes no Forge Neo file: remove the extension and everything is as it was

• Checks your setup at startup and tells you which file is missing instead of failing with a generic error

• Other models are not affected

Recommended Settings ⚙️

• Euler, schedule type MOSS, 50 steps, CFG 3. The moss UI preset sets these values

• Leave the negative prompt empty with the SFT model. Use sides that are a multiple of 16

Installation 📦

1. Open Forge Neo WebUI and go to Extensions → Install from URL

2. Paste: https://github.com/eduardoabreu81/moss-image-forge-neo

3. Click Install and restart the WebUI

4. From Comfy-Org/MOSS-Image, place the INT8 checkpoint (SFT or Base), the Qwen3.5 9B text encoder and flux2-vae in the usual folders, then restart

5. Pick the moss UI preset, the checkpoint, and the text encoder and autoencoder under VAE / Text Encoder

Documentation: File list · What's new · GitHub · License: AGPL-3.0. The MOSS-Image weights are released by the OpenMOSS Team under Apache-2.0 and are not included.