Unofficial guide. Cicadas is independent of OpenMOSS and does not run MOSS Image.
MOSS Image: online demo and ComfyUI
MOSS Image is OpenMOSS’s text-to-image family, with Base and SFT checkpoints. It accepts ordinary descriptions or a structured layout that places text and objects. Start with the official demo below, or use the ComfyUI instructions for a local workflow.
Base or SFT
Base is the starting checkpoint; SFT was tuned on curated examples. The official side-by-side is a useful place to inspect text grouping before choosing.
Text and layout
The project describes a 5B backbone and layout prompts that specify objects, text and bounding boxes. Those are developer capabilities, not a Cicadas quality test.
Choose where to run it
Use the project’s demo for a first look, ComfyUI for a local workflow, or follow its API link. Each destination has its own access and compute conditions.
How to try MOSS Image
- Open the official demo and inspect its current availability before uploading anything.
- For local use, choose a checkpoint and follow the official ComfyUI or Python instructions for your hardware.
- Keep the checkpoint and settings with the result. Inspect spelling, object placement and visual artifacts before using it.
MOSS Image Base vs SFT
Both checkpoints share the model family and inference interface. SFT is a version fine-tuned on curated data. The project shows a Base/SFT comparison; we have not run an independent comparison, so this page does not rank them.
Read the model overview ↗Using MOSS Image in ComfyUI
The current README recommends native MOSS support in an updated ComfyUI build with ComfyUI-format weights. Its optional repository loader remains available for local Hugging Face directories and older plugin workflows. Follow the native instructions first rather than installing a plugin by default.
Open the official ComfyUI instructions ↗Layout prompts need the right coordinate format
MOSS layout JSON describes the canvas and each text or object region. The project example uses y_min, x_min, y_max, x_max coordinates relative to the canvas center. Use its example with matching image dimensions; do not paste bounding boxes from another model without converting their format.
See the layout example and coordinates ↗Weights, code and hosted use have different conditions
The project labels MOSS Image weights Apache-2.0 and notes that pipeline code and third-party components retain their own licenses. Read the files for the checkpoint and components you use. A downloadable model does not mean a hosted demo or API has unlimited free compute.
Check the model and license sources ↗MOSS Image questions
Can I generate a MOSS Image picture on Cicadas?
No. This page explains the model and links to the developer’s tools. It does not upload your prompt or photo to an image provider, and opening this guide uses no Cicadas credits.
Where can I try MOSS Image online?
Use the official OpenMOSS Hugging Face Space linked above. Check its current queue, access rules and data handling there; this guide does not promise that the demo is always available or free.
Does MOSS Image make video?
The model discussed here generates images. To animate an image afterward, use a separate image-to-video tool; that is a different model and a separately priced job.
Do I need a ComfyUI plugin?
The current official README describes native ComfyUI support. The repository’s loader plugin is an alternative for specific local-directory or existing plugin workflows, not a requirement for every user.
Is this the official MOSS Image website?
No. OpenMOSS maintains the model and the official links above. Cicadas publishes this independent guide and does not claim affiliation or endorsement.
From an image to a video
By the Cicadas team · sources checked 2026-10-21 · · Testing and sources