What is MOSS Image?
MOSS Image is OpenMOSS’s text-to-image model family. It can turn written prompts into illustrations, product visuals and posters, with particular attention to English and Chinese lettering and the arrangement of objects and text.
The 5B image backbone accepts natural language, tags and structured JSON layouts. Its documented prompt context is up to 4,096 tokens; this describes input length, rather than 4K image resolution. Read the OpenMOSS project overview for the underlying model details.
Our online generator runs MOSS-Image-SFT through fal.ai. You can create an image from a text prompt without downloading weights or setting up a local GPU.
Try MOSS Image onlineMOSS-Image-Base vs MOSS-Image-SFT
Base and SFT are separate official checkpoints. SFT is fine-tuned from Base using curated data. Choose the checkpoint your workflow expects, and keep its sampling configuration with it.
| Model | Role | Where to use it |
|---|---|---|
| MOSS-Image-Base | Foundation checkpoint for generation and further fine-tuning. | Local tools and workflows that explicitly select Base. |
| MOSS-Image-SFT | Curated-data fine-tune of Base. | This site’s fal-powered generator, or a local SFT workflow. |
Both have official model cards on Hugging Face: MOSS-Image-Base and MOSS-Image-SFT. The released weights are labeled Apache-2.0; consult the relevant license and service terms for your use.
What is Kromoss?
Kromoss is a community model by lodestones, fully fine-tuned from MOSS-Image-Base at 1024 px. Its model card describes compatibility with the MOSS architecture, text encoder and VAE, and provides ComfyUI-format BF16 and INT8 checkpoints.
The same repository offers DMD and DMD2 8-step LoRAs distilled from Base. These are optional local workflow components; they do not turn this site’s SFT endpoint into a Kromoss or Turbo generator.
Start with the Kromoss model card and files, then follow the Kromoss ComfyUI setup for the distinct checkpoint and LoRA settings.
MOSS Image GitHub and Hugging Face resources
Use the upstream source for code and the model publisher’s page for weights. These links serve different purposes:
- OpenMOSS/MOSS-Image on GitHub — source code, documentation and release notes.
- OpenMOSS-Team/MOSS-Image-Base — original Base weights and model card.
- OpenMOSS-Team/MOSS-Image-SFT — original SFT weights and model card.
- Comfy-Org/MOSS-Image — repackaged single-file weights and native workflow links.
- lodestones/Kromoss — the community fine-tune and its optional acceleration LoRAs.
This website is an independent service. OpenMOSS publishes the original model family; lodestones publishes Kromoss. Our prompt library credits its community sources separately.
Run MOSS Image online or in ComfyUI
For an image now: open the MOSS Image generator, write a prompt, choose a shape and generate. Sign-in is required. New accounts receive two welcome credits once, enough for one image; subsequent images use two credits each. See pricing and credit options.
For local control: use the ComfyUI workflow guide. You manage the model files, hardware, precision and sampling. The online service and local workflows can produce different results even with similar prompts.
To begin a prompt, describe the subject, composition, lighting and exact visible words. For example:
An illustrated travel poster for Lisbon. A yellow tram climbs a steep street in warm afternoon light. Large headline "LISBOA" at the top, cream paper, terracotta and teal palette, generous margins.Common questions about the MOSS Image model
Is Kromoss the same as MOSS-Image-SFT?
They are different fine-tunes in the same model ecosystem. This site runs SFT; Kromoss is available from lodestones for compatible local workflows.
Can I edit an existing image here?
This workspace provides text-to-image generation. It does not accept reference-photo uploads or offer photo editing.
Is there a MOSS Image Turbo model?
The Kromoss repository documents 8-step acceleration LoRAs for local use. That is separate from this website’s fal MOSS Image endpoint, which serves SFT. We do not offer a Turbo model selector.
Do I need to install ComfyUI?
Only for the local route. The online generator works in your browser, and the prompt library can help you write a first brief.