Sources read 2026-11-05
MOSS Image Base vs SFT · Guide
A short checkpoint comparison: what changes from Base to SFT, where to get the official files, and where IDM fits. Keep the full installation guide on the existing MOSS page.
Here the generator is Qwen Image 2.1: 8 credits per 1K image, as on /moss-image. It runs neither MOSS-Image-Base nor MOSS-Image-SFT; IDM adapters are not loaded.
64 free credits = 8 standard images. No card required.
Output board
Your generated variations will appear here.
- Made by
- OpenMOSS
- Run on Nano AI
- Qwen Image 2.1 · 8 credits per image
- Page updated
- November 5, 2026
Source: huggingface.co
Published source example
The official Base and SFT whiteboard pair, in source order. No claim of matched seed or a Nano AI controlled comparison. Read 2026-11-05. Not generated by Nano AI; not a measured result from the generator above.
MOSS Image Base vs SFT: What the sources establish
Sources read 2026-11-05. OpenMOSS describes SFT as supervised fine-tuning of Base, using the same architecture and pipeline. In the published whiteboard example, the author reports clearer text grouping and layout with SFT. That example is not evidence that SFT always wins.
A short checkpoint comparison: what changes from Base to SFT, where to get the official files, and where IDM fits. Keep the full installation guide on the existing MOSS page.
At a glance
- Source read
- 2026-11-05
- Original project
- OpenMOSS
- Generator here
- Qwen Image 2.1
- Output here
- 1K · 8 credits per image
Choose the right workflow
Original project
OpenMOSS describes SFT as supervised fine-tuning of Base, using the same architecture and pipeline. In the published whiteboard example, the author reports clearer text grouping and layout with SFT. That example is not evidence that SFT always wins.
Try here
Here the generator is Qwen Image 2.1: 8 credits per 1K image, as on /moss-image. It runs neither MOSS-Image-Base nor MOSS-Image-SFT; IDM adapters are not loaded.
Published example
The official Base and SFT whiteboard pair, in source order. No claim of matched seed or a Nano AI controlled comparison.
MOSS Image Base vs SFT: checkpoint differences
Source: OpenMOSS SFT model card, read 2026-11-05. No universal ranking or matched-seed test is claimed.
| MOSS-Image-Base | MOSS-Image-SFT | |
|---|---|---|
| Training stage | Base checkpoint | Supervised fine-tuning of Base |
| Architecture / pipeline | MOSS-Image | Same as Base |
| Published whiteboard pair | Official Base example above | Author reports clearer text grouping and layout in this example |
Create an original result
- 1
Prepare your idea
Choose an original prompt below. For an edit, upload only your own authorized reference photo.
- 2
Check the model and credits
The model shown above is the service used here; the source project’s features do not transfer to it.
- 3
Generate and review
Sign in, submit and inspect the result. Published examples are references, not a promise to reproduce them.
Prompts to adapt
Original templates to adapt. Published examples above are separate source material.
Studio
An original botanical exhibition poster titled "GARDEN NOTES", a green leaf illustration, cream paper, generous margins, no people, no logos.
Detail
A simple menu titled "TEA BREAK", two lines "Green tea $3" and "Oat latte $5", a small teacup drawing, no people, no logos.
Setting
An original gardening book cover titled "SMALL GARDENS", three terracotta pots, cream and green palette, no people, no logos.
Variation
An original gallery card titled "PAPER & LIGHT", a folded paper sculpture, pale blue background, clean typography, no people, no logos.
FAQ
MOSS Image Base vs SFT · Questions before you start
Source facts and the separate hosted workflow.
Which model runs here?
Here the generator is Qwen Image 2.1: 8 credits per 1K image, as on /moss-image. It runs neither MOSS-Image-Base nor MOSS-Image-SFT; IDM adapters are not loaded.
What does the original project do?
OpenMOSS describes SFT as supervised fine-tuning of Base, using the same architecture and pipeline. In the published whiteboard example, the author reports clearer text grouping and layout with SFT. That example is not evidence that SFT always wins.
Did Nano AI generate this example?
The official Base and SFT whiteboard pair, in source order. No claim of matched seed or a Nano AI controlled comparison. Read 2026-11-05. Not generated by Nano AI; not a measured result from the generator above.
Are speed or quality claims tested here?
No local GPU or third-party inference benchmark was run. A static source image cannot establish motion, audio synchronization or an end-to-end speedup.
Where should I go next?
Use the related pages below for the existing model overview or another workflow.
More from Nano AI
Related guides
Nano AI is independent of the named projects and vendors. Public sources were read on 2026-11-05. Published media belongs to its source; this page does not add credits to third-party accounts.