Open cards · Read 2026-11-03 · Not either model
MOSS-Image vs Krea 2: license, size, and style controls
MOSS-Image vs Krea 2 is a comparison search. The longer MOSS-Image guide stays on /moss-image, and the longer Krea 2 notes stay on /krea-2. This page only lines up the license, the size, the style controls, and the step counts printed on the public cards read on 2026-11-03.
This generator runs Qwen Image 2.1 for 8 credits at 1K. It is not MOSS-Image and not Krea 2. Flux 2 and Nano Banana 2.1 are also 8 credits, so this is not a cheaper substitute.
64 free credits = 8 standard images. No card required.
Output board
Your generated variations will appear here.
- Made by
- OpenMOSS
- Released
- October 16, 2026
- Run on Nano AI
- Qwen Image 2.1 · 8 credits per image
- Page updated
- November 3, 2026
Source: huggingface.co
MOSS-Image vs Krea 2 from the cards opened today
The MOSS-Image-Base card on Hugging Face, read on 2026-11-03, says the weights and bundled inference code are released under the Apache License, Version 2.0. The model-size line says 5B params. The card says it supports natural-language prompts, tag sequences, and JSON layouts. The printed sampling default is 50 steps and CFG 3. The GitHub README read the same day says the standalone pipeline code is under the Apache License 2.0, names a 5B backbone, and says prompts can run up to 4K tokens. The introduction line on that README is dated 2026-10-16. Neither of those two pages describes a style-reference upload or a moodboard control.
krea.ai/krea-2, read on 2026-11-03, is dated May 12, 2026 and says Krea 2 is an image model built for aesthetics, with style references and moodboards. The page says Krea 2 is now open source, and it says K2 generates outputs in 15 seconds or less. That timing is Krea's claim. This site has not measured it. The Hugging Face card for krea/Krea-2-Raw, last updated June 22, 2026, says the architecture is a diffusion transformer with 12 billion parameters. The same card's model-size line says 13B params. Both figures are left as printed. The license on that card is the Krea 2 Community License, which is not the Apache License, Version 2.0. The card says Raw is not recommended for inference and points at raw.safetensors. Its printed text-to-image example uses 52 steps, cfg 3.5, and 1024. Those commands are text prompts. They do not show a style-reference argument. Style references and moodboards are described on the product page, not in that command block.
This page does not repeat the download tutorial on /moss-image or the hosted-variant notes on /krea-2. No benchmark table was copied, because neither card opened today printed a score this page is willing to treat as a measured comparison. The generator is Qwen Image 2.1, the same stand-in as /moss-image, at 8 credits and 1K. /krea-2 runs Flux 2, also 8 credits and 1K. Nano Banana 2.1 is 8 credits and 1K as well. Eligible new accounts get 64 credits, which is 8 images of any of those three. None of the three is a cheaper substitute for the others.
MOSS-Image vs Krea 2 at a glance
- Read
- 2026-11-03
- MOSS license
- Apache License, Version 2.0, on the Base card
- MOSS size
- 5B params on the Base card. 5B backbone on the GitHub README
- MOSS controls on those pages
- Text, tags, JSON layout. Up to 4K tokens. No style-reference field described
- MOSS steps they print
- 50 steps, CFG 3
- MOSS date they print
- 2026-10-16 on the GitHub introduction line
- Krea product page
- Style references and moodboards. Announced 2026-05-12. Says it is now open source
- Krea card
- 12 billion parameters, and a 13B model-size line. June 22, 2026
- Krea license
- Krea 2 Community License. Not Apache 2.0
- Krea steps they print
- 52 steps, cfg 3.5, 1024, file raw.safetensors
- Generator here
- Qwen Image 2.1, 8 credits, 1K. Same price as Flux 2 and Nano Banana 2.1
Four differences that are on the pages
The licenses are not the same
MOSS-Image-Base says Apache License, Version 2.0. The Krea 2 Raw card says Krea 2 Community License. The product page's 'open source' line does not rename that license.
The sizes they print are not the same
MOSS prints 5B. Krea's overview prints 12 billion parameters, and the card header prints 13B. This page does not average them.
Style references are a Krea product-page feature
krea.ai/krea-2 describes style references and moodboards. The MOSS card and README describe text, tags, and JSON layout. The Raw inference commands that were printed are text prompts at 52 steps.
This button is not a cheaper third model
Qwen Image 2.1, Flux 2, and Nano Banana 2.1 are each 8 credits at 1K. The button runs Qwen Image 2.1 because this page hangs off /moss-image.
MOSS-Image, Krea 2, and this page
Columns follow the public cards and krea.ai/krea-2 as read on 2026-11-03. No third-party benchmark is included. The 15 seconds line is Krea's claim.
| MOSS-Image | Krea 2 | This page | |
|---|---|---|---|
| License they print | Apache License, Version 2.0 | Krea 2 Community License | Not those weights |
| Size they print | 5B | 12 billion, and a 13B line | Not stated for the stand-in |
| Style control they print | JSON layout. No moodboard on the card | style references and moodboards | Up to 5 reference images |
| Steps they print | 50 steps, CFG 3 | 52 steps, cfg 3.5, 1024 | 1K output |
| Date they print | 2026-10-16 | 2026-05-12, card June 22, 2026 | Read 2026-11-03 |
| Price here | Not hosted | Not hosted. Their 15 seconds is their claim | Qwen Image 2.1 is 8 credits |
How to try the priced Qwen Image 2.1 endpoint
- 1
Start from a prompt on this page
Pick one still life. Change the object so it is yours. Do not add a person. This does not download MOSS-Image or raw.safetensors.
- 2
Leave Qwen Image 2.1 selected
Qwen Image 2.1 costs 8 credits at 1K. Flux 2 on /krea-2 is also 8 credits at 1K. Neither one is OpenMOSS-Team/MOSS-Image-Base.
- 3
Generate and keep the file
Sign in. Eligible new accounts get 64 credits, which covers 8 images. The file is not a MOSS-Image or Krea 2 output.
Stills for the Qwen Image 2.1 trial
Original prompts. They are not the sample prompts printed on the MOSS or Krea cards.
Bowl
A white ceramic bowl of oranges on a pale table, one small card beside it with the word ORANGE in plain letters, no people, no logos.
Window
An empty shop window at dusk with one unbranded poster of a mountain, no people, no logos, no other readable text.
Bottles
Three unlabeled glass bottles on a wooden tray, soft side light, no people, no logos, no readable text.
Map
A folded paper map and a blank compass on a desk, overhead light, no people, no logos, no readable text.
FAQ
MOSS-Image vs Krea 2 FAQ
Short answers from the cards read on 2026-11-03. The install guides stay on the other two pages.
Which model does this button run?
Qwen Image 2.1, for 8 credits at 1K. It is not MOSS-Image and not Krea 2. Flux 2 and Nano Banana 2.1 are also 8 credits, so this is not a cheaper substitute.
What license does each card print?
MOSS-Image-Base says Apache License, Version 2.0 for the weights and bundled inference code. krea/Krea-2-Raw says Krea 2 Community License. The product page also says Krea 2 is now open source. That sentence is not a second license name.
How big do they say they are?
The MOSS card says 5B params. The Krea overview says 12 billion parameters, and the model-size line says 13B. This page does not pick one Krea figure or run a benchmark.
Do both accept style references?
krea.ai/krea-2 describes style references and moodboards. The MOSS card describes natural language, tag sequences, and JSON layouts, up to 4K tokens on the README. The Raw card's printed commands are text prompts at 52 steps, cfg 3.5, and 1024.
Where are the longer guides?
Download and sampling notes for MOSS-Image stay on /moss-image. Krea 2's own page is /krea-2, and that page's generator is Flux 2 at the same 8 credits. This page does not copy those tutorials.
Is Nano AI affiliated with OpenMOSS or Krea?
No. Nano AI is not affiliated with OpenMOSS or Krea. The generator runs Qwen Image 2.1. The weights id OpenMOSS-Team/MOSS-Image-Base is not hosted here, and raw.safetensors is not hosted here.
More from Nano AI
Related pages
Nano AI is an independent service and is not affiliated with OpenMOSS or Krea. Card facts are from huggingface.co/OpenMOSS-Team/MOSS-Image-Base, the OpenMOSS GitHub README, krea.ai/krea-2, and huggingface.co/krea/Krea-2-Raw as read on 2026-11-03. The generator runs Qwen Image 2.1 at 1K.