MOSS Image Review: Polish Test, Pricing and License
MOSS Image scored 8.5/10 in our Polish-language test. See how it handled Chinese and Polish text, how fast it was and what it costs on fal.ai.
👁 117 reads
MOSS Image is an open-weight image model from the OpenMOSS team, released on October 16, 2026, under the Apache-2.0 license. It is designed to handle Chinese and English text in images. It arrived on fal.ai two days later. On October 20, I tested it in Polish: it scored 8.5/10 across 10 prompts, matching FLUX 3 Klein. It produced the clearest Chinese neon sign of all the models in the test, but rendered a Polish milk-bar menu with [corrupted lettering in the original Polish test].
At a glance (October 20, 2026)
- Creator: OpenMOSS Team. MOSS-Image-Base and MOSS-Image-SFT weights are on Hugging Face under the Apache-2.0 license, which allows commercial use.
- Strength: Text in images, especially Chinese and English; supports natural-language prompts, tags or JSON layouts.
- On fal.ai: $0.05 per megapixel, with image sides from 512 to 2,048 px and 50 steps by default.
- My result: 8.5/10 across 10 Polish-language prompts, about $0.06 and 53 seconds per image (median).
How much does MOSS Image cost?
On fal.ai, you pay $0.05 per megapixel of output. My 1,280 × 832 image (1.06 MP) cost $0.056. A 2,048 × 2,048 image costs $0.20. For comparison, FLUX 3 Klein on fal.ai costs about $0.01 for a similar image, while Seedream 5.0 Flash costs $0.027.
The model has open weights, so you can also run it yourself with Diffusers or ComfyUI, or try it for free in the OpenMOSS Space on Hugging Face. I did not test it locally - I used only the fal API.
My test: 10 Polish-language prompts
The same prompts were also given to six other models (see the new models on fal.ai roundup). I generated 1,280 × 832 images with default settings, using one attempt per prompt. I checked the text at full resolution. The prompt descriptions below are translations of the original Polish test prompts; the test itself was conducted in Polish.
| Prompt (translation of the Polish test prompt) | MOSS | FLUX 3 Klein | Notes on MOSS |
|---|---|---|---|
| Kraków at night | 8.0 | 6.0 | St. Mary’s Basilica is recognizable |
| An apartment-block kitchen | 8.5 | 8.0 | A glass in a wicker holder |
| Bakery sign: “ZDZISIA’S BAKERY” | 9.5 | 10 | Correct, but the sign is small |
| Jar labeled “BUCKWHEAT HONEY” | 9.5 | 8.5 | Label is correct |
| Portrait of a 62-year-old woman | 9.0 | 9.0 | Freckles, natural-looking skin |
| Hands making pierogi (Polish dumplings) | 8.5 | 7.5 | Hands are correct |
| Milk-bar menu | 6.0 | 10 | Misspelled “cheese pancakes” |
| Morskie Oko lake | 6.0 | 6.5 | Looks like the Dolomites |
| “JAZZ BY THE VISTULA” poster | 10 | 10 | Flawless |
| Chinese and English neon sign | 10 | 9.5 | Clearest text |
| Average | 8.5 | 8.5 | MOSS: 53 s; Klein: 6 s |
Polish text in images
MOSS handles short Polish text well: “Fresh doughnuts from 6:00,” “Meadow Apiary,” “BY THE VISTULA” and “89” came out correctly (translations of the original Polish test phrases, not examples from an English-language test). Longer text was less reliable: in a four-item menu, the model misspelled two words. It rendered the original Polish currency units abbreviation correctly in the menu, unlike Krea 3 and Seedream 5.0 Flash.
Photorealism and recognizable places
Realism is good: the images showed freckled skin, hands with five fingers and warm lighting. Recognizable places were more hit-and-miss. MOSS rendered Kraków’s Main Market Square correctly, but Morskie Oko came out looking like a lake in the Dolomites, surrounded by sharp, pink peaks.
MOSS Image: strengths and weaknesses
Strengths:
- Best Chinese text in the test, and very good short Polish text;
- Open weights under Apache-2.0, allowing commercial use and local deployment;
- Accepts JSON graphic layouts (I did not test this feature).
Weaknesses:
- Slow on fal.ai - a median of 53 seconds per image, the slowest of the seven models tested;
- Longer Polish text contains errors;
- Five times more expensive than FLUX 3 Klein for the same average score.
Who should use MOSS Image?
- Designers creating bilingual materials with Chinese text.
- Anyone looking for an open-weight model with strong text rendering and a license that permits commercial use.
- If you care only about Polish text and speed, FLUX 3 Klein is a better choice (local test).
I cover other recently released open models in my open AI models roundup, and models you can run on your own computer in the local AI hub.
Frequently asked questions
Can MOSS Image render Polish text?
It handled short Polish text, including Polish characters. In a longer text sample - a four-item menu - it made two spelling errors. The test was conducted in Polish.
Is MOSS Image free?
The weights are free under the Apache-2.0 license. On fal.ai, you pay $0.05 per megapixel; you can also try it at no charge in the Hugging Face Space.
Who is behind MOSS?
The OpenMOSS Team publishes MOSS model weights on Hugging Face. It released the Base and SFT versions on October 16, 2026.
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →Sources
- My test, October 20, 2026: 10 images generated through the fal.ai API, costing $0.56.
- fal.ai: MOSS Image model page and pricing (checked October 20, 2026).
- Hugging Face: OpenMOSS-Team/MOSS-Image-SFT (Apache-2.0 license).
More on this from QUANTSLANT
state of ai·How OpenAI Scaled ChatGPT Storage for One Billion UsersOpenAI’s technical write-up offers a rare look at the storage infrastructure behind ChatGPT’s growth to more than one billion users.