For photos, MonAI's strongest models are Qwen Lightning v2512 and RedCraft Z-Image, followed by RealVisXL v5, Firenze XL v2.1, epiCRealism XL and CyberRealistic v12. Below, the same five prompts run on each model with the same seed and size, using each model's own MonAI style.
Select a picture to see it full size. Copy prompt copies that model's exact prompt; Try opens MonAI with the model, prompt and settings filled in.
Jump ahead to:
¶ Which model should I use?
| Model | MonAI style | Best for |
|---|---|---|
| Qwen Lightning v2512 | True Cinematic Photo (guidance 1.2, 8 steps) | The most natural, candid-looking photos, and it follows the whole prompt, including the scene around the subject. It can write words in the picture: the bakery sign reads "Fresh Bread". |
| RedCraft Z-Image | True Photography (guidance 1.2, 6 steps) | Natural, true-to-life photos. Like Qwen, it follows long, plain descriptions, handles complex scenes with several people or objects and can write words in the picture. |
| RealVisXL v5 | RealVis5 Photo (guidance 5.5, 25 steps) | Crisp detail and strong contrast, especially in faces and textures. A dependable all-rounder for portraits and scenes. |
| Firenze XL v2.1 | Firenze v2 Photo (guidance 4.7, 28 steps) | Warm, film-like colour and careful composition. MonAI's own house model, also good for paintings and illustration. |
| epiCRealism XL | Epic Photo (guidance 4.4, 25 steps) | Soft, natural light and gentle skin tones; a favourite for portraits. |
| CyberRealistic v12 | Cyber v12 Photo (guidance 5, 27 steps) | A brighter, cleaner look that often frames the subject wider. Built on Illustrious XL, so it also understands anime-style tags. |
¶ Prompting tips
- Write sentences. All six models follow plain descriptions of the subject, setting and light.
- Complex scenes: for a scene with several people or objects, use Qwen Lightning or RedCraft Z-Image. RealVisXL, Firenze and epiCRealism are often unable to render complex scenes; they are at their best with one clear subject, such as a portrait.
- Text in the picture: Qwen Lightning and RedCraft Z-Image can write short words on signs, labels and posters. Put the words in quotes, as in
a sign that says "Fresh Bread". In the bakery comparison, RealVisXL, Firenze, epiCRealism and CyberRealistic misspelled the sign. - Negative prompts: the MonAI styles for Qwen and RedCraft Z-Image use no negative prompt; describe what you want instead. The other four use their style's negative prompt. See Negative Prompts by Model.
- CyberRealistic is built on the anime model Illustrious XL. Write
womanormanrather than1girlor1boy, and describe the clothing you want.
¶ How the pictures were made
Every picture is 768 × 960 with seed 3591. Each model uses its default MonAI style, which adds that style's words to the end of the prompt and sets its negative prompt, guidance scale and steps. Show prompt above each set gives the prompt; Copy prompt gives the exact prompt for each model. Each picture was rendered once. With another seed, a model can look different.
Source: the MonAI wiki · Updated · First published