Kolors Text-to-Image: 6 Prompt Tests (1024px)
Kolors is a diffusion-based text-to-image model from the Kuaishou Kolors team. The project highlights strong Chinese and English text rendering and solid prompt understanding. This post runs six prompts that stress text, portraits, product photos, and big sci-fi scenes.
Model link
Test setup
- Size: 1024×1024
- Steps: 30
- Guidance scale: 3.5
- Samples: 1
- Negative prompt: “bad, blurry, watermark”
Test 1: Chinese text rendering

This prompt checks small text legibility plus macro detail. If characters warp, try fewer style adjectives and ask for a larger sign.
Test 2: Neon sign typography in a complex scene

Text inside busy lighting often breaks in subtle ways. A good result keeps the word readable while preserving reflections and rain mood.
Test 3: Portrait + sunset lighting

Portrait prompts show how well the model handles skin, hair, and color grading. Keep the prompt simple if faces drift.
Test 4: Epic sci-fi scale

This checks composition and worldbuilding. The best outputs keep coherent lighting and readable structures even with lots of elements.
Test 5: Clean product photo

Product-style prompts rely on believable shadows and materials. If props drift, state exact counts and keep the scene minimal.
Test 6: Minimal poster typography

Poster prompts are a fast way to test layout and text. If the letters are not perfect, generate the background and add type in a design tool.
Quick takeaways
| What was tested | What worked | What to watch |
|---|---|---|
| Chinese and English text | Readable layouts in many cases | Exact typography can still deform |
| Complex scenes | Strong mood and composition | Busy prompts can introduce artifacts |
| Product lighting | Soft studio shadows | Prop counts can drift without anchors |