The Model Landscape
Not all AI image models are created equal. Each has unique strengths, speeds, and price points. Here's your guide to choosing the right one.
Image Models
Nano Banana (1 credit)
Google's lightweight model. Blazing fast generation with surprisingly good quality.Best for: Quick iterations, testing prompts, casual use Speed: Very fast (~2-4 seconds) Quality: Good
Nano Banana Pro (2 credits)
The upgraded version with higher quality output while maintaining speed.Best for: Higher quality with fast speed, production use Speed: Fast (~3-6 seconds) Quality: Very good
Flux Schnell (2 credits)
Black Forest Labs' fast model. Excellent balance of speed and quality.Best for: General purpose, quick high-quality images Speed: Fast (~3-5 seconds) Quality: Very good
Flux Dev (5 credits)
More parameters and control. Supports up to 50 inference steps for highly detailed output.Best for: Detailed artwork, specific compositions, fine control Speed: Medium (~8-15 seconds) Quality: Excellent
Flux Pro (10 credits)
The premium tier. Best-in-class image quality with support for multiple source images.Best for: Professional work, highest quality output Speed: Slower (~15-30 seconds) Quality: Outstanding
SDXL (3 credits)
Stability AI's proven workhorse. Wide style range and excellent prompt following.Best for: Diverse styles, img2img, established workflows Speed: Medium (~8-12 seconds) Quality: Very good
SD 3.0 (5 credits)
The latest Stable Diffusion architecture with improved text rendering and composition.Best for: Text in images, complex compositions Speed: Medium (~10-15 seconds) Quality: Excellent
Choosing the Right Model
| Need | Recommended Model | |------|-------------------| | Speed over quality | Nano Banana | | Best overall | Flux Schnell | | Maximum detail | Flux Dev or Flux Pro | | Budget-friendly quality | Nano Banana Pro | | Text in images | SD 3.0 | | Image-to-image | SDXL or Flux Dev |
Pro Tip: Iterate Then Upscale
Start with a fast, cheap model (Nano Banana) to nail your prompt. Once you're happy with the composition, either regenerate with a higher-quality model or use the Upscale AI tool to enhance the result.