This guide covers the two ways to make an image, the five models you can choose from, and how to get the best result from each.
Quick start
Open the prompt box, or go to Generate
Make sure you're in image mode
Write a prompt, and for image-to-image, upload your starting image first
Set your aspect ratio, resolution, and model
Click the paper plane icon to generate
Your image appears in the Library when it's ready. You can queue another generation while the first one renders.
The two ways to make an image
Mode | What you bring | Best for |
Text to image | A prompt | New scenes, concepts, ad creative, infographics — anything you don't already have a photo for |
Image to image | A photo + a prompt | Editing, restyling, background changes, product placement, variations of a shot you have |
Choosing a model
Magicfit is multi-model, so you can match the engine to the job. Each model has a personality, pick by what you're making, not just by what's newest.
Model | Built on | Strengths | Best for |
Nano Banana | Gemini 2.5 Flash Image | Fast, fun, reliable edits; strong character consistency | Quick edits and everyday generations |
Nano Banana 2 | Gemini 3.1 Flash Image | Pro-grade reasoning at Flash speed; up to 4K; consistency for up to 5 people; real-world grounding | High-volume, high-quality work that still needs to be fast |
Nano Banana Pro | Gemini 3 Pro Image | The highest-fidelity option; "thinking" pass before rendering; best-in-class text rendering; up to 4K; consistency across many inputs | Complex compositions, infographics, posters, anything text-heavy or multilingual |
GPT Image 2 | OpenAI | Near-perfect text rendering; strong instruction-following; "thinking mode"; multilingual (incl. CJK, Hindi, Bengali); up to 4K | Dense text, precise layouts, mockups where every word must be right |
Recraft v4.1 | Recraft | Design-grade output; clean icons, logos, and vector-like forms; brand color control; strong from short prompts | Brand assets, packaging, product mockups, e-commerce imagery |
Quick picks:
A poster or infographic with lots of text → Nano Banana Pro or GPT Image 2
A clean product shot or packaging mockup → Recraft v4.1
Fast, good-enough generations at volume → Nano Banana 2
A quick edit to an existing image → Nano Banana
A character or person who must look the same across shots → Nano Banana Pro or Nano Banana 2 (consistency for up to 5 subjects)
Text rendering matters more than you think. If your image contains words use Nano Banana Pro or GPT Image 2. Older or faster models garble text far more often.
Settings, explained
Every setting sits in the prompt box. What's available depends on the model you pick.
Setting | What it does |
×1 | How many images to generate. Use + and − to change it. Start with 1–2 to test your prompt before spending credits on variations. |
1:1 | Aspect ratio. Click to change. Options depend on the model (common: 1:1, 16:9, 9:16, 4:3, 3:4). |
2K | Resolution. Click to change. Higher resolutions cost more and take longer. Several models go up to 4K. |
nano banana pro | Model. Click to switch. This changes which features, resolutions, and aspect ratios are available. |
Writing prompts that work
A reliable structure for any image prompt:
Subject → Setting → Composition → Lighting → Style
A few rules that apply across both modes:
Be specific about the subject. "A bottle" is weak; "a frosted glass serum bottle with a matte black pump" gives the model something to render
Name the lighting. "Soft window light from the left" does more than any list of adjectives
Describe composition — close-up, wide, overhead, centered, rule-of-thirds
Put any text in quotes, and keep it short. "A label reading 'COLD BREW'" renders more reliably than a long paragraph
Add a style line at the end — editorial, minimal, cinematic, flat illustration
Shorter prompts work well on newer models (Recraft v4.1, the Nano Banana family) — you don't need to over-engineer
Each mode's article has a copy-paste template tuned for that workflow
Credits
Credits are spent when you generate
The exact credit cost is shown next inside the prompt box
Image cost scales with model × resolution, and the number of images you request
Failed generations are refunded automatically
Credits don't roll over between billing cycles
No refund on credits
Common mistakes
A vague subject. The more specific your description, the closer the result
No lighting direction. One lighting sentence transforms an image
Cramming long text into the prompt. Keep on-image text short and quoted
Generating ten variations before testing one