Text to Image

Generate images from text descriptions using various AI models.

How It Works

1. Write a text prompt describing your image
2. Select AI model(s) from the model selector
3. Choose settings (aspect ratio, quantity, etc.)
4. Click Generate

Settings

• Aspect Ratio: Choose image dimensions (1:1, 16:9, 9:16, etc.)
• Resolution: Tier selector on supported models — 1K (~1024px), 2K (Full HD), or 4K (UHD, SeeDream 4.x only). Availability is model-specific; some models are flat across tiers (Flux 2, Nano Banana, Imagen, Qwen) and some charge more at higher tiers (e.g. SeeDream 5.0 at 4K costs 2× the base).
• Quantity: Number of images to generate (1-4)
• Model Selection: Choose one or multiple AI models
• Stylization: Control creative interpretation (model-specific)
• Reference Images: Upload images to guide generation — see "Reference Images vs Source Images" below for what this actually does

Available AI Models

• FLUX.2: High-quality, versatile image generation
• GPT Image 2: OpenAI text rendering, infographics, photorealism
• Qwen Image 2 (Standard / Pro): Strong prompt following and text
• Nano Banana 2 / Nano Banana 2 Pro: Google multi-reference edits with web-grounded text
• Recraft 4: Logos, vector-style, brand design
• Grok Imagine Quality: xAI high-fidelity image generation
• SeeDream 5.0 Lite / Seedream 4.5: ByteDance high-resolution generation (up to 4K)
• Higgsfield Soul: Stylized, aesthetic image generation
• Midjourney V8 + Niji V7: Artistic and anime-style generation

See the full Models catalog for every available model.

Features

Multi-Model Generation

• Generate with multiple models simultaneously
• Compare results from different AI models
• Each model has unique strengths and styles

Moodboards

• Create collections of reference images
• Apply moodboard style to generations
• Manage and reuse moodboards across projects

Image Presets

• Quick style templates
• Save time with pre-configured settings
• Apply consistent styles across generations

Reference Images vs Source Images — Read This Before You Upload

This is the single most common confusion in Kolbo. The two concepts look similar but produce completely different results.

Reference Images (this page — Text to Image)

Reference Images guide style, mood, and composition. The model studies them and generates a brand-new image inspired by them. It does not copy their pixels.

Use Reference Images when you want to:

• Match an aesthetic ("make it feel like this brand's other ads")
• Borrow a color palette or lighting style
• Imitate a composition or pose

Do NOT use Reference Images when you want to:

• Embed a real logo, icon, or watermark into the output — the model will redraw a similar-looking shape, not your actual logo
• Keep a specific person's face exactly the same — use Visual DNA, or switch to Image Editing with the face photo as a Source Image
• Composite multiple uploaded assets together — that's an Image Editing workflow

Source Images (Image Editing tool)

Source Images get pixel-accurately composited into the output. Use Image Editing when you need a specific uploaded asset to actually appear in the final image. See the Image Editing docs for the full workflow.

Rule of thumb: If the user looking at the output should be able to see the exact pixels of an asset you uploaded, you need Image Editing, not Text to Image.

Tips

1. Be Specific: Detailed prompts get better results
2. Try Different Models: Each model interprets prompts differently
3. Reference vs Source: Reference Images inspire; Source Images embed. See above.
4. Experiment: Test different settings and prompts