Krea 2 Image Generation Guide: Style Transfer, Moodboards, and Cinematic Control
Wanderson Jackson
Updated August 2026. 9-min read. Krea 2 is a foundation image model trained from scratch for style transfer and creative control, available on Avocado AI for 1-2 credits per image.
Krea 2 is the first foundation image model from Krea AI, the creative AI suite that combines image generation, video, and 3D tools in a single workspace. Unlike Krea's earlier models, which were fine-tuned variants built on top of existing architectures like Flux, Krea 2 was trained from scratch. The result is a model purpose-built for creative workflows: style transfer from reference images, moodboard-driven generation, and tunable creativity controls.
The "foundation" label matters. Most AI image models on the market are either general-purpose (GPT-Image 2, DALL-E) or fine-tuned forks of open-source architectures (Flux, SDXL). Krea 2 is Krea's attempt to own the full stack, the way Midjourney did with its V-series. The differentiator is creative control: rather than just generating images from text, Krea 2 is designed to absorb a visual style and apply it consistently across multiple generations.
On Avocado AI, Krea 2 is available in two variants: Krea 2 Large (2 credits per image) for higher fidelity work, and Krea 2 Medium (1 credit per image) for faster iteration. Both are accessible from Starter plan upward and can be used within the Workspace alongside 15+ other image models.
Key Capabilities
Generation modes:
Text-to-image generation with natural-language prompts
Style reference: upload a reference image and Krea 2 applies its visual style to new generations
Moodboard generation: combine multiple references to define a consistent aesthetic direction
Edit mode: modify specific regions of generated images with targeted prompts
Output resolution: 1024x1024 native (varies by variant and prompt)
Two quality tiers on Avocado: Large (higher detail) and Medium (faster generation)
Special features:
Style transfer: upload a reference image and Krea 2 extracts its color palette, lighting, composition, and texture patterns
Moodboard workflows: combine multiple references to create a hybrid aesthetic
Tunable creativity: adjust how closely the model follows the prompt versus exploring creative variations
Node editor: Krea's native platform offers a visual node-based workflow for complex generation pipelines
LoRA training: train custom style models on Krea's native platform using your own images
What Krea 2 does NOT do:
On-image text rendering (GPT-Image 2 and Ideogram V3 are stronger here)
Native SVG or vector output (Recraft V4 Vector handles this)
Video generation (Krea offers video through other models like Hailuo and Kling, not Krea 2 itself)
Prompt Engineering Guide
The Krea 2 prompt formula
Krea 2 responds best to structured prompts that separate the subject from the aesthetic. Unlike models that benefit from single-sentence descriptions, Krea 2's style-transfer architecture means it pays close attention to mood, lighting, and texture descriptors.
Formula:
[Subject and action] + [Setting and environment] + [Lighting and mood] + [Style reference or aesthetic keyword] + [Composition notes]
Core tips
Name a specific aesthetic reference. Krea 2 has been trained on a broad visual vocabulary. Referencing styles like "1970s Kodachrome," "editorial fashion photography," or "Studio Ghibli background art" gives the model a clear target. Avoid vague terms like "artistic" or "high quality."
Use moodboards for batch consistency. If you need 20 product images with the same visual language, upload 3-5 reference images as a moodboard rather than trying to describe the style in words each time. Krea 2's moodboard feature extracts the shared visual DNA from multiple references.
Separate subject from style in your prompt. Write "A ceramic vase on a marble countertop" as the subject, then add "warm studio lighting, shallow depth of field, editorial product photography" as the style layer. Krea 2 processes these as distinct inputs.
Be specific about lighting. Krea 2 is particularly sensitive to lighting descriptors. "Golden hour side lighting through a window" produces very different results from "flat overhead fluorescent." Name the light source, direction, and quality.
Adjust creativity for iteration vs. final output. Lower creativity settings produce more predictable, prompt-faithful results (good for client work where you need consistency). Higher creativity settings let the model explore (good for brainstorming and mood discovery).
Texture words carry more weight than color words. Instead of specifying "warm orange tones," describe "rough terracotta, sun-bleached linen, weathered wood grain." Krea 2 interprets material descriptions more precisely than abstract color references.
Keep negative prompts minimal. Krea 2's architecture handles negative prompts differently from Flux-based models. Overloading negatives can reduce output quality. Focus on positive descriptors instead.
Use the edit mode for refinements. Rather than regenerating a full image to fix one area, use Krea 2's edit mode to target specific regions. This saves credits and preserves the parts of the image that already work.
Example prompts
Product photography:
"A minimalist perfume bottle on a polished black surface. Single soft light from the upper left, creating a subtle reflection. Deep shadows, generous negative space, editorial product photography. 3:4 aspect ratio."
Cinematic scene:
"A lone figure walking through a fog-covered forest at dawn. Diffused golden light filtering through tall trees, ground mist, volumetric atmosphere. Wong Kar-wai color palette with muted teals and warm amber highlights. 16:9 widescreen."
Lifestyle brand content:
"A woman sitting at a sunlit cafe table, hands wrapped around a ceramic mug. Shot on 35mm film with natural grain, warm morning light, candid moment. Soft focus background with bokeh from string lights. Relaxed, authentic mood. 4:3 aspect ratio."
Pricing
On Avocado AI (credit-based)
Model
Credits per image
Plan availability
Krea 2 Large
2
All tiers (Intro and above)
Krea 2 Medium
1
All tiers (Intro and above)
On the Starter plan (300 credits/month for EUR 39/month), a Krea 2 Large image costs roughly EUR 0.26. A Krea 2 Medium image costs roughly EUR 0.13. Credits roll over for one year.
On Krea AI natively (compute-unit-based)
Plan
Monthly price
Compute units/month
Free
$0
Limited daily allowance
Basic
$9
5,000
Pro
$35
20,000
Max
$70+
40,000+
Business
$200+
80,000+
Krea's pricing uses compute units, not per-image credits. The cost per image varies depending on the model and settings used. Krea 2 is available on all Krea plans, including the free tier.
Cost-saving strategies:
Use Krea 2 Medium for brainstorming and iteration (1 credit on Avocado), then switch to Large for final outputs
Combine Krea 2 with Avocado's Workspace to generate, edit, and compose in one session without exporting between tools
Use moodboards to reduce failed generations: a well-curated reference set means fewer regenerations
Strengths and Trade-offs
Strengths
Style transfer is genuinely useful. Unlike most image models that treat style as a prompt-only exercise, Krea 2 can absorb a reference image's visual language and apply it to new subjects. This is valuable for brand consistency across multiple product lines or campaign assets.
Moodboard workflows reduce iteration. Instead of describing the same aesthetic in every prompt, upload references once and Krea 2 maintains the visual DNA. This cuts prompt engineering time significantly for batch work.
Competitive per-image cost on Avocado. At 1 credit for Medium and 2 for Large, Krea 2 is priced alongside Nano Banana 2 and GPT-Image 2. For cinematic and artistic output, it offers strong value.
Tunable creativity is a real differentiator. Most models give you one mode: follow the prompt. Krea 2 lets you dial creativity up for exploration or down for precision. This matters for client work where predictability is a requirement.
Foundation model means unique output. Because Krea 2 was trained from scratch rather than fine-tuned on Flux or SDXL, its output has a distinct visual character. The images don't look like "another Flux variant."
Trade-offs
Weak at text rendering. If your images need on-image text (logos, labels, captions), GPT-Image 2 or Ideogram V3 are better choices. Krea 2 treats text as a visual element rather than a typographic one.
Not the fastest model. Krea 2 Large takes longer to generate than Nano Banana 2 or Flux variants. For high-volume, speed-critical workflows (e.g., batch ad creative), faster models may be more practical.
Square output by default. Krea 2 on Avocado tends to return 1024x1024 output regardless of requested aspect ratio. If you need 16:9 or 9:16, plan to crop and resize with PIL or an image editor after generation.
Style transfer quality depends on reference images. Low-resolution or visually noisy reference images produce unpredictable results. The moodboard feature works best with curated, high-quality references.
Ecosystem is Krea-native. The best Krea 2 features (moodboards, LoRA training, node editor) live on Krea's own platform, not on Avocado. On Avocado, you get the core generation model without the full creative workflow suite.
How It Compares
Krea 2 vs. GPT-Image 2: GPT-Image 2 is the stronger model for photorealistic output and on-image text rendering. It handles complex scenes with multiple subjects more reliably. Krea 2 wins on stylistic consistency and mood-driven generation. For product photography with text overlays, GPT-Image 2. For brand mood imagery and editorial aesthetics, Krea 2.
Krea 2 vs. Recraft V4: Recraft V4 is purpose-built for design work: logos, icons, UI elements, and vector output. Krea 2 is better for photographic and cinematic imagery. If you need SVG output or tight style control for brand assets, Recraft V4. If you need atmospheric, mood-driven visuals, Krea 2.
Krea 2 vs. Nano Banana 2: Nano Banana 2 is the default, general-purpose model on Avocado. It is faster and costs 1 credit across all variants. Krea 2 Large at 2 credits produces more nuanced, stylistically rich output. For quick drafts and clean product shots, Nano Banana 2. For editorial and cinematic work, Krea 2.
Krea 2 vs. Seedream V5 Pro: Seedream V5 Pro (ByteDance) excels at dense layouts and multilingual text rendering across 14 languages. Krea 2 is stronger on Western editorial and cinematic aesthetics. For Asian-market content or text-heavy designs, Seedream V5. For lifestyle and brand imagery targeting Western audiences, Krea 2.
FAQ
What is the difference between Krea 2 Large and Krea 2 Medium?
Krea 2 Large (2 credits) produces higher-detail output with more nuanced textures and lighting. Krea 2 Medium (1 credit) is optimized for speed and works well for iteration and brainstorming. Use Medium for drafts, Large for finals.
Can I use Krea 2 on Avocado AI's free plan?
Avocado AI does not have a free plan. Plans start at EUR 19.99/month (Intro, 100 credits). Krea 2 is available on all tiers. On Krea's own platform, Krea 2 is available on the free tier with limited daily compute.
Does Krea 2 support style reference images?
Yes, Krea 2 was designed for style transfer. Upload a reference image and the model extracts its visual style (color palette, lighting, texture, composition) and applies it to your new generation. This works on both Avocado and Krea's native platform.
How does Krea 2 handle aspect ratios?
Krea 2 on Avocado tends to return 1024x1024 output by default. If you need a different aspect ratio, generate at 1:1 and crop with PIL or an image editor. Supported ratios include 1:1, 16:9, 9:16, 4:3, 3:4, 21:9, 3:2, and 2:3.
Is Krea 2 good for product photography?
Krea 2 produces strong product imagery for editorial and lifestyle contexts. For clean, studio-lit product shots with precise backgrounds, GPT-Image 2 may be more consistent. Krea 2 shines when you want atmospheric, mood-driven product visuals.
Can I train a custom LoRA with Krea 2?
LoRA training is available on Krea's native platform, not on Avocado AI. If you need custom style models, use Krea's Train LoRA feature with your own image dataset, then generate through their platform.
What makes Krea 2 different from other AI image models?
Krea 2 is a foundation model trained from scratch, not a fine-tuned variant of Flux or SDXL. Its differentiator is creative control: style transfer from reference images, moodboard workflows, and tunable creativity settings. Most competing models treat style as a text prompt only.
How many credits does Krea 2 cost on Avocado AI?
Krea 2 Large costs 2 credits per image. Krea 2 Medium costs 1 credit per image. On the Starter plan (EUR 39/month, 300 credits), that works out to roughly EUR 0.13 to EUR 0.26 per image depending on the variant.
How to Pick in Under 30 Seconds
Need on-image text or logos? Use GPT-Image 2 or Ideogram V3 instead.
Need SVG or vector output? Use Recraft V4 Vector.
Need fast batch generation for ads? Use Nano Banana 2 (1 credit, fastest).
Need cinematic, mood-driven imagery with style consistency? Krea 2 is the right choice.
Need to match a brand's visual style across 20+ images? Use Krea 2 with a moodboard of reference images.
Working on a tight budget? Krea 2 Medium at 1 credit matches Nano Banana 2 pricing.
Need photorealistic output with complex scenes? GPT-Image 2 at high quality is more reliable.
Need multilingual text rendering? Seedream V5 Pro supports 14 languages natively.
If you want one workspace that runs Krea 2 alongside 15+ other image models in a single credit pool, start with Avocado AI. Check out our pricing for details. Every plan includes access to the full model catalog, with credits that roll over for one year.
Wanderson Jackson is the founder of Avocado AI, a creative workspace that brings together image, video, and audio generation models under one roof.
