Creations: AI Image Studio
Three models. Infinite outputs. One creative loop.
AIKIZI's Image Studio turns any idea into a finished visual. Three generation models, each with a different engine and character, feed into the same decode-and-refine loop that powers the rest of the platform.
You write a prompt, choose a model and resolution, optionally attach reference images, and the system returns a finished creation in seconds. Every output can be decoded instantly, turning the analysis back into your next generation prompt. The loop is the feature.
The three models are not redundant. Each sits at a different point on the speed-vs-fidelity curve, and each produces a visibly different aesthetic signature. Choosing the right one is part of the creative process.
Three Models
The flagship. Highest fidelity, best prompt adherence, strongest at complex scenes with multiple subjects. Ideal for portfolio-grade output and reference-image workflows.
Fast iteration model. Lower latency, slightly softer detail, excellent for rapid exploration and prompt testing. Good baseline when you are still finding the direction.
Painterly, high-contrast aesthetic. Distinctive look that leans cinematic. Strong at environments, mood pieces, and stylized character work. A different visual lane.
Aspect Ratios & Resolution
Every model supports ten aspect ratios and three resolution tiers. The combination determines both the visual canvas and the token cost.
Aspect Ratios
1:1 4:3 3:4 16:9 9:16 3:2 2:3 4:5 5:4 21:9
Square for social and avatars. Standard and wide landscape for editorial and cinematic. Portrait ratios for mobile-first and character work. Ultra-wide for panoramic compositions.
Resolution Tiers
| Tier | Output | Cost |
|---|---|---|
| 1K | ~1024px long edge | 4-8 tokens |
| 2K | ~2048px long edge | 8-16 tokens |
| 4K | ~4096px long edge | 12-24 tokens |
Higher resolution does not just mean more pixels. The models produce finer micro-detail, sharper edges, and more coherent textures at 4K. It is not upscaling; it is a higher-quality generation pass.
Portrait Gallery
Reference Images
Attach up to 3 reference images (10 on Pro) alongside your prompt. The model uses them as visual anchors, not as inputs to copy. References guide composition, color palette, mood, and subject posture without overriding the prompt.
This is where Nano Banana Pro separates from the other two. Its reference-image conditioning is markedly stronger, producing outputs that feel like they belong in the same visual universe as the references while still being original compositions.
Reference images work best when they share a coherent aesthetic. Mixing wildly different styles dilutes the signal. Think of references as a mood board, not a collage.
The Decode-to-Creation Loop
This is the core workflow that makes the Studio more than a prompt box. Four steps, each feeding the next.
STEP 01
Generate
Write a prompt, pick a model and resolution, optionally attach references. Hit create.
STEP 02
Decode
The output lands in your library. One tap runs a full 10-class AI analysis: subjects, mood, camera, lighting, style DNA, color swatches, and more.
STEP 03
Refine
The decode result includes a generation-ready prompt in the Create tab. Edit it, adjust the model or resolution, and generate again.
STEP 04
Iterate
Each generation can be decoded again. The loop tightens with every pass. Three or four rounds typically converge on something you could not have prompted from scratch.
Token Economics
| Action | Cost | Notes |
|---|---|---|
| Image Gen (1K) | 4-8 tokens | Cheapest exploration |
| Image Gen (2K) | 8-16 tokens | Good balance |
| Image Gen (4K) | 12-24 tokens | Portfolio quality |
| Image Decode | 3-5 tokens | Full 10-class analysis |
| Download from Explore | 10 tokens | Other creators' work |
A typical create-decode-refine cycle costs 7-13 tokens per round at 1K resolution. On the Basic plan (180 monthly credits), that is roughly 14-25 full cycles per month. Pro (540 credits) gives 40-75+ cycles.
What You Can Build
The Studio is not a toy. Creators on the platform are producing editorial photography, character concept art, product mockups, social media content, album artwork, and book covers. The decode loop means every output teaches you something about what the model sees and how to steer it next time.
The three-model system matters here. Nano Banana Pro for final deliverables. NB2 for fast sketching and direction-finding. Seed Dream when you want something that looks nothing like the other two. Switching models mid-session is a creative tool, not just a technical option.
Blueprints (Realms) add another layer. Admin-curated prompt templates that lock the creative direction while letting you customize the details. Think of them as creative constraints that produce surprisingly varied outputs.
Landscape Gallery
AIKIZI IMAGE STUDIO — THREE MODELS, ONE LOOP