Which AI image model is best on Labnana?

There is no single answer — the models differ by task. Nano Banana Pro leads on fidelity, in-frame typography, and 4K; GPT-Image-2 is the most literal about following a written brief; Seedream 5.0 Pro is the only one that revises a marked region instead of regenerating the frame; Nano Banana 2 is the low-cost option for iteration.

Can I switch models without losing my prompt?

Yes. The model dropdown sits inside the prompt box, so switching preserves your text. Settings a model does not support — 4K on Seedream 5.0 Pro, for example — fall back to the nearest supported option rather than failing.

How much can a free account generate?

A new account receives 2 free Nano Banana Pro generations at 1K or 2K, plus 1 more from each daily check-in and 2 for every new user who registers through a referral. Daily check-in also grants 5 universal credits, which are spendable on any model. 4K always requires a paid plan.

Which models support 4K?

Nano Banana Pro, Nano Banana 2, and GPT-Image-2 generate at 4K on a paid plan. Seedream 5.0 Pro supports 1K and 2K only.

How many reference images can I attach?

Up to 5 per generation on Nano Banana and Seedream 5.0 Pro, and up to 4 on GPT-Image-2 and GPT-Image-2.5.

Can one prompt be run across several models?

Yes, and it is the most direct way to compare them. Every prompt card loads into the generator, where the model can be switched and the same text re-run.

Are generated images watermarked?

No. Labnana applies no watermark on free or paid plans. Before using images in paid work, review the model providers’ usage policies.

How long does a video take to generate?

Usually five to ten minutes, with Seedance 2.0 Fast returning sooner than Pro. You can leave the page — the clip lands in My works · Video when it is ready.

Can Seedance 2.0 keep a person consistent from a reference portrait?

Yes. On reference to video you can attach up to 9 references of the same person and the model holds that face across the clip. References and a first frame cannot travel in the same request, which is why they live on separate pages.

Which video model suits vertical and feed crops?

Seedance 2.0 supports 9:16 and 3:4 portrait video, while Wan 3.0 supports the same two portrait ratios. Pick the closest supported ratio and crop the result if your feed requires another format.

Can I control the first and last frame of a video?

Yes. Seedance 2.0 Pro and Fast and both Wan 3.0 tiers accept a first frame, an end frame, or both. Given two ends, the model builds the shot between them.

Labnana

1 imageGPT-Image-2

Model directory · Labnana

AI Models on Labnana

Frontier image and video models from Google, OpenAI, ByteDance, and Alibaba, with the capability boundaries stated rather than implied.

Labnana runs two kinds of models in one workspace. On the image side, model families that differ in what they are actually good at: fidelity and in-frame typography, literal adherence to a written brief, or region-level editing of a finished image. On the video side, models that turn a still, a sentence or a reference set into a short clip. Choosing correctly is usually worth more than refining a prompt.

This page states each model’s capabilities, supported resolutions and formats, and free quota. Switch between the image and video catalogues below — the workflow notes and free-account entitlements sit with the image models, where generation starts for most people.

Models with a prompt library

Each model below has a dedicated library of published generations, every entry paired with the prompt that produced it.

Google · Gemini 3 Pro Image · Gemini 3.1 Flash Image

Nano Banana

Nano Banana is Labnana’s name for Google’s Gemini image models: Pro (Gemini 3 Pro Image) for fidelity, in-frame typography, and 4K, and 2 (Gemini 3.1 Flash Image) for faster, lower-cost iteration and extreme aspect ratios. Both accept the same prompt, so a brief can move between them without rewriting.

Typographic text in-frame
Character consistency across a series
Reference-guided generation
Up to 4K

Resolutions: 1K / 2K / 4K
Aspect ratios: 10 ratios; Nano Banana 2 adds 1:4, 4:1, 1:8, 8:1
Reference images: Up to 5 per generation
Free quota: 2 Pro generations on sign-up, +1 per daily check-in

Browse prompt library

基于参考图图一和图二所示的两位动漫风格女孩生成图片,确保背景是写实风格，只有少女为动漫风格，聚焦人物上半身。
写实画面描述：傍晚的宿舍，小桌上摆着一个咕嘟冒泡的小电锅，热气袅袅升起，被夕阳染成暖金色。 — Published generations from this model

Create one polished, source-grounded creative image. Create a strict top-down knolling composition i — Published generations from this model

OpenAI · GPT-Image-2.5 Flare · GPT-Image-2.5 Sunburst

GPT-Image-2.5

GPT-Image-2.5 is OpenAI’s current image model, and it is two models rather than one. OpenAI describes Flare as the small model optimised for speed and Sunburst as the base model optimised for quality. Both are in the Labnana generator at the same price, so switching between them costs a dropdown change.

Two tiers, one price
Precise editing
Subject preservation
Transparent backgrounds
Up to 4K

Resolutions: 1K / 2K / 4K
Aspect ratios: 10 ratios from 1:1 to 21:9
Reference images: Up to 4 per generation
Free quota: No dedicated quota — runs on universal credits

Browse prompt library

中式修仙世界,山顶,角色在亭子前，清晨黄金时刻，逆光拍摄，背向镜头，少女伸懒腰的剪影，身形轮廓清晰，天际线被晨光照亮，光线透过轻纱般的衣服透露出人体的轮廓，镜头光晕，浅景深，暖金色调，干净通透。
【反 — Published generations from this model

根据参考图角色生成一幅清新唯美的日系二次元简笔插画画面：角色小嘴轻轻张开，带着好奇又软萌的表情。抛弃原图服装特点，改为浅蓝绿色盐系马甲，内搭白色衬衫；头发散落并随意披在肩上，带两到三个随机可爱发卡。画 — Published generations from this model

OpenAI · GPT-Image-2

GPT-Image-2

GPT-Image-2 is the most literal model on Labnana: it holds to long, structured instructions, which makes it predictable when layout, style, and content are all specified up front. Two channels are available in the generator — the standard route and a Stable route for when consistency matters more than variation.

Literal instruction following
Clean illustrative rendering
Reference-guided generation
4K output

Resolutions: 1K / 2K / 4K
Aspect ratios: 10 ratios (1:1 – 21:9)
Reference images: Up to 4 per generation
Free quota: Campaign-based — remaining count shown in the generator

Browse prompt library

提取角色，二次元CG，带有一点手绘风格。
捕捉角色侧身看镜头的瞬间。
角色身高148cm
角色胸前的挂坠饰品是立体的、球状的
鲜明展现角色气质。画面中不出现文字 — Published generations from this model

Extract character, anime CG, with a slight hand-drawn style.
Capture the moment the character look — Published generations from this model

ByteDance · Seedream 5.0 Pro

Seedream 5.0 Pro

Seedream 5.0 Pro reads coordinates, which makes it the only model on Labnana capable of region-level editing: change a marked area of a finished image while the background, lighting, and remaining subjects hold. It also handles Chinese instructions natively and covers 1K and 2K output.

Region-level editing
Native Chinese instructions
Reference-guided generation
1K and 2K output

Resolutions: 1K / 2K
Aspect ratios: 10 ratios (1:1 – 21:9)
Reference images: Up to 5 per generation
Free quota: No model quota — universal credits

Browse prompt library

海边，图一角色倚靠在栏杆上，双手拿着相机看向画面一侧，被风吹乱的发丝，衣襟随风摆动，远处海面波光粼粼，倒映天空云层间隙间洒下的丁达尔光，无数海鸥掠过前后景形成视觉冲击感动态模糊效果，人物散发朦胧微光形 — Published generations from this model

生成一幅充满生活叙事感的写实摄影级画面，使用例图中的手绘速写本与咖啡馆场景。
画面主体是一只手举起一本绿色单色速写本，本子上的手绘内容与其身后真实咖啡馆场景形成"画中画"呼应。
一.画面主体与动作
1 — Published generations from this model

How generation works on Labnana

One workspace, one credit balance, one library. The workflow below is identical whichever model you select.

1
Write a prompt, or load one
The prompt box is pinned to the top of every page. Any card in the prompt libraries loads into it with the original model, aspect ratio, and reference images already set.
2
Set the model and output
Switch models from the dropdown inside the prompt box without losing your text, then set resolution and aspect ratio. The credit cost for the current settings is shown on the generate button before the run.
3
Keep working on the result
Generations are saved to your library and stay private until you publish them. With Seedream 5.0 Pro you can revise a marked region of a finished image and keep every version side by side.

What a free account gets

A free account can generate without payment and keeps accruing allowance through daily check-in and referrals. Free quota is model-specific; universal credits work everywhere.

On sign-up: 2 free Nano Banana Pro generations at 1K or 2K
Daily check-in: 5 universal credits, plus 1 Nano Banana Pro generation
Referral: +2 Nano Banana Pro generations for every new user who registers
Universal credits: Spendable on any model, including GPT-Image-2 and Seedream 5.0 Pro
Resolution limits: Free quota covers 1K and 2K; 4K requires a paid plan on every model
Watermarking: None, on free or paid plans

GPT-Image-2 free generations are issued through platform campaigns; the model dropdown shows the remaining count on your account. Seedream 5.0 Pro has no model-specific quota and bills against universal credits.

View plans and credits

Specifications at a glance

Resolutions and free quota as the generator applies them today. Credit costs are shown live on the generate button and are therefore not duplicated here.

Model	Provider	Resolutions	Free quota	Best for
Nano BananaGemini 3 Pro Image · Gemini 3.1 Flash Image	Google	1K / 2K / 4K	2 Pro generations on sign-up, +1 per daily check-in	In-image text, character consistency, photoreal detail
GPT-Image-2.5GPT-Image-2.5 Flare · GPT-Image-2.5 Sunburst	OpenAI	1K / 2K / 4K	No dedicated quota — runs on universal credits	Precise edits, subject preservation, transparent assets
GPT-Image-2GPT-Image-2	OpenAI	1K / 2K / 4K	Campaign-based — remaining count shown in the generator	Spec-driven briefs, clean illustration, layout control
Seedream 5.0 ProSeedream 5.0 Pro	ByteDance	1K / 2K	No model quota — universal credits	Region-level edits, local corrections, Chinese instructions

Also selectable in the generator

Nano Banana 2 (Gemini 3.1 Flash Image) is the faster, lower-cost tier of the Nano Banana family, and the tier that adds the 1:4, 4:1, 1:8, and 8:1 aspect ratios. Its prompts are indexed on the Nano Banana page.
GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst are OpenAI’s newest image models, both selectable in the generator on the same channel and at the same price as GPT-Image-2. What each one is for is on the GPT-Image-2.5 page.

Selecting a model

Typography that has to be readable

Nano Banana, on its Pro tier. Posters, packaging, signage, UI mockups — quote the exact wording and state its role. It is also the only family here that pairs 4K output with a standing free quota.

A brief with fixed requirements

GPT-Image-2. It holds to long, structured instructions, which makes it the predictable option when layout and style are both non-negotiable.

One region of a finished image is wrong

Seedream 5.0 Pro. It is the model behind Edit Image: mark the region, state the change, and the rest of the frame is preserved.

Exploring an idea before committing

Nano Banana, on its 2 tier. Lower cost per image and faster turnaround for working through variations, then switch to Pro for the final render.

Video models

Labnana also generates video. The parameter bar filters itself to whatever the selected model supports — resolution, duration, aspect ratio and the end frame slot all follow the model, so an unsupported setting never reaches submit.

Seedance 2.0 · ByteDance

Seedance 2.0

Seedance 2.0 is the tier pair on Labnana: Pro renders up to 1080p, Fast stops at 720p and comes back sooner. Everything else is shared — six aspect ratios, four to fifteen seconds, and an optional end frame that pins where the shot finishes. Switching tiers keeps your prompt and your uploads.

First and end frame control
Up to 1080p on Pro
4–15 second clips
Prompts up to 1,500 characters

Learn about this model

Wan 3.0 · Alibaba

Wan 3.0

Wan 3.0 comes in two tiers on Labnana and renders audio together with the picture rather than handing you a silent clip. It runs from two to thirty seconds — twice the length of anything else here — at 480p, 720p or 1080p, and takes both a first and an end frame. Prime matches the standard tier’s capabilities and comes back sooner.

Sound rendered with the picture
Clips from 2 to 30 seconds
480p, 720p and 1080p
First and end frame

Learn about this model

MiniMax H3 · MiniMax

MiniMax H3

MiniMax H3 renders four to fifteen seconds at 768p or 2K, across six aspect ratios including 21:9. Every clip comes back with 32 kHz stereo audio; there is no switch to turn it off, because the sound is part of what the model generates rather than a track laid underneath. It reads reference images, reference video and reference audio, and it is the only model on Labnana that will work from an end frame alone.

32 kHz stereo, always on
4–15 seconds at 768p or 2K
Six ratios, 21:9 included
End frame on its own

Learn about this model

Video specifications at a glance

Capabilities as the generator applies them today. Credit costs are shown live on the generate button and are therefore not duplicated here.

Model	Provider	Duration	Resolution	Aspect ratios	End frame	Best for
Seedance 2.0 Pro	ByteDance	4–15s	480p / 720p / 1080p	6 (16:9, 4:3, 1:1, 3:4, 9:16, 21:9)	Supported	Final renders, faces and people, 1080p delivery
Seedance 2.0 Fast	ByteDance	4–15s	480p / 720p	6 (16:9, 4:3, 1:1, 3:4, 9:16, 21:9)	Supported	Drafts, prompt iteration, quick turnaround
Wan 3.0	Alibaba	2–30s	480p / 720p / 1080p	5 (16:9, 4:3, 1:1, 3:4, 9:16)	Supported	Long clips, shots that need sound
Wan 3.0 Prime	Alibaba	2–30s	480p / 720p / 1080p	5 (16:9, 4:3, 1:1, 3:4, 9:16)	Supported	The same output on a shorter wait
MiniMax H3	MiniMax	4–15s, whole seconds	768p / 2K, 24 fps	6 (21:9, 16:9, 4:3, 1:1, 3:4, 9:16)	Supported, and works without a first frame	Clips that need sound, ultrawide framing

Where you generate video

Every model above runs in the same box. Describe the shot, add reference material if you have it, and switch between reference and first-frame mode without leaving the page.

AI Video GeneratorGenerate a clip from a written prompt, or supply reference images, a reference clip or a first frame so the subject, the motion and the opening frame follow your own material. Both modes run in the same composer.

Popular comparisons

Nano Banana vs GPT-Image-2.5 GPT-Image-2 vs GPT-Image-2.5 Nano Banana vs Seedream 5.0 Pro Seedance 2.0 vs Wan 3.0 MiniMax H3 vs Wan 3.0

Frequently asked questions

Select a model and start generating

The prompt box at the top of this page runs every model listed here.

View pricing

Models with a prompt library

Nano Banana

GPT-Image-2.5

GPT-Image-2

Seedream 5.0 Pro

How generation works on Labnana

Write a prompt, or load one

Set the model and output

Keep working on the result

What a free account gets

Specifications at a glance

Also selectable in the generator

Selecting a model

Typography that has to be readable

A brief with fixed requirements

One region of a finished image is wrong

Exploring an idea before committing

Video models

Seedance 2.0

Wan 3.0

MiniMax H3

Video specifications at a glance

Where you generate video

Popular comparisons

Frequently asked questions

Which AI image model is best on Labnana?

Can I switch models without losing my prompt?

How much can a free account generate?

Which models support 4K?

How many reference images can I attach?

Can one prompt be run across several models?

Are generated images watermarked?

How long does a video take to generate?

Can Seedance 2.0 keep a person consistent from a reference portrait?

Which video model suits vertical and feed crops?

Can I control the first and last frame of a video?

Select a model and start generating