Labnana
Sign In
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
Create one polished, source-grounded creative image. Create a richly layered European old-town slow-Create one polished, source-grounded creative image. Create a strict top-down knolling composition i生成一幅极致密集的涂鸦艺术（doodle art）风格肖像插画，使用例图中的画风与构成方式。 画面主体是一个大胡子男性头像，其面部与全身由数以百计的小型独立涂鸦角色、眼睛与几何图形密铺拼接而成。 一.[Anime Style] An anime girl taking a selfie in front of the real-world Akihabara Radio Kaikan buildi
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 5 per generation
Free quota
Campaign-based — remaining count shown in the generator
Browse prompt library
Extract character, anime CG, with a slight hand-drawn style. Capture the moment the character look竖版插画，复古手绘板风格，非厚涂，高对比度。整幅画摒弃平滑渐变，以老旧羊皮纸般的斑驳肌理、丝网印刷的粗糙网点与干涩的刮擦排线构成全部影调。画面色彩极度克制，仅使用深邃的暗黑与大面积高饱和的金黄色，两种Replace the character in image1 with the character in image2, and maintain consistency in art style.[Overall Style] vertical poster, semi-painted thick coating combined with letterpress printing tex
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生成一幅极致密集的涂鸦艺术（doodle art）风格肖像插画，使用例图中的画风与构成方式。 画面主体是一个大胡子男性头像，其面部与全身由数以百计的小型独立涂鸦角色、眼睛与几何图形密铺拼接而成。 一.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生成一幅写实静物摄影，主体是一本摊开的黑白钢笔速写本斜放在凌乱的白色工作桌上，跨页画满人物角色速写，周围散落钢笔、铅笔与骰子。 一.画面主体与动作 1.主体与摆放：一本无线胶装的速写本摊开成跨页，以
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. 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. 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. 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.

ModelProviderResolutionsFree quotaBest for
Nano BananaGemini 3 Pro Image · Gemini 3.1 Flash ImageGoogle1K / 2K / 4K2 Pro generations on sign-up, +1 per daily check-inIn-image text, character consistency, photoreal detail
GPT-Image-2GPT-Image-2OpenAI1K / 2K / 4KCampaign-based — remaining count shown in the generatorSpec-driven briefs, clean illustration, layout control
Seedream 5.0 ProSeedream 5.0 ProByteDance1K / 2KNo model quota — universal creditsRegion-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.

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