NextLabs
Open sourceMIT · for every model on OpenAI Next

Every model that makes things,in one open lab.

Try, compare and call the models that make images.
Your key stays in your browser; every run keeps its request.

Open the workbench Browse the gallery GitHub
Seedream 5.0 Pro@openai-nextSeedream 5.0 Pro@openai-nextNano Banana Pro (GA)@openai-nextSeedream 5.0 Pro@openai-nextSeedream 5.0 Pro@openai-nextGPT Image 2.5@openai-nextSeedance 2.0@openai-nextSeedance 2.0@openai-nextNano Banana Pro (GA)@openai-nextSeedream 5.0 Pro@openai-nextSeedream 5.0 Pro@openai-nextSeedance 2.0 Mini@openai-nextVeo 3.1@openai-nextNano Banana 2 (GA)@openai-nextGPT Image 2.5@openai-nextSeedream 5.0 Pro@openai-next
Seedream 5.0 Pro@openai-nextSeedream 5.0 Pro@openai-nextNano Banana Pro (GA)@openai-nextSeedream 5.0 Pro@openai-nextSeedream 5.0 Pro@openai-nextGPT Image 2.5@openai-nextSeedance 2.0@openai-nextSeedance 2.0@openai-nextNano Banana Pro (GA)@openai-nextSeedream 5.0 Pro@openai-nextSeedream 5.0 Pro@openai-nextSeedance 2.0 Mini@openai-nextVeo 3.1@openai-nextNano Banana 2 (GA)@openai-nextGPT Image 2.5@openai-nextSeedream 5.0 Pro@openai-next
OpenAISoraGoogleGeminiNano BananaByteDanceDoubaoKlingMiniMaxHailuoxAIGrokSunoAnthropicClaudeDeepSeekQwenZhipuMoonshotKimi
OpenAISoraGoogleGeminiNano BananaByteDanceDoubaoKlingMiniMaxHailuoxAIGrokSunoAnthropicClaudeDeepSeekQwenZhipuMoonshotKimi

Quiet tools, dense where you work

Keyboard first: ⌘K finds any page or model, G then a letter jumps between tools.

GPT Image 2.5

1536×1024·14.2s·$0Saved to My works

API sk-…7f2a≈ $0.050Run⌘↵

One run, one graph

The request goes from your browser to OpenAI Next with your own key — the Lab never holds it. What comes back is kept with the request that made it.

Your promptand parameters
Your browserholds your key
OpenAI NextAPI · DRAW
GPT Image 2.5image
Seedance 2.0video
Suno v4music
My worksand the gallery
runnable here
0
verified live
0
vendors
0

Every model, its own hand

Each one is in the catalog with its real parameters, and each was run for real before it got here.

  • ByteDance's image model with strong Chinese text rendering.

    ByteDance · verified Oct 11Try it API

  • OpenAI's image model: a prompt, a reference picture if you like, and a transparent background on request.

    OpenAI · verified Oct 11Try it API

  • ByteDance's multimodal video model: text, first/last frame, reference image/video/audio, with synced audio.

    ByteDance · verified Oct 12Try it API

  • Google's image model, also known as Nano Banana Pro: ten shapes from 21:9 to 9:16, up to 4K.

    Google · verified Oct 12Try it API

  • OpenAI's newer image model, on the API platform, with the same controls as GPT Image 2.

    OpenAI · verified Oct 11Try it API

  • Google's faster image model, also known as Nano Banana 2: the same shapes, from 512 px to 4K.

    Google · verified Oct 12Try it API

Cat and monarchCat and monarch

Not just pictures, from the gallery

See the gallery

Late studio, felt piano

Suno v4

0:00 / 0:00

Chronicle

Every model, as it arrived

Releases, prices and what you can run here, from every vendor — synced daily from models.dev, Epoch AI and both OpenAI Next platforms.

Open the chronicle

Paste an OpenAI Next key and every model on the platform is one click away. It stays in this browser.

Works with API or DRAW keys — we detect which.

Open source, MIT.Run your own Lab, or read how every request is built. GitHub

~