What's New

Introducing Skills (beta)

Skills teach Agent how you like things done. Type / in a chat to pick from 30 built-in guides, from product shots to film stocks, or write your own for the whole workspace. It's in its early stages, so give it a try and tell us what you think.

Two new video models

MiniMax H3 makes 4 to 15 second clips at 768p or 2K, with sound. Start from a prompt, animate a first frame, set a last frame to land on, or hand it up to nine reference images to follow.

Gemini Omni 1.1 Flash makes 10 second clips from 360p up to 4K, in 16:9 or 9:16, also with sound. It takes a start frame, an end frame, or reference images.

Both always generate audio, so there is no toggle to remember.

A tidier feed

The feed can get busy once you have hundreds of generations. Click any date header to collapse it into a compact preview, and older days now start collapsed, so skimming back in time is much faster. Spot the one you were looking for? Click it, and the day expands with it right in view.

Everything is faster

The image picker now opens instantly, even in workspaces with thousands of images. Scrolling stays smooth, and thumbnails load at the right size, right when you need them.

Images tray

Every image an agent chat creates now lives in a tray above the composer, so you never scroll back through the thread to find one. Expand it into a grid, open any image full screen, or drag one into the composer as a reference.

No more waiting your turn

You can keep typing while Agent is working. Send a message and it queues up to run next, or press Steer to redirect the current turn without starting over.

A sprite sheet packs a whole set of characters, poses, or frames into one image. Until now, getting one of them out on its own meant careful manual cropping and masking.

You can now cut a sheet into clean transparent cutouts in one go. Generate a sheet with any model, or upload one you already have, and Batch Banana detects each sprite and cuts it out. Review the cutouts, merge any that got split, and save the ones you want to your library as separate transparent images. Use them as game sprites, or drop them into a poster, an ad, or any design.

Projects

Projects are a quick way to keep separate work separate. Pick one from the top bar and everything you generate is tagged to it, and your feed shows only what belongs to it. Agent chats are scoped the same way, so each project keeps its own threads. Run a client job, a game, and a campaign side by side without their images mixing.

Agent is a new way to create in Batch Banana. Instead of writing prompts yourself, have a conversation: explore directions, ask it to search the web for inspiration, and let it write the prompts and generate images or video for you. Attach images to give it direction, then keep iterating in the same chat until it's right.

Everything Agent generates lands in your feed like any other generation.

Prompt sets

Give Agent a bigger brief and it proposes a whole set of prompts. Edit them, pick which ones to run, then run the set and refine from the results.

Agent can do things

Generating isn't all of it. Agent can also create a project, invite someone to your workspace, search your library, or pull in an image from a URL as a reference. Anything with real consequences asks you first.

Pricing

Chatting with Agent is free. The images and video it generates cost the same credits as ones you run yourself, and bigger runs show their price and ask before spending.

You can now generate video in Batch Banana. Switch the composer from Image to Video and write a prompt like you always do, or attach an image to animate it. Pick aspect ratio, duration, resolution, and audio, and see the credit cost before you submit.

Finished video plays in your feed and lives in your library next to your images.

Currently supporting

• Flux 3 - Black Forest Labs' flagship model: clips up to 20 seconds with synchronized audio, lip-synced dialogue, and keyframe control
• Seedance 2.5 - ByteDance's flagship multimodal video with native audio, clips up to 30 seconds
• Seedance 2.0 - ByteDance's multimodal video with synchronized audio, clips up to 15 seconds
• Veo 3.1 - Google's premium quality with synchronized audio and start and end frame control
• Kling 3.0 - Kuaishou's cinematic clips up to 15 seconds at up to 4K, with optional native audio
• Gen-4.5 - Runway's high-fidelity motion and strong prompt adherence

Nano Banana 2 Lite is Google's fastest and cheapest image model, now in Batch Banana at 8 credits per image, half the price of Nano Banana 2. It generates an image in about four seconds, nearly three times faster than the full model.

That speed changes how you work. Iterate freely: try directions, refine a prompt over many rounds, make quick edits. When you land on something you like, switch to a heavier model for the final render.

Lite generates and edits with up to 14 reference images and every aspect ratio in the app, at 1K resolution. New workspaces start with it selected.