Nano Banana 2.1 and better shot-to-shot planning

Hi there,

In the last email, shots started being planned as a sequence. This update is about what happens when those planned shots are rendered: each one now goes to the video model with everything the scene already knows.

Also in this update: Google Nano Banana 2.1 replaces Nano Banana 2, and it now costs fewer Media credits per still.

1. Planned shots render with the cast, the place and the voices

When you send a planned scene to Composer, a shot with people in it, or a line to say, is rendered in the video model's reference mode. It gets its own frame, the frame it should end on, the portraits of everyone on screen, the scene's master view, the place, and the voice of each speaker. Each one is labelled, so the model knows which face and which voice belong to whom. A silent shot that has to join the one before it without a visible cut still starts from the exact frame.

2. Each shot continues from the take before it

A scene is rendered one shot at a time, in order. Each shot is given the finished take of the one before it, so it continues from what was actually rendered, not from a drawing of it. The first shot of a new scene starts fresh.

3. Takes are checked against the plan

When a take lands in Composer, its last frame is compared with what the plan expected. If something is off, the shot shows it and offers "Render again with fix", and the next shot is told about it so the mistake is not carried forward. Nothing is rendered again unless you ask.

4. A missing portrait stops the shot before you pay for it

If a character on screen has no portrait yet, the shot is held and tells you whose portrait to add, instead of rendering without the face.

5. Nano Banana 2.1 replaces Nano Banana 2

Google's newer image model is now behind the Nano Banana option for storyboard frames, character portraits and reference sheets. A still from text costs about 5 Media credits and a still made from references about 7, down from 8. In our side-by-side test it followed the prompt more closely: asked for an empty office with two signs, Nano Banana 2 (left) added a person and extra writing, and 2.1 (right) drew what was asked. Projects already set to Nano Banana 2 use 2.1 from now on. There is nothing to change.

Smaller things you'll notice

• A planned scene now records what the scene is about: its objective, its turn and its mood. Every shot prompt in that scene carries it.
• The setup page for Claude and ChatGPT now says which client ID belongs to which assistant. If ChatGPT showed "redirect_uri does not match", the Claude ID was entered there. The right one is on the page. It also helps to refresh all tools in the Frameliq plugin settings in Claude or ChatGPT.
• Claude and ChatGPT render planned shots the same way the app does, with the same references.

No fal key? You don't need one

Most people who stall in Frameliq stall at the same step: images and video ask for a fal key, and they don't have one. There are two ways past it:

• Creator Plus, $29/mo: 800 Media credits a month are included, so stills, video and voices work with no fal account.
• Creator, $19/mo: add your own fal key in Settings and pay fal directly for what you generate.

Hit reply and tell us how your scenes hold together now. We read every reply.

Happy filmmaking,

The Frameliq Team