What's New in Visuals
This is the biggest batch of new models we've added in months, and a new way to decide what a generation costs you.
There are 5 new image and video models, the newest text models, and 2 new voices. On top of that, the Veo 3.1 family and Kling now let you choose how good you want the output to be. That choice is really a price choice: the higher setting costs more credits, so you can spend on the take you're going to release and go cheap while you're still figuring it out.
The biggest batch of new models in months
Video
Luma Ray 3.2
Seedance 2.0 and Seedance 2.5, both from ByteDance
Image
Gemini Omni Flash
Nano Banana 2 Lite
Text
Opus 5, Sonnet 5 and GPT-5.6. These are available for text cards and for writing.
Gemini Omni and the Seedance models can work from more than a still image. They take a reference video, and Gemini Omni takes audio too. Those inputs live on the visual canvas. On the image and video generator pages you work from an image reference.
Pick how good you want it, and what it costs
Some models now let you set the quality of what comes out.
The whole Veo 3.1 family has quality modes. Pick 720p or 1080p.
Kling on motion video has a quality and resolution selector.
Read this as a cost control, because that's what it is. The higher setting costs more credits than the lower one for the same clip. So run the low setting while you're testing ideas, and save the high setting for the version that goes out.
What the visual canvas got (Alpha preview)
This was in alpha preview at the time, available to a small group. It's open to everyone now.
The card toolbar picked up 2 new controls, one for the watermark and one for the output format, so you set both on the card rather than after the fact.
This is also where a model's video and audio inputs live. If you want to hand Seedance a reference video, or give Gemini Omni an audio track to work from, the visual canvas is the place that takes one.
2 new voices for voiceovers
There are 2 new OpenAI voices to pick from: marin and cedar.
Also fixed
Project links and video model settings survive a save. You'd set them, save, and find them reset. They stay put now.
Spanish captions keep their opening question mark. It was being stripped out. It stays in.
Lyric video clips land where the timing says. Some were being placed 15 frames off. That's fixed.
Fewer prompts get blocked for no reason. Prompts that were perfectly fine were sometimes flagged as unsafe. That happens a lot less now.