GPT Image 2.6 arrives in Seelab in two versions, Kling v4 joins the Video space, Nano Banana 2.1 becomes the default image model, and Gemini Omni Flash takes over from Veo on the Google side. Here is what changes, and when to use each model.

New: GPT Image 2.6, Sunburst and Flare

OpenAI's new image generation is available in the model selector, in two versions. According to OpenAI, it produces images 50% faster than 2.5.

• Sunburst: detailed compositions and precise edits. At normal or high quality, it keeps details stable from one edit to the next: ideal for evolving a visual step by step. Sharp, well-contrasted text.
• Flare: speed, for iterating. Built for posters, product visuals and layouts. Low quality by default: switch to normal or high for the final render.
• The same grid for both: 1k, 2k or 4k, 10 formats, 3 quality levels, from 2 to 11 credits per image. At high quality 4K, Sunburst 2.6 costs up to 3 times less than 2.5 (11 credits instead of 36).

Same prompt, same settings (3:4, 2k, normal quality): Flare gives warmer light and a more classic label, Sunburst a tighter framing and bolder text.

Kling v4: shots up to 30 seconds, and video you can edit

Kuaishou's new video engine is available in the Medias and Edit tabs of the Video space.

• Generation: shots from 3 to 30 seconds, in 720p or 1080p, in 16:9, 9:16, 1:1 and cinema 21:9. Fabrics, liquids, hair and human motion are rendered with very realistic physics.
• With references: up to 10 images and 5 videos in a single generation.
• Editing: it is now the default engine of the Edit tab. Upload a 3 to 30 second video and change the product, a character, the setting, the camera angle or the style, while keeping the original motion and framing.

Google: Veo ends, Gemini Omni Flash takes over

Google has retired Veo: its video generation now runs on Gemini Omni. In Seelab, Gemini Omni Flash 1.1 takes over.

• Videos from 3 to 10 seconds, in 720p or 1080p, in 16:9 and 9:16.
• Sound is always generated, in sync with the image.
• The prompt is optional when you start from an image: ideal to bring an approved visual to life.
• Also available in the Edit tab.

Nano Banana 2.1: the new default image model

Google's latest image generation offers the best quality-to-credits ratio of the Nano Banana family.

• It follows long, detailed instructions better and renders text more cleanly.
• Up to 14 reference images, including 4 characters who keep their face from one edit to the next.
• 1k, 2k or 4k, 14 formats including the 8:1 banner, from 2.3 to 7.7 credits per image.

Models change, your Seelab model stays

Market models evolve every month. Your Seelab model is trained on your products: it guarantees the shape, the label and the colors, whatever the setting. One training works in both image and video.

The right method: the market model for the scene, your Seelab model for the product. And if you'd rather delegate, the Seelab team models your products for you.

Your turn: open Creation or Video and pick the new model from the list.

Learn more

• Image models in detail
• Which image model to choose?
• Video models in detail
• Edit a video
• What is a Seelab model?

Updated on: 09/11/2026