AI image generation: photorealism, art, and text on an image
Photorealism, art and text on an image are three different jobs for an AI. Which models to pick in the Images section and how to spend starter tokens wisely.
Webhooks in the Avatar Video section: get a notification on your server when a video is ready
Webhooks are a tab in the Avatar Video section: the moment a render finishes, a POST request goes to your server, so you never have to poll. What the signing secret is, how to register an HTTPS endpoint and how to verify the signature.
Z-Image: fast image generation and what it is good for
Z-Image is the lightest way to make an image on NeuralSpace: 0.7 tokens per picture, text-only prompts, five aspect ratios and JPEG/PNG. Here is what this model handles well and when to pick another.
Kaiming He's team shows an AI pretrained on cat photos solves ARC almost as well as LLMs
In a new ECCV 2026 paper, a vision model with no text at all — pretrained on ordinary ImageNet photos — scores 63.4% pass@2 on ARC-1, and 70.2% as an ensemble. A specialized 8B-parameter LLM system scores 71.6%.
Media library in the Avatar Video section: store files once, reuse them in every video
The media library is a tab of the Avatar Video section that stores your uploaded images, video and audio: a file drops into generation or translation with one click. What you can upload, how to start and what it costs.