Important AI drops, source-backed.
A daily ranked feed of model releases, API changes, platform launches, and research notes from the last 30 days.
Verified fallback · Last verified: June 6, 2026
ChatGPT Images 2.0 is the current OpenAI image-generation reference point
OpenAI announced ChatGPT Images 2.0 on April 21, 2026, with improved precision, control, multilingual text rendering, and visual reasoning.
▸ The official announcement names the product ChatGPT Images 2.0.
▸ OpenAI positions the release around precision, control, visual reasoning, and stronger text-in-image output.
▸ The OpenAI model catalog lists GPT Image 2 as a specialized image generation and editing model.
OpenAI system card: Images 2.0 thinking mode can use live web search data
The deployment safety card says thinking mode adds reasoning and tool use to image generation, including live web search data integration.
▸ The system card describes enhanced world knowledge, instruction following, and dense text generation.
▸ It says thinking mode can integrate live web search data into the image-generation process.
▸ It also documents extra safeguards for realism and sensitive imagery risks.
Nano Banana Pro is Gemini 3 Pro Image, not a generic "best image model" claim
Google introduced Nano Banana Pro on November 20, 2025 as an image generation and editing model built on Gemini 3 Pro.
▸ Google identifies Nano Banana Pro as Gemini 3 Pro Image.
▸ Google says it supports more accurate text rendering, stronger world knowledge, and creative controls.
▸ Google says the model is available across products including Gemini, Google Ads, Workspace, AI Studio, and Vertex AI surfaces.
Google developer access for Nano Banana Pro started as a paid preview
Google says Gemini 3 Pro Image started rolling out for developers through the Gemini API in AI Studio and Vertex AI.
▸ Google describes developer access as a paid preview rollout.
▸ The announcement ties the model to Gemini API, Google AI Studio, and Vertex AI.
▸ Google says grounding with Google Search can retrieve data based on the prompt.
Seedance 1.0 is the verified ByteDance video model in public docs
ByteDance lists Seedance 1.0 as a text-to-video and image-to-video model with multi-shot generation and 1080p output.
▸ ByteDance says Seedance 1.0 supports multi-shot video generation from text and image.
▸ The official model page says it can create 1080p videos with smooth motion and cinematic aesthetics.
▸ The page describes native multi-shot storytelling, style consistency, and prompt-following controls.
Seedance 1.0 technical report gives one concrete performance number
The arXiv technical report says Seedance 1.0 generated a 5-second 1080p video in 41.4 seconds on an NVIDIA L20.
▸ The report describes text-to-video and image-to-video training in one system.
▸ It reports about 10x inference speedup through distillation and system optimization.
▸ The specific 41.4-second result is for a 5-second 1080p video on NVIDIA L20 hardware.
Train the model watchlist.
Like or dislike updates so Kreatrix learns which sources, categories, and model drops matter to you.