AI Daily — August 3, 2026
Models & Research
ByteDance Doubles Seedance Clip Length to 30 Seconds — ByteDance has launched Seedance 2.5, its latest audio-video generation model. Single-pass output goes from 15 to 30 seconds, with multi-round extension for multi-minute sequences. Reference input expands to 30 images, 10 video clips, and 10 audio clips per pass. The release adds timestamp-level editing for targeted changes after generation, plus upgrades to green screen and clay render referencing. ByteDance also points to industrial uses including synthetic training data for robotics and long-tail simulation for autonomous driving. The company acknowledges remaining weaknesses in complex motion physics and multi-subject interaction. Seedance 2.5 is live on Jimeng AI and Doubao Pro, with API access coming later. ByteDance ↗
My takeaway: I think a 30-seconds single pass puts a full commercial spot within reach of one generation. The previous Seedance 2.0 showed superior performance and ranked at the top almost all time. In the comparison videos I checked, characters in 2.5 looked more natural to me than in 2.0. According to the other video I checked, this 30-seconds generation was not cheap (about $9.8 CAD)
MiniMax Launches H3 With 2K Video and Native Stereo Audio — MiniMax Ships Open-Weights H3 With 2K Video and Native Stereo Audio — MiniMax has launched H3, a general-purpose multimodal generation model that takes text, images, video, and audio as unified context. Output is video with native stereo sound, up to 15 seconds at 2K, offered as the default rather than an upscale tier. The company says reference and editing tasks run through natural language instead of separate task-specific models. MiniMax also claims its per-second price at 2K is under a third of mainstream models. Weights are now public on Hugging Face, and ComfyUI shipped day-zero support. MiniMax ↗ ComfyUI ↗
My takeaway: MiniMax H3 is an open-weights model, but that doesn't mean the quality is low. On the Artificial Analysis, it ranks in the top three on the text-to-video, image-to-video, and video editing leaderboards. The more surprising part is local inference. ComfyUI says pruning and int8 quantization cut the memory footprint by 66 percent, from 123.6GB to 42.5GB, and that dynamic VRAM offloading lets a 2K video model run on a GPU like the RTX 3060.
Summaries are AI-generated and may contain errors — always verify against the linked original. Each story links to its source, which holds the copyright. Outlet names are shown for attribution only and do not imply any endorsement or affiliation.
Disclaimer: The views expressed in My Takeaway are my own personal opinions and general observations on industry trends. They are not intended to criticize, disparage, or make factual claims about any specific company, product, or platform. Any platform names mentioned are referenced solely for illustrative and informational purposes.