Model01
MiniMax H3 ships as an open, omni-reference video model
MiniMax launched H3 with open weights, unified text, image, video and audio context, native stereo sound and output up to 2K. An open model that edits characters, scenes, dialogue and voice in one context turns a hosted feature into infrastructure other tools can build around. Studios should benchmark identity, lip-sync and audio continuity together, then keep the winning workflow portable between local ComfyUI graphs and hosted APIs.
Source MiniMax @MiniMax_AI · 2026.07.31 ↗
Industry02
Luma adds Seedance 2.5 to its incoming model shelf
Luma confirmed Seedance 2.5 was coming to its platform, following similar announcements from Runway, Pika and Artlist. The same frontier model appearing across competing interfaces means model access is commoditizing while workflow, asset memory and review become the differentiators. Teams should choose the host by project organization and editability, not by assuming one platform has exclusive model quality.
Source Luma AI @LumaLabsAI · 2026.07.31 ↗
