Seedance 2.5 Introduces Draft Mode to Reduce AI Video Generation Costs

ME News report, September 22 (UTC+8), Beating AI Bulletin: ByteDance has launched the Draft mode for Seedance 2.5 API. Users can first generate low-cost 480P draft clips to confirm composition, motion, dialogue, and camera movement; once satisfied, they can generate the selected draft into a 1080P final video. Ji Meng has also integrated into “Seedance 2.5 (Draft Mode).” This differs from the conventional approach of testing with a low-resolution prompt and then regenerating a high-resolution version using the same prompt. When generating the 1080P version, the model references the original draft’s ID and reuses parameters such as the prompt, reference materials, seed, aspect ratio, and duration to re-infer the output, preserving the original structure, composition, and motion as closely as possible. Local textures, small text, and crowd counts may still vary—meaning it is neither frame-accurate replication nor a simple upscaling of 480P to 1080P. According to official estimates, for a 5-second 1080P video without video input and with an average of four attempts needed to select a satisfactory result, Draft reduces costs by 57% and nearly doubles overall generation speed. If twenty attempts are required to select one result, cost savings can reach up to 77%. A similar Draft feature previously appeared in Seedance 1.5 Pro; this update integrates this low-cost iterative workflow into Seedance 2.5 and enables selected drafts to be further rendered into corresponding 1080P final videos. The most expensive aspect of AI video generation is typically the repeated “gacha” sampling phase—and this update directly targets those costs. (Source: BlockBeats)