dreamina

Video queue: _vip = fast queue

Video generation has two queues, gated by model tier:

• _vip models (seedance2.0_vip, seedance2.0fast_vip) -> fast queue / priority: reach queue_status: Generating in ~1-2 min, finish in a few min.
• non-vip models (seedance2.0, seedance2.0fast, seedance2.0mini) -> slow queue / free: can sit at queue_status: Queueing for 30-40+ min, sometimes effectively stuck.

Rule: any video that must land promptly -> ALWAYS use a _vip model. Non-vip only for "don't care when it finishes". _vip also costs more and is the only tier reaching 1080p/4K. There is NO CLI cancel -> a slow-queue task, once submitted, can't be aborted (only ignored).

Report credits after every generation

Every generator call -> after it returns, report credit spend as a per-step line:

第N步 <cmd>: 消耗 <credit_count> credits, 余额 <total_credit>

credit_count -> from the task result JSON. Balance -> dreamina user_credit (total_credit). Multiple calls in a turn -> one line each, plus a total-consumed summary at the end.