Update app.py

Update app.py

Accept the 2 s durations the demos offer

Tidy comments and docstrings

Log which ZeroGPU identity each request's booking is charged to

Book the rewrite and the encode separately so each fits ZeroGPU's per-booking ceiling

Size the GPU booking to the work: an xlarge booking is priced 2x and the demos call this Space anonymously

Migrate to canonical diffusers PR 14371 and the public MiniMaxAI/MiniMax-H3 weights

Optional Context-IR prompt rewriting

Group canvases by ratio, fast to full

Fast canvases for 4:3, 3:4 and 21:9

Generalize as a Qwen3-VL conditioning service

Canvas label rename (wire contract)

The reference half moved to minimax-h3-reference

Add 544x544 square fast canvas (wire contract with generator)

Add the ref2va conditioner half and its /encode_ref2va endpoint

Match the generator canvas table (labels are forwarded to /encode)

Sync the split MiniMax-H3 Spaces

Sync the split MiniMax-H3 Spaces