CoreLesson 735 min

Regeneration: The 768P-to-2K Pipeline

By the end of this lesson you can
  • Choose between the source_task_id and content+base_video regeneration methods, and explain why sending both, or neither, fails
  • Validate a candidate base_video against all five hard gates before spending anything
  • Resend the correct prompt when the 768P source was generated through Context-IR

The prices from h3-f2 make the argument. Generating directly at 2K costs $0.13/second. Generating at 768P costs $0.08/second, and regenerating a finished 768P clip to 2K costs $0.05/second of regenerated output. Per clip, $0.08 + $0.05 equals $0.13 — parity. The saving is in iteration: five 8-second drafts at 768P run $3.20, and mastering only the keeper adds $0.40 — $3.60 total, against $5.20 for the same five drafts generated straight to 2K. Draft cheap, master once.

The module doing the mastering, H3-Regenerate-2K, is not a separate super-resolution network — it uses H3-Base itself to regenerate the clip at 2K, in context. It is also one of the two modules that is not open-sourced; MiniMax's announcement says "this module is not yet open-sourced. We will release it once it is ready." And note the ceiling: regeneration currently supports only 2K as the target resolution.

Core — locked

Unlock “Regeneration: The 768P-to-2K Pipeline”

The 4-lesson Foundation module is free and always will be — you have already read the part most guides get wrong. The paid lessons go deeper, with labs, knowledge checks, and every technical claim verified against live documentation.

MiniMax H3 — Complete Course
The whole course, one purchase
$129one-time

Already bought it? Sign in with the email you used at checkout.