Minimax H3 · Ref2VA Character / Head Swap

Full Custom · Guide LoRA · Add Guide motion

Freedom / Unlocked — Strict control for power users

Swap a performer using one reference picture + one guide video.

• Character mode — full identity (face, body, outfit, style) from Picture 1

• Head mode — head / face / hair / accessories only; body & clothes stay from the video

Motion, camera, timing, and scene energy follow the guide video (Add Guide — not raw ref_videos).

Output 24fps. Tune duration, megapixels, and steps for credit vs quality.

━━━━━━━━━━━━━━━━━━━━

HOW TO USE

1. Upload Picture 1 (identity) + Video 1 (motion)

2. Set LoRA + Prompt mode to the SAME value:

1 = Character · 2 = Head

3. Duration must match the guide video length (see table)

4. Describe (optional): extra face/character notes, or leave default

DURATION (24fps frame counts)

2s

5s

6s

8s

10s

12s

15s

MEGAPIXELS

low 0.4 — lighter / lower cost (recommended start on shared VRAM)

medium 0.6 — balanced

high 0.9 — sharper · higher credit use

Aspect ratio: pick the closest preset to your guide video.

STEPS

low 4 — faster / lighter

medium 6 — balanced

high 8 — more refine · slower

Keep steps modest; start low if jobs fail on Tensor.

DESCRIBE

Character: Match the full character identity from Picture 1: face, body, clothing, accessories and style.

Head: Match the face, hair, head accessories and head identity from Picture 1 as closely as possible.

━━━━━━━━━━━━━━━━━━━━

Research & responsible use

This tool is a general-purpose AI generation and research utility for experimentation, testing, and creative work.

Use it in line with applicable laws. Do not generate or share illegal, harmful, or rights-infringing content. You are responsible for your prompts, uploads, and how outputs are used.

By using this tool, you accept its experimental nature and agree to use it responsibly.