All local H3 guides
Checkpoints / Local H3

MiniMax H3 FL2VA vs Ref2VA

Choose between MiniMax H3 FL2VA and Ref2VA for text-to-video, image-to-video, first/last frames, video-to-video, character references, and voice references.

Reviewed
2026-08-09
Reading time
5 min
Evidence
3 official / 0 community
Short answer

Use FL2VA when the request has text only or needs a first frame, last frame, or both. Use Ref2VA when images, videos, or audio act as references for identity, style, motion, camera, or voice. T2V and I2V are FL2VA tasks. V2V is a Ref2VA use case.

01

The decision table

Your inputCheckpointTask
Prompt onlyFL2VAT2VA
Opening frameFL2VAFL2VA
Ending frameFL2VAFL2VA
Opening and ending framesFL2VAFL2VA
Character or product reference imagesRef2VARef2VA
Motion or camera reference videoRef2VARef2VA / V2V
Voice or sound reference plus visual inputRef2VARef2VA
02

A keyframe is not the same as a reference

A first or last frame owns a temporal endpoint. H3 must begin at that composition, end at it, or build the motion between two supplied states. A reference image instead contributes identity, object, setting, palette, or another declared property without necessarily appearing as frame zero.

03

Ref2VA input limits

More references are not automatically better. One identity owner, one environment owner, and one motion or voice owner are easier to diagnose than several files competing for the same property.

  • Up to 9 images.
  • Up to 3 video clips, each 2 to 15 seconds, with no more than 15 seconds total video.
  • Up to 3 audio clips, each 2 to 15 seconds, with no more than 15 seconds total audio.
  • Audio cannot be the only reference type.
  • No more than 12 files across all reference types.
04

Deployment consequences

The checkpoint families have separate transformer weights. In SGLang, serve FL2VA and Ref2VA with different model-variant values and usually different ports. In ComfyUI, the T2V and I2V templates load the FL2VA diffusion model while the R2V template loads the Ref2VA model.

  • Download only the family needed by a dedicated service.
  • Keep route names aligned with the loaded variant.
  • Reject reference requests at an FL2VA endpoint instead of silently ignoring materials.
  • Test every conditioning path with an input whose effect is easy to verify.
Frequently asked questions
Is MiniMax H3 T2V FL2VA or Ref2VA?

Text-to-video runs on the FL2VA checkpoint family with no image condition.

Which MiniMax H3 checkpoint supports image-to-video?

FL2VA supports optional first and last keyframes. Ref2VA supports images as semantic references. Choose based on whether the image is a timeline endpoint or a reusable reference.

Which checkpoint supports video-to-video?

Use Ref2VA and provide a video or video-with-audio reference. SGLang documents V2V as a Ref2VA use case rather than a separate task value.

Do FL2VA and Ref2VA have different quality levels?

They are task-specific conditioning checkpoints, not quality tiers. The right choice depends on the input relationship.

Sources and evidence

Read the original material

Community figures describe the named machine and workflow only. They are not MiniMax performance guarantees.

  1. Official documentationMiniMax H3 model variants and input specifications
  2. Official documentationSGLang H3 task and variant mapping
  3. Official documentationComfyUI H3 templates