NewUnified .char models are now supported in ComfyUI
COMPARE/compare/char-vs-refmod

RefMod vs .char: which to use for consistent characters

Use RefMod if you only generate with MiniMax H3 and want the smallest, fastest setup with no training. Use .char if you want one character file that works across several image and video models, carries body, wardrobe and voice, and scores every take for drift.

TL;DR

ROWRefMod.char
Training neededNoNone on FLUX.2 and H3 R2V. An adapter trained once on Krea 2 and other H3 nodes
Photos requiredNot stated1 to 2
Models supportedMiniMax H3 onlyFLUX.2, Krea 2, FLUX.1, Z-Image Turbo, MiniMax H3, LTX-2.5, plus Nano Banana Pro and H3 R2V via fal
Portable fileYes, about 1 to 1.6 MB, for H3Yes, one .char file
Face / body / wardrobe / voiceFace and look, no voiceFace, body, wardrobe and an optional voice clip
Runs locallyYes, in ComfyUIYes, in the OmniChar app and ComfyUI, or hosted on OmniChar Cloud
Open sourceNot statedYes, GPL-3.0
CostFree on your own GPUFree to run locally. OmniChar Cloud uses credits
Drift measurementNoneContinuity score out of 100 on every take

What each one is

THE ALTERNATIVE

RefMod

RefMod is a community project by malcolmrey that stores a character as a small file of about 1 to 1.6 MB for MiniMax H3. It needs no training and loads quickly. It works with H3 only and is labelled experimental.

OMNICHAR

.char

A .char is one portable file that holds reference photos, an identity fingerprint, a locked description, an optional trained adapter and an optional voice clip. It works in the OmniChar app, the ComfyUI node and OmniChar Cloud. FLUX.2 and H3 R2V use the references directly; Krea 2 and other H3 nodes use an adapter trained once. Every take gets a continuity score out of 100.

Strengths

Where RefMod wins

  • Tiny: about 1 to 1.6 MB
  • Fast to set up, with no training step
  • The simplest option if MiniMax H3 is the only model you use

Where .char wins

  • Works across image and video models, not only H3
  • Carries a trained adapter and an optional voice clip in the same file
  • Keeps body and wardrobe consistent, not only the face
  • Scores every take out of 100 so you can see drift

When to use which

Use RefMod ifyour whole pipeline is MiniMax H3 and you are fine relying on an experimental tool.
Use .char ifthe same character needs to appear in FLUX.2 stills and LTX-2.5 or H3 video, or a voice clip has to travel with it.

Questions

Is RefMod better than .char?

It depends on your models. RefMod is smaller and faster to set up, but it works with MiniMax H3 only. A .char works across several image and video models.

Who made RefMod?

RefMod is by malcolmrey. It is described as experimental.

Does .char need training?

Not on FLUX.2 or H3 R2V, which use the reference photos directly. Krea 2 and other H3 nodes use an adapter that is trained once and stored inside the .char.

Can I measure drift?

A .char gives every take a continuity score out of 100, so you can see when a take moves away from the character.

Is .char open source?

Yes. OmniChar is GPL-3.0 and runs in the OmniChar app, the ComfyUI node and OmniChar Cloud.

Build a character once and use it in the OmniChar app, ComfyUI and OmniChar Cloud