MiniMax H3 Morning Brief 2026-11-19 | Top Pick: Slap (and Self Slap)

This issue reviews 3 new H3 versions, with 2 rated A and 1 rated C. Top pick: The H3 port of the slap action LoRA, where a character slaps themselves or is slapped by an off-screen hand. Both trigger phrases and strength ranges are provided; the author claims it offers better consistency, sound, and reaction to being hit compared to the base model. LongForge v7.9 allows a single long reference video to be used across multiple scenes and uses the corresponding audio segments from the original video as reference, rated A. Lora-Daddy H3 Studio 4.3 has only a one-sentence change description and all examples are hidden, rated C.

Zhenzhen's AI Radar · MiniMax H3 · 2026-11-19 Morning Brief

This issue features 3 new versions: Strongly Recommended 0 · Recommended 2 · Worth Watching 0 · Listed Only 1

Editor's Note This issue reviews 3 new H3 versions, with 2 rated A and 1 rated C. Top pick: The H3 port of the slap action LoRA, where a character slaps themselves or is slapped by an off-screen hand. Both trigger phrases and strength ranges are provided; the author claims it offers better consistency, sound, and reaction to being hit compared to the base model. LongForge v7.9 allows a single long reference video to be used across multiple scenes and uses the corresponding audio segments from the original video as reference, rated A. Lora-Daddy H3 Studio 4.3 has only a one-sentence change description and all examples are hidden, rated C.

Recommended (2)

[A] Slap (and Self Slap) · MiniMax H3 V1

LoRA · Port · Author thegipper · Published 11-18 21:37 · Downloads 440

The H3 version of the slap action LoRA: enables a character to slap themselves or be slapped by a hand reaching in from off-screen; the author claims it primarily improves action consistency, sound, and reactions after being hit.

You can use the same prompt to compare with and without the LoRA, focusing on listening to the impact sound and observing the reaction after being hit.

Images

Video 1232x1232

Reasons for Recommendation

  • Ported from Wan 2.2 (I2V / T2V, each version split into high-noise and low-noise files) to H3, where the H3 version consists of a single file; the author states that while the base model can sometimes perform this action, adding the LoRA improves consistency, sound, and reactions
  • Suitable for short drama shots requiring slap actions and sound effects, such as emotional conflicts or comedic scenes
  • Both trigger phrases and strength settings are provided, 2 example videos are visible, it is free, and the license permits commercial sale

How to Use

  • Write prompts in natural language, using the trigger phrases "She slaps herself in the face." or "She is slapped in the face by a hand from offscreen"
  • Start with a strength of 0.8, and lower it to 0.5–0.6 when there are many actions in the scene

Risks and Limitations

  • 4 sensitive-level examples are hidden
  • Both trigger phrases use "She" as the subject; the author has not specified performance when switching to male subjects or multiple people
  • The H3 version description is only three sentences long, and the author has not provided step counts

Trigger Words: She slaps herself in the face., She is slapped in the face by a hand from offscreen

[A] MiniMax H3 LongForge FL2VA & REF2VA (LONG VIDEO) · LongForge v7.9

Workflow · Version Update · Author AIBOB1199 · Published 11-19 08:37 · Downloads 13

ComfyUI workflow v7.9 for stitching H3 long videos by scene and exporting finished films with audio: Adds "Follow scenes" for REF2VA, allowing a single long reference video to be used across multiple scenes, automatically extracting the corresponding segment for each scene.

Images

Video 1536x2688

Reasons for Recommendation

  • New capability in this version: Reference videos can exceed 15 seconds, with each scene extracting its corresponding segment based on its position in the timeline (single-scene reference limit is 362 frames); "Paired audio" synchronously extracts the corresponding audio segments from the original video as reference
  • "From reference" automatically calculates width and height based on target pixel count and the aspect ratio of the first reference video or image (rounded to multiples of 32), handling mobile-shot videos according to rotation metadata; "Check mapping" allows viewing which reference segment each card uses before generation
  • Suitable for needs where a single original clip drives multi-scene short films, such as action replication or dance covers; 1 example videos and 2 images are visible, it is free, and documentation is available in Russian and English

How to Use

  • Copy the node pack folder into ComfyUI/custom_nodes, install requirements.txt using ComfyUI's built-in Python, restart, and open H3_LongForge_FL2VA_REF2VA_v7.9.json, selecting the H3 model, text encoder, and two VAEs
  • Add videos to Reference Media, select "Follow scenes · REF2VA" for Video use, and set End = 0 to follow through to the end of the original video; to use the original audio, select "Paired audio" and enable Reference audio
  • Maintain aspect ratio from the original: Set Resolution to Megapixels and Aspect ratio to From reference
  • Use GENERATE NEXT → ONE SCENE → Run to generate scene by scene, then EXPORT FILM → Run to composite the final film with audio

Risks and Limitations

  • The author has not written update notes; changes are only documented in the model description
  • Requires ComfyUI supporting native H3 and V3 node APIs, as well as FFmpeg; the complete template also depends on external nodes such as KJNodes, Veda, and NVIDIA RTX VFX
  • References only guide H3; final audio is regenerated by the model rather than directly copied from the original track
  • The license permits image output and in-site rental only, not commercial sale

Listed Only (1)


Data: Civitai public API (base model MiniMax H3), inclusion window 11-18 20:31 → 11-19 09:01 (Beijing Time). Recommendation reasons have been reviewed by editors; download counts reflect attention levels only.

Civitai

Explore the latest AI Radar updates