MiniMax H3 Evening Report 2026-11-16 | First Recommendation: MiniMax H3 ref2va/fl2va Fast Generation Workflow

This issue reviews 5 new H3 versions: 1 rated A, 3 rated B, and 1 rated C. The top recommendation is SillyBilly101's ref2va/fl2va fast generation workflow v3.0: it runs both reference-to-video and first-last frame modes with a single image set, automatically assembles H3 segmented prompts, and crops overly long references. This version changes the logic to check duration before counting, and disabled references no longer occupy any of the 12 slots. Speed is identical to v2. KinetiX's combat action LoRA KOF V2 has the fastest downloads but all examples are hidden and the description only contains external links, so it is rated B; the Colored Pencil Style LoRA and Pure Black Material Slider have clear parameters but are both paid, so they are rated B; Kinetic_Sakuga, which only has a single word in its description, is rated C.

Zhenzhen's AI Radar · MiniMax H3 · 2026-11-16 Evening Report

This issue features 5 new versions: Strongly Recommended 0 · Recommended 1 · Worth Watching 3 · Listed Only 1

Editor's Note This issue reviews 5 new H3 versions: 1 rated A, 3 rated B, and 1 rated C. The top recommendation is SillyBilly101's ref2va/fl2va fast generation workflow v3.0: it runs both reference-to-video and first-last frame modes with a single image set, automatically assembles H3 segmented prompts, and crops overly long references. This version changes the logic to check duration before counting, and disabled references no longer occupy any of the 12 slots. Speed is identical to v2. KinetiX's combat action LoRA KOF V2 has the fastest downloads but all examples are hidden and the description only contains external links, so it is rated B; the Colored Pencil Style LoRA and Pure Black Material Slider have clear parameters but are both paid, so they are rated B; Kinetic_Sakuga, which only has a single word in its description, is rated C.

Recommended (1)

[A] MiniMax H3 ref2va/fl2va fast generation workflow · v3.0

Workflow · Version Update · Author SillyBilly101 · Published 11-16 20:55 · Downloads 2

A ComfyUI workflow for fast H3 video generation designed for 12 GB VRAM GPUs (the author uses 3060): it switches between ref2va (reference-to-video) and fl2va (first-last frame) modes within the same image, provides separate input boxes for each segment of the H3 prompt with automatic assembly, and automatically checks and crops video/audio references.

Images

Video 1184x1760

Reasons for Recommendation

  • Version Update: The author states in the description that v3.0 does not change speed, only making two changes—reference checking now first verifies if the total duration of video or audio exceeds the limit before counting the quantity, and disabled references no longer occupy any of the 12 reference slots; the layout has been rearranged to make it easier to replace prompt sections or acceleration methods with those from other workflows.
  • Serves low-VRAM users for reference-to-video, first-last frame, and video editing: the author lists time comparisons for 5 second and 10 second image-to-video generation at 0.5 / 1.0 / 2.0 MP. Since v2, it supports reference cropping and scaling, automatically truncates references longer than 15s, and offers an optional slower refine step to improve visuals around fast-moving objects.
  • Includes 4 example videos; free, with a license allowing output, rental, and sale. The model page has accumulated 964 downloads.

How to Use

  • Import H3_FastGen_V3.json; load the fl2va model and turn on the “use fl2va?” switch for first-last frame mode. Turn it off and load the ref2va model for reference-to-video mode.
  • Requires the 4 Step LoRA (placement location marked in the image), as well as node packs like Impact Pack, KJ Nodes, and Dynamic Prompts. See the model page for the complete list.

Risks and Limitations

  • The version notes section is empty; changes are only described in the general description. This tool has not been run by us; it has many node dependencies, and the MiniMax H3 Audio T8 node requires manual installation.
  • Timing data comes only from the author's single 3060 setup; 7 examples are hidden.

Worth Watching (3)

[B] KinetiX Combat Semantic Bridge (MiniMax-H3) · KOF V2

LoRA · Version Update · Author make676767 · Published 11-16 15:48 · Downloads 75

The second version of the KOF (King of Fighters) edition of the KinetiX combat action LoRA: the title of the author's usage guide article claims it brings the physical movements of real combat into H3 generation. This version includes a companion workflow resource pack.

Images

Illustration 1448x1086

Reasons for Recommendation

  • Version Update: The same page previously featured v1.0, v1.1 (expanded training set), and KOF V1.0. The author did not specify the changes in KOF V2.
  • Serves combat and fight action videos; it received 75 downloads and 7 likes within 5 hours of release. The download speed is higher than that of H3 models over the past 7 days 96%, indicating high attention.
  • Free, with a license allowing output, rental, and sale, provided attribution is given.

How to Use

  • Load kinetix_combat_bridge_kof_v2.safetensors and use it with the workflow in the same-version workflow_resources zip package.
  • The author did not provide trigger words, strength settings, or prompt writing guidelines on the model page, only linking to an X post and a Civitai article.

Risks and Limitations

  • Visible examples are 0 (6 hidden), making it impossible to preview the effect; the author did not write update notes.
  • The model page body contains only external links; usage instructions are entirely in off-site articles and Discord.
  • The card image is an AI-generated illustration, not the actual model output; the author has no publicly available standard-level examples.

[B] ColoredPencil [Minimax / MOSS / Krea2 / SD] · Minimax v1.0

LoRA · Port · Author TeraWatt · Published 11-16 11:32 · Downloads 2 · Paid

The H3 version of the Colored Pencil hand-drawn style LoRA: it transforms images into colored pencil illustrations with characteristic strokes, layered colors, and paper texture. The trigger word is ColoredPencil_TeraWatt.

Images

Video 1216x1792

Reasons for Recommendation

  • Port: The same page already has MOSS Image, Krea 2, and SD series versions. The H3 version is the video version of the same style; the author states that all example videos were generated using text-to-video mode.
  • Serves hand-drawn style short videos for picture books, children's illustrations, characters, animals, and plant encyclopedias.
  • Includes 7 example videos for preview.

How to Use

  • Load the LoRA (BF16) and include the trigger word ColoredPencil_TeraWatt in the prompt.
  • The author did not provide strength or step count settings.

Risks and Limitations

  • Paid download (permanent payment); the license does not include sale rights.
  • 12 examples are hidden; the author did not describe the differences between the H3 version and the image version.

Trigger Word: ColoredPencil_TeraWatt

[B] Pure Color Purifier Slider · Jet Black v1.0

LoRA · New Model · Author make676767 · Published 11-16 09:23 · Downloads 2 · Paid

KinetiX's Pure Black Material Slider LoRA: positive values compress black clothing, tactical gear, and carbon fiber armor into light-absorbing matte jet black, resisting greenish or grayish tints caused by neon and other colored lights; negative values conversely create color-shifted gray fading.

Images

Video 960x544

Reasons for Recommendation

  • Solves the problem of black clothing being tinted and shadows turning gray under strong colored lighting; the author claims it simultaneously preserves faces, skin tones, and background lighting.
  • Full strength range provided: +1.0 to +1.8 is the author's recommended range, +3.0 to +5.0 is for extreme jet black, and -1.5 to -3.0 is for color-shifted gray fading.
  • Includes 10 example videos, along with positive and negative prompt snippets in both Chinese and English, and a same-seed A/B test template.

How to Use

  • No trigger word. Load KinetiX_mat_jetblack_purifier_h3_v1.safetensors and control the direction using strength; can be combined with the positive prompt snippets provided by the author.
  • It is recommended to run comparisons with a fixed seed at -1.5 / 0.0 / +1.5 / +3.0; the author did not provide step count settings.

Risks and Limitations

  • Paid download; it received 2 downloads within 12 hours of release, with no user feedback yet.
  • Only targets black materials, so its use case is narrow.

Listed Only (1)

  • Kinetic_Sakuga · v1.0 — LoRA: An H3 style LoRA named Kinetic_Sakuga. The description contains only the single word “Sakuga,” suggesting from the name that it is an action expression style based on Japanese animation sakuga (high-quality animation) techniques. Video 864x480

Data: Civitai Public API (Base Model MiniMax H3), collection window 11-16 08:31 → 11-16 21:01 (Beijing Time). Reasons for recommendation have been reviewed by editors; download counts only reflect attention levels.

Civitai

Explore the latest AI Radar updates