Zhenzhen's AI Radar · MiniMax H3 · 2026-11-21 Evening Report
This issue features 4 new versions: Strongly Recommended 1 · Recommended 0 · Worth Watching 2 · Listed Only 1
Editor's Note This issue reviews 4 new H3 versions: 1 S, 2 B, and 1 C, with no exclusions. Top pick: KinetiX Ground Impact Bounce Slider. Positive values make hit characters bounce back into mid-air like pinballs after hitting the ground or walls; negative values result in heavy landings with zero rebound. No trigger words required. Intensity tiers and bilingual prompt templates are fully documented. The Daily Grind workflow v1.2 only fixed an error when disabling preview videos, downgraded to B; the mechanical transformation LoRA Transform Delux has clear usage but is paid, listed as B; the KinetiX Combat Semantic Bridge Smash version is paid and lacks version notes, listed as C.
Strongly Recommended (1)
[S] KinetiX Dynamic Impact Ricochet Bounce Slider · v1.0
LoRA · New Model · Author make676767 · Published 11-21 12:48 · Downloads 79
A "Ground/Wall Impact Bounce" slider LoRA for combat actions: Positive values make knocked-back characters bounce high into mid-air after slamming into the ground or walls, facilitating aerial follow-ups; negative values cause characters to slam flat with zero rebound.
Suitable for a three-panel comparison using the same prompt at -3 / 0 / +2.5 to observe the difference between bouncing and dead stops.
Images

Reasons for Recommendation
- New capability: Solves the "slam and lie flat" issue in H3 action shots. Adjustable positive or negative values allow the same LoRA to create both Smash Bros.-style pinball bounces and heavy, dead-stop landings.
- The author provides complete intensity tiers: +0.8 to +1.5 for mild enhancement, +1.8 to +3.0 as the recommended range, +3.5 to +5.0 for the most exaggerated effect, and -2.0 to -5.0 for dead stops with no rebound.
- Includes directly pasteable English phrases for both positive and negative effects, along with a full production template featuring shot breakdowns, ambient sound, and music fields (recommended weight +2.5).
- 2 sample videos are visible; free to use, with a license allowing image generation, rental, and sales. 79 downloads within 8 hours of release.
How to Use
- No trigger words; start with an intensity of +2.5 after loading the LoRA. Increase to +3.5 or higher for more exaggerated effects, or set to -2.0 to -5.0 for non-bouncing landings.
- Add the author's positive phrase (high-elasticity ground bounce ricochet, body rebounding sharply back into mid-air…) or negative phrase (dead heavy thud stopping flat on ground…) to your prompt.
- Sampling steps and CFG are not specified by the author; use the default values from the H3 workflow.
Risks and Limitations
- Only 2 sample videos are visible; 3 sensitive-level examples are hidden. The author claims it does not distort visuals or character appearance, but no comparison videos are provided.
- Effects are targeted at hit-landing scenarios in combat and action scenes; not useful for daily life shots. The model card begins with the author's X and Discord promotions.
Worth Watching (2)
[B] The Daily Grind - Minimax H3 Workflow · v1.2
Workflow · Version Update · Author darkroast175696 · Published 11-21 10:14 · Downloads 58
The Daily Grind is a ComfyUI workflow that combines common H3 tasks like text-to-video and reference-to-video into a single graph. Version v1.2 fixes an issue where disabling the "Preview Video" function caused workflow errors.
(Editor downgraded to B: This version is only a bug fix with no new features.)
Images

Reasons for Recommendation
- Version update: v1.2 only fixes bugs—disabling Preview Video no longer interrupts the workflow, and some nodes have been rearranged. The generation process itself remains unchanged.
- Tool utility: Common settings are centralized on one screen, eliminating the need to enter subgraphs to change parameters. Switching between text-to-video and reference-to-video does not require changing templates. It facilitates retrieving seeds, prompts, and other reproduction data.
- Optional two-stage rendering: Place action LoRAs and style/detail LoRAs in different stages to avoid blurry actions or style LoRAs disrupting audio tracks. The workflow includes scene suggestions.
- Free, with 4 sample images. The model has accumulated 247 downloads.
How to Use
- Download h3-daily-driver-v1-2.zip and import it into ComfyUI. Nodes displayed in red either belong to toggleable groups or are mandatory installations.
- For first-time H3 users, download the recommended model files via the links in the model card. Parameters like steps should follow the workflow's default values, as the author did not specify them in the model card.
Risks and Limitations
- This is a minor update. Users already on v1.1 who do not disable preview videos do not need to update.
- Examples are images, with no video samples. Dependent nodes and models must be verified by the user; this tool has not been run by us. It is a workflow JSON, not new weights.
[B] Transform Delux MMH3 · Transform Delux MMH3
LoRA · New Model · Author riiahworld · Published 11-21 15:28 · Downloads 2 · Paid
A mechanical transformation VFX LoRA: Causes the outer shells of characters or robots to split into plates, revealing glowing circuits and mechanical structures. Arms can reassemble into weapons and then revert to their original form.
(Editor upgraded to B: Usage is clearly written and the direction is clear, but it requires payment.)
Images

Reasons for Recommendation
- New model: Serves text-to-video shots for sci-fi transformations and mecha deployments. The author provides trigger words, a complete example prompt, and a chronological prompt structure.
- Prompt structure: Trigger word → Subject and initial state → Sequential transformation actions → Revealed internal structure → Final form or reversion → Camera and scene → Optional sound effects.
- 2 sample videos are visible.
How to Use
- Place the trigger word vfx_transform at the beginning of the prompt, then describe the transformation process chronologically. The author specifically notes that r3tr0sks is the trigger word for another Z-Image-Turbo LoRA and should not be mixed.
- Intensity and steps are not specified by the author.
Risks and Limitations
- Paid download (permanent payment). Only 2 downloads within 6 hours of release.
- Intensity is not specified; 2 sensitive-level examples are hidden.
Trigger Word: vfx_transform
Listed Only (1)
- KinetiX Combat Semantic Bridge (MiniMax-H3) · Smash v1.0 — LoRA: The new Smash version of the KinetiX Combat Semantic Bridge series (which integrates fighting game hit action semantics into H3). The author did not specify what this version does.

Data: Civitai Public API (Base Model MiniMax H3), collection window 11-21 08:31 → 11-21 21:01 (Beijing Time). Recommendation reasons are editor-reviewed; download counts only reflect attention levels.