Ashmotv
/

transformation_lora

Model card Files Files and versions

Tr@nsf0rmation_style — MiniMax H3 Video Transformation LoRA

This LoRA is fine-tuned on the MiniMax H3 (Hailuo) diffusion transformer architecture to generate dynamic, seamless visual morphing and transformation transitions between two images or scenes. It guides the model to fluidly deconstruct and reassemble individual elements, textures, outlines, and subjects from the initial frame into the final composition.

🎬 Video Showcase

Sample 01: Scene Transformation

Prompt: [Shot 1] Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another.

Sample 02: Full Subject Transformation

Prompt: [Shot 1] Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another.

Sample 03: Multi-Element Scene Reassembly

Prompt: [Shot 1] Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another. The bald person in grey T-shirt and blue jeans from <Picture 1> Tr@nsf0rmation into the Girl in <Picture 2>. The person in the Yellow shirt from <Picture 1> Tr@nsf0rmation into the yellow grass from <Picture 2>. The person in the green shirt from <Picture 1> Tr@nsf0rmation into the blank background in <Picture 2>.

Sample 04: Visual Transformation

Prompt: [Shot 1] Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another.

Sample 05: Fluid Transition

Prompt: [Shot 1] Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another.

Sample 06: Semantic Morphing

Prompt: [Shot 1] Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another.

Sample 07: Complex Subject Swap

Prompt: [Shot 1] Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another.

Sample 08: Pencil Stroke Evolution

Prompt: [Shot 1] black pencil outlines from Picture 1 Tr@nsf0rmation into Picture 2 with individual pencil strokes undergoing Tr@nsf0rmation into one another.

Sample 09: Dynamic Line Dissolve

Prompt: [Shot 1] Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another.

Sample 10: High-Fidelity Flow

Prompt: [Shot 1] Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another.

🚀 Usage Guide

Trigger Word & Prompt Structure

Main Trigger Word: Tr@nsf0rmation_style
Action Connector: Tr@nsf0rmation into

The LoRA operates best when describing the starting state (<Picture 1>) transitioning into the final state (<Picture 2>), explicitly calling out how subjects, materials, or individual visual elements transform.

Recommended Prompt Template:

integrated_multimodal_description: [Shot 1] Tr@nsf0rmation_style, Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another. [Subject A from <Picture 1>] Tr@nsf0rmation into [Subject B from <Picture 2>].

Example Prompts:

Multi-element correspondence:

integrated_multimodal_description: [Shot 1] Tr@nsf0rmation_style, Picture 1 Tr@nsf0rmation into Picture 2 with individual elements and subjects Tr@nsf0rmation into one another. The bald person in grey T-shirt and blue jeans from <Picture 1> Tr@nsf0rmation into the Girl in <Picture 2>. The person in the Yellow shirt from <Picture 1> Tr@nsf0rmation into the yellow grass from <Picture 2>.
Pencil & line-art transformation:

integrated_multimodal_description: [Shot 1] Tr@nsf0rmation_style, black pencil outlines from Picture 1 Tr@nsf0rmation into Picture 2 with individual pencil strokes undergoing Tr@nsf0rmation into one another.
Direct object transformation:

Tr@nsf0rmation_style, a ripe red tomato with green leaves Tr@nsf0rmation into a cartoon boy with brown hair in a cap and vest holding a rake over his shoulder.

⚙️ Recommended Inference Settings

Parameter	Recommended Value	Notes
Base Model	MiniMax H3 FL2VA / Ref2V	e.g. `minimax_h3_fl2va_pruned_int8_convrot.safetensors`
LoRA Strength (st2500)	`1.0` – `1.2`	Step 2500 checkpoint: fully converged transformation effect
LoRA Strength (st1600)	`1.0` – `1.4`	Step 1600 checkpoint: softer, slightly lighter stylization
Stacked LoRA Setup	`st2500 @ 1.2` + `st1600 @ 1.4`	Used in the showcase generations for maximum morphing adherence
Sampler	`res_multistep` or `euler`	ResMultistep provides stable temporal transitions
Scheduler	`simple`	Flow-match scheduler
Steps	`12` (with DMD turbo) / `25–30` (standard)	Works well with 8-step/12-step distilled turbo adapters
Sigma Shift	`shift_video = 8.0`, `shift_audio = 3.0`	MiniMax H3 custom sigma shift node
Resolution	`852x480`, `1024x576`, or upscaled to `1080p`	Compatible with 3D Latent Upscaler (`minimax_h3_latent_upscaler_3d_bf16`)

📦 Checkpoints Included

Transformation_style_c1-st2500_comfyui.safetensors: Trained to 2500 steps. Strongest transition cohesion and element-level morphing.
Transformation_style_c1-st1600_comfyui.safetensors: Trained to 1600 steps. Intermediate checkpoint ideal for subtle blend transitions or when combining with other style LoRAs.

🛠️ ComfyUI Integration

Place the LoRA files inside your ComfyUI models directory:
```
ComfyUI/models/loras/
```
Connect a LoraLoaderModelOnly node after your MiniMax H3 UNet loader (minimax_h3_fl2va_pruned_int8_convrot.safetensors).
Set strength_model between 1.0 and 1.2.
Provide <Picture 1> (start frame) and <Picture 2> (target frame) into your First-and-Last-frame conditioning nodes.
In your prompt node, format your description using Tr@nsf0rmation_style and Tr@nsf0rmation into.

Downloads last month: 72