Heptagram Turbo v1.0

A community-enhanced MiniMax H3 checkpoint by eatyet.

Heptagram Turbo v1.0 is a community fine-tuned and merged MiniMax H3 checkpoint, combining multiple community H3 checkpoints with integrated TaoMate and HyperFlow acceleration LoRAs.

The acceleration components are incorporated into the checkpoint, providing a streamlined generation workflow without requiring separate acceleration LoRA loading.

Model Information

Model Name: Heptagram Turbo v1.0
Creator: eatyet
Base Architecture: MiniMax H3
Model Type: Fine-tuned / Merged Checkpoint
Precision: INT8
Generation: Text-to-Video (T2V), Image-to-Video (I2V), Reference-to-Video (R2V)
Audio: Native audio-video generation inherited from MiniMax H3

Key Features

Multiple community H3 checkpoints merged into one model.
TaoMate acceleration integrated into the checkpoint.
HyperFlow acceleration integrated into the checkpoint.
Reduced sampling steps for faster video generation.
Supports T2V, I2V and R2V workflows.
Designed for streamlined local inference with ComfyUI.

Recommended Settings

Parameter	Recommended Value
Sampler	res_multistep
Scheduler	simple
Steps	8
Supported Step Range	4–8
Sigma Shift	12/3
Model Precision	INT8

These settings are based on the published Heptagram Turbo v1.0 workflow configuration. Results may vary depending on prompts, input references and hardware.

Supported Generation Modes

T2V: Text-to-Video generation.
I2V: Image-to-Video generation.
R2V: Reference-to-Video generation.

Credits

Original architecture: MiniMax AI
Base model: MiniMax H3
Acceleration components: TaoMate and HyperFlow
Model merging, fine-tuning and release: eatyet

License

This model is derived from MiniMax H3 and is distributed subject to the MiniMax H3 Community License Agreement.

Please refer to the accompanying LICENSE and NOTICE files for the applicable terms and restrictions.

Model tree for eatyet/HeptagramV_1.0

Base model

MiniMaxAI/MiniMax-H3

Finetuned

(202)

this model

eatyet
/

HeptagramV_1.0