Modelpublic

SyFeee/Bernini-LTX-Research-Checkpoints

sourceHugging Faceotherupdated 1mo agoView on Hugging Face

0likes

Model Card

Bernini LTX Research Checkpoints

This repository contains 12 experimental Bernini adaptation runs trained against the LTX-2.3 22B development checkpoint. These are early research artifacts, not production-validated releases.

Important Loading Requirement

Each run contains two files that must be loaded together:

—lora_weights_step_*.safetensors: the rank-64 LoRA adapter.
—bernini_P_step_*.safetensors: the matching Bernini projection weights.

The LoRA alone is incomplete. Keep both files from the same run and step; projection weights are not interchangeable across runs.

Included Runs

Run	Step	Research focus
`bernini_av_phaseA`	1000	Reference-to-video plus joint audio
`bernini_av_phaseA_704`	1000	Phase A at the 704 training setup
`bernini_av_phaseB_704`	1000	Reference-voice conditioning at 704
`bernini_combined_pilot`	1800	Combined conditioning pilot
`bernini_duo_pilot_inline`	1000	Inline two-subject conditioning
`bernini_fight_motion_pilot`	1500	High-motion pose-control pilot
`bernini_multishot_704`	1000	Multi-shot continuity at 704
`bernini_p2_baseline`	1500	P2 baseline
`bernini_p2_plan`	1500	P2 planning-conditioned run
`bernini_r2v_msr_baseline`	1500	R2V MSR baseline
`bernini_r2v_plan_pure`	1500	R2V plan-only variant
`bernini_r2v_plan_ref`	1500	R2V planning with references

Each runs/<run>/training_config.yaml records the corresponding experiment configuration. Paths in these archival configs reflect the original training machine and must be adapted locally.

Scope and Use

The checkpoints explore identity/reference conditioning, planning, multi-subject control, motion control, multi-shot behavior, and audio generation. Quality varies by run. Validate identity, motion, audio, and temporal consistency on representative prompts before deployment.

Use of these derivatives remains subject to the LTX-2 Community License and the terms of the base model.