Skip to content

feat: audio-vae-cpu, save/load latent, optimized scripts, LoRA docs#1639

Open
Daniel-Uzcategui wants to merge 2 commits into
leejet:masterfrom
Daniel-Uzcategui:pr-audio-vae-cpu
Open

feat: audio-vae-cpu, save/load latent, optimized scripts, LoRA docs#1639
Daniel-Uzcategui wants to merge 2 commits into
leejet:masterfrom
Daniel-Uzcategui:pr-audio-vae-cpu

Conversation

@Daniel-Uzcategui

Copy link
Copy Markdown

Features

--audio-vae-cpu

Run audio VAE on CPU to free GPU VRAM for video VAE. Audio VAE is lightweight and runs faster on CPU anyway, so this is a net win for VRAM-constrained setups.

--save-latent / --load-latent

Save the denoised latent tensor before VAE decode, or load a pre-computed latent to skip text encoding and denoising. Combined with --audio-vae-cpu, enables fast iteration on VAE parameters without re-running denoising.

Optimized generation scripts

Batch scripts for 480p/720p/768p generation with tuned VAE tiling parameters.

LoRA documentation

LORA_DUAL_CHAR.md — guide for dual-character LoRA usage (two subjects from one LoRA).

Other changes

  • skip_model_check parameter on resolve_and_validate for latent-loading workflows
  • cast_audio_activation_for_weight for CPU audio VAE compatibility
  • Allow hires_denoising_strength=0 (skip denoising) for load-latent workflows
  • Updated TEST_MATRIX.md with complete generation log

🤖 Generated with Claude Code

Daniel-Uzcategui and others added 2 commits June 11, 2026 22:00
…kflows, update TEST_MATRIX

When using --load-latent, the denoising step is skipped entirely. Change
the hires_denoising_strength clamp from (0.0001, 1.0] to [0.0, 1.0] and
the validation range accordingly so users don't need to pass a dummy
non-zero denoising strength.

TEST_MATRIX updated with complete generation log across all styles.

Co-Authored-By: Claude <noreply@anthropic.com>

@wbruna wbruna left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please review the contribution conventions for the project, especially the part about AI-assisted code.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants