Upload 2 files

Delete app.py

Update app.py

Update app.py

add keyframes, singlepass checkbox, settings import/export

update transition defaults

add transition lora

adjust resolution parameters

update resolution defaults

fix rounding error

update resolution handling

add optional lora support

Unified presets with sigma textbox

Tune identity defaults

Add scene chaining and audio reference prep

Add audio reference + per-lora audio strength via MultiLoRALoader

Add mistic better motion + physics v2 loras, sulphur defaults 0.15/0.15, log gen settings

Pre-populate workflow cache in parent so fork COW skips re-conversion

Fix kijai omninft bf16 URL case (repo + filename)

Expose OmniNFT + sulphur v1 loras, two-regime GPU budget, comment trim

Expose K/V strength as a tunable slider in a new funpack accordion

Cache K/V zero-prefix tensors to fix VAE NVML asserts; drop kv checkbox info blurb

Use delegating CompressedTimestep subclass when n_ref misaligns with ppf

Fix K/V timestep extension to match vx exactly when n_ref not divisible by ppf

Extend v_cross_pe in K/V AV compat patch for audio-to-video cross-attention

Patch FunPack KV wrapper for LTX-AV: extend modulation tensors alongside positional embeddings

Add K/V identity conditioning via FunPack wrapper node

Hide input mode UI, revert title to i2v, raise dreamly/singularity defaults

Add prompt relay (timeline-based prompts) with strict parser and silent fallback

Add singularity lora as optional toggle, default off

Pass all reference images to enhancer; mark MSR modes WIP; reorder input mode radio

Add multi-reference (original) input mode with bundled workflow + normalize lora path separators

Route MSR conditioning from IC-LoRA guide direct; match canonical per-guide strength; normalize lora path separators

Fix MSR conditioning chain and add per-image guide multi

Drop Gemma enhancer, fix MSR node ids (552->740, 556->596)

Pre-download Gemma enhancer at startup so no model downloads inside GPU fork

Add MSR multi-reference mode and selectable Gemma-4 enhancer for MSR

Tune defaults: sulphur 0.3, vbvr 0.5, similarity 0.3, sigmas tuned

Register sigma change handler last so generate/enhance fn_indexes stay stable

Add refine sigma preset selector (original/tuned/custom) with validation

Remove audio toggle, native AV always on

Expose identity anchor tuning, auto-align cache step to frame count, drop audio gpu-time penalty

Switch vbvr to archive mirror, add dreamly/synth/plora optional loras

Fix get_gpu_duration signature to match generate lora args

Rename optional lora to sulphur experimental, add vbvr lora

Add optional eros experimental lora, zerogpu budget controls, updated header

Set enhancer GPU duration to 80s

Bump enhancer GPU duration to 120s for cold model load

Link against system CUDA 13 libs and bundle them with the binary

Build llama-server with pip CUDA 12.8 toolkit to match runtime libs