Character Cartridges (.joypack v1.0): spec, packer, loader node, and WREN - the first portable video-character file (verified: same face + rhotic General American voice in an unseen scene, zero training)

v1.7: the 24fps rule (fps steers accents) - default 24 + warning; fp8 loader passthrough fix; AutoFinish console-flash fix

v1.6: voice casting system + sampler-path audit fixes

INT8 ConvRot support: the loader now reconstructs int8_tensorwise comfy_quant checkpoints to bf16 at load (dequant + regular-Hadamard un-rotation, verified ~0.95% from the bf16 source). The published INT8 surgical build now loads directly in the multishot workflow.

RAM preflight: account for fp8 file -> bf16 upcast at load (~2x on-disk size). Previous estimate missed the reported crashing config (25GB fp8 file on a 64GB box).

Add host-RAM preflight warning before model load. Three users reported Windows hard-crashing (RAM 99% -> reboot) during first load; the loader materializes the whole checkpoint in host RAM, which fits with a large pagefile and kills a 64GB box with the Windows default. Warns with concrete fixes instead of dying silently.

Clearer error for ComfyUI-quantized (INT8 ConvRot / fp8mixed) checkpoints: name the bf16/GGUF builds of the SAME model that work in this pack, and say the INT8 file is for ComfyUI's native loaders. Two users hit the old wording.

Linux hard-crash fix: WAV sidecar uses soundfile first, torchaudio removed. Avoids the torchcodec/FFmpeg dlopen that could trip a glibc TLS loader abort (ld.so _dl_allocate_tls_init) and kill ComfyUI mid-run on systems with FFmpeg <7.

wav save: fall back torchaudio -> soundfile -> builtin wave (Linux user without FFmpeg shared libs hit 'Could not load libtorchcodec' on torchaudio.save)

Generate VALIDATE_INPUTS: turn ComfyUI's cryptic widget-scramble type error into an actionable 'delete + re-add the node' message

Generate: self-heal shifted widget values from older graph layouts (NaN hires_factor / absurd reference_zoom) with loud console guidance

docs: hires routing table everywhere (hires_factor = routing switch); PREVIEW-labeled SaveVideo; routing-framed tooltip; refreshed v1.5 zip

prune failed churn-experiment hires modes (frozen/faint/converge); dropdown = subtle/medium/strong/spatial with tournament-informed tooltip

autofinish default upscale = bicubic+CAS (deterministic, zero temporal churn; tournament winner by eye); RTX kept as legacy option; spatial/faint/converge/frozen hires modes + SeedVR2 findings

fix: video RoPE clock uses the render fps (hardcoded 24 at 25fps renders = 4%/s AV rope skew = the ~10s lip-sync drift)

hires v2: per-shot shared noise fields (kills the window-boundary texture switch) + tenstrip 4-step upscale ladder as 'strong' mode

audio memory default ON: cross-shot voice consistency is the point of multishot; sync/voice tradeoff documented in tooltip

enable_audio_memory default OFF (every perfect-sync render ran without it)

head_trim_frames default 14 (production-proven; shot-start sync fix)

v1.4: LoRA Stack node (dropdown+strength UI, chainable), shipped wired into the loader

v1.4: multiline lora_path box; operator's node layout; widget repair

v1.4: multi-LoRA loading; shipped Reference Image node; instructions updated

v1.3: VHS glitch reaches the master (worker port + transition sidecar); v2a_grad_scale note corrected

LLMEnhance: local endpoints no longer require an api_key (placeholder sent automatically); clearer error + tooltip; document the field in the workflow README

v1.1: A/V sync fix, AutoFinish nodes, gemma_file dropdown, manual_path, framing/movement doctrine, ready-to-run multishot workflow

Lazy text-encoder load: build only on conditioning-cache MISS (saves 21-33GB host RAM + load time on cache-HIT runs)

hires-fix: 33f windows + zero resident blocks during the pass (VAE encoder fatal abort at 65f/1920x1088 with 12GB pinned)

fix: _ref_sched unbound on no-reference runs (no-character scripts)

encoder_fp8: strip unused vision tower + absolute-headroom fits check -> GPU encode engages on 24GB cards

conditioning disk cache (encode once per script per machine) + CPU-fp8 encoder warning

hires-fix second pass: per-shot latent re-encode + DMD tail re-denoise at target res (real detail synthesis; 65f windows, 3090-safe)

guard: fp8-compute toggles require bf16 source file (fp8 files load all-fp8 with cast skipped -> randn crash)

encoder_fp8: halve the Gemma wrapper footprint (fp8 storage, upcast at encode; enables GPU encode pass on 32GB)

fp8_scaled_mm mode (native fp8 matmul, no upcast tax) + resident_blocks partial offload

Generate: schedule-aware ref dedup, cap 6 scheduled entries

TextEncode: hot-swap encoder to GPU for the encode pass (fits-check + OOM fallback)

per-character reference scheduling: RefPicker emits ref_shots (first-appearance shot per character), Generate injects each ref at its character's entry shot with the long trim - multi-character single-run videos get every character properly seeded

load order: DiT (quantize/fuse peak) first, VAEs, text encoder LAST + gc between stages - encoder-first stacked 24GB under the fuse peak and caused an mmap access violation at VAE load on a 96GB box

references seed shot 1 only: from shot 2 the bank carries identity+staging; re-injecting a wide full-scene ref every shot made the subject relocate between shots

no-offload pre-flight: refuse cannot-fit DiT with a catchable error (cudaMallocAsync otherwise hard-aborts the whole process), warn on tight activation budgets

reference_zoom: over-zoom refs before conditioning - a ref at exactly render size is pixel-continuable and the model opens the shot ON it (long literal ghost no trim covers); 1.2 default breaks the continuation shortcut while keeping identity + wide composition

head_trim_first_shot: shot 1 conditions on refs alone (empty bank) and opens ON the reference - give it its own longer trim (auto = 2x head_trim, min 24)

LoRA hardening: kohya lora_down/up support + alpha scaling, fuse telemetry + zero-match warning, refuse comfy_quant checkpoints with clear error; lora_file dropdown

RefPicker: character dropdown, no silent scan-fallback, cache fix; head-trim applied to all outputs; lora_file dropdown on ModelLoader; gemma_path guard

Upload 11 files