Replaces single-sample steps with batched sampling via random.choices().
Tensors are stacked to [B, T, C] before the forward pass; t is now [B].
Default grad_accum lowered to 1 since real batching gives stable gradients.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Raw curve shown in light blue, EMA-smoothed (beta=0.9) overlay in darker
blue. Both saved as PNG at end of training. The node IMAGE output now
returns the smoothed version. Live preview also uses the smoothed overlay.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
500 warmup steps is 25% of a 2000-step run — too long. 100 steps lets
the full lr kick in much earlier without sacrificing stability.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The third element in ComfyUI's preview tuple is max_size in pixels, not
JPEG quality. Passing 85 was capping the live loss curve at 85×40px.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
torch.enable_grad() alone is insufficient: operations on inference tensors
(created inside ComfyUI's outer inference_mode context) produce inference
tensors even inside enable_grad, breaking autograd. inference_mode(False)
exits the inference context so the deepcopy, apply_lora, and training loop
run with a fully clean autograd context.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
torch.enable_grad() re-enables grad tracking but nn.Parameters created while
torch.inference_mode() is active are inference tensors that can't enter autograd
regardless. Splitting into _train_inner() and calling it inside enable_grad()
ensures the deepcopy, apply_lora, and the training loop all run with a clean
autograd context.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
ComfyUI executes all nodes inside torch.no_grad(), which prevents gradient
tracking and makes loss.backward() fail. torch.enable_grad() overrides it.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
apply_lora() is called after generator.to(device), so lora_A/lora_B were
being created on CPU while the rest of the model was on CUDA.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
STFT hop-size rounding produces ±1 latent frame vs the expected seq length.
Clamp to seq_cfg.latent_seq_len after transpose so generator.forward assertion passes.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Recent torchaudio defaults to torchcodec as the audio backend, which requires
FFmpeg shared libraries. Falls back to soundfile for envs where torchcodec
can't load (e.g. containerised ComfyUI without system FFmpeg).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
torch.stft requires float32 input — casting vae_utils to bf16 caused silent
failures during dataset pre-loading. Also adds traceback.print_exc() so future
clip-load errors are visible in the ComfyUI log.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
At every save_every steps, run a quick 8-step no-CFG inference pass on
a random training clip and save the decoded waveform as
sample_stepXXXXX.wav next to the checkpoint. Uses the existing
generator.unnormalize + feature_utils.decode + vocode pipeline from
the sampler. Failure is non-fatal (logged and skipped).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Send updated loss curve to ComfyUI frontend every 50 steps via
pbar_train.update_absolute() with a JPEG preview tuple — same
mechanism as KSampler's denoising previews.
- Fix x-axis step labels for resumed runs (previously always started
at 0; now correctly shows start_step + offset).
- Split _draw_loss_curve (returns PIL Image) from _pil_to_tensor
(converts for ComfyUI IMAGE output).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Runs the full training loop inside ComfyUI. Reuses the already-loaded
CLIP model from the inference model for text encoding; loads only a
minimal VAE encoder separately (freed after dataset pre-loading).
Outputs:
- SELVA_MODEL with LoRA applied (ready to connect directly to Sampler)
- adapter_path STRING (for SelVA LoRA Loader in future sessions)
- loss_curve IMAGE (PIL-rendered line chart of training loss per 50 steps)
Progress is shown via ComfyUI ProgressBar (two phases: dataset loading,
then training steps). Resume is supported via resume_path input.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Step checkpoints now save optimizer state, scheduler state, and step
number alongside the LoRA weights. Pass --resume path/to/adapter_stepXXXXX.pt
to continue training from that checkpoint. --steps always means total steps,
so resuming from 1000 with --steps 2000 trains 1000 more steps.
adapter_final.pt format is unchanged (state_dict + meta only) so
SelvaLoraLoader remains compatible.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- _resolve_named_path: replace / \ and null in name to prevent path
traversal outside cache_dir (would cause a confusing FileNotFoundError
at np.savez time instead of at path resolution).
- train_lora: load_npz was called twice per clip when prompt was in
prompts.txt; consolidate to a single call before prompt resolution.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When name is provided, features are saved as name.npz (or name_001.npz,
name_002.npz etc. if the file already exists) instead of a content hash —
useful for building a named training dataset. Hash-based caching is
unchanged when name is left empty.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Input is now pre-extracted .npz files (from SelvaFeatureExtractor) paired
with clean audio files (same stem). Visual features no longer re-extracted
during training.
- FeaturesUtils loaded with enable_conditions=False (VAE only) — Synchformer
and T5 are no longer loaded, saving ~3-4 GB VRAM.
- CLIP text encoder loaded separately via patch_clip so text prompt can differ
from the one used during feature extraction.
- Prompt priority: prompts.txt override > embedded in .npz > directory name.
- Removed: torchvision video loading, frame sampling/resizing, net_video_enc,
synchformer path check.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- LoRALinear now creates lora_A/lora_B with dtype matching the base
linear's weight, preventing a float32/bf16 mismatch at forward time
when the generator is loaded in bf16 or fp16.
- Remove unused `import math` from train_lora.py.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Teaches the model new/partial sound classes from custom video+audio pairs.
Only ~10 MB of adapter weights are trained vs ~4.4 GB for the full model.
selva_core/model/lora.py
LoRALinear: wraps nn.Linear with frozen base + trainable A/B matrices.
B initialised to zero → zero adapter contribution at init.
apply_lora(): walks named_modules, replaces matching nn.Linear in-place.
Default target: "attn.qkv" (all 21 SelfAttention QKV projections in
large_44k). Add "linear1" to also wrap post-attention output projections.
get_lora_state_dict() / load_lora() for ~10 MB save/load.
train_lora.py (standalone script, no ComfyUI dependency)
Data format: directory of video files + optional prompts.txt
("filename: description"). Falls back to directory name as prompt.
Pre-extracts features for all clips into RAM, then trains from those.
Training loop: encode audio→latent (need_vae_encoder=True), flow
matching MSE loss on velocity prediction, backward on LoRA params only.
Saves adapter_stepNNNNN.pt checkpoints + adapter_final.pt with metadata.
Key verified interfaces used:
encode_audio() → DiagonalGaussianDistribution; .mode().clone() required
normalize() is in-place
forward(latent, clip_f, sync_f, text_f, t) takes raw tensors
nodes/selva_lora_loader.py (SelVA LoRA Loader ComfyUI node)
Loads .pt adapter, deep-copies the generator, applies LoRA, loads weights.
strength param scales lora_B to adjust adapter contribution at inference.
Reads rank/alpha/target from embedded metadata if present.
Returns a patched SELVA_MODEL bundle for use with the existing Sampler.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Replace zero-fill with neutral gray (0.5) fill so masked background
pixels stay in-distribution: 0.5 maps to ~0 in CLIP normalized space
and exactly 0 after sync's [-1,1] normalization
- Add mask_strength float (0–1) for partial background suppression
- Add mask_clip / mask_sync booleans to toggle masking independently
on the CLIP (384px) and TextSynchformer (224px) encoding paths
- Fix temporal mask sampling: use fps-accurate index formula (same as
_sample_frames) instead of proportional int(i*M/N)
- Include mask_strength, mask_clip, mask_sync in cache hash when mask
is connected, so changing any param correctly busts the cache
- Log lines now report masked/skipped state and strength per path
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Both nodes moved models to GPU before work then back to CPU after.
Any exception (OOM, cancellation, bad input) would skip the cleanup,
leaving models on GPU permanently until ComfyUI restarts.
Wrap the entire work block in try/finally so offload_to_cpu cleanup
always runs regardless of how the node exits. Also removes the unused
`mode` variable in SelvaSampler.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- selva_sampler: wrap decode+vocode in their own OOM catch — previously
OOM during mel decode or vocoding gave a raw CUDA traceback instead
of the actionable hint
- selva_feature_extractor: sync frames log line now shows (masked) when
a mask is active, matching the CLIP log line
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Allows per-frame or static segmentation masks to be applied before CLIP
and sync encoding, zeroing background pixels. Useful when multiple objects
compete for the same sound and text prompting alone is insufficient.
- _apply_mask(): resizes mask spatially (nearest-exact), samples temporally
to match sampled frame count, multiplies into frames
- _hash_inputs(): includes mask bytes in cache key (begin/mid/end sampling)
- INPUT_TYPES: mask added to optional inputs with tooltip
- extract_features(): mask=None parameter, applied after _resize_frames for
both CLIP (384px) and sync (224px) paths, before normalization
- Log line notes when masking is active
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Model Loader:
- bf16 support check — auto-falls back to fp16 on unsupported GPUs
- DESCRIPTION and OUTPUT_TOOLTIPS
Feature Extractor:
- Store variant in features dict and .npz cache
- Progress bar (3 steps: CLIP encode, T5 encode, sync encode)
- Expand cache hash to 32 hex chars
- DESCRIPTION and OUTPUT_TOOLTIPS
Sampler:
- Variant mismatch validation against extracted features
- Cancellation support via throw_exception_if_processing_interrupted()
- OOM catch with actionable error message
- normalize toggle (optional BOOLEAN, default true) for peak normalization
- Remove empty optional: {} block
- DESCRIPTION and OUTPUT_TOOLTIPS
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Replace PreviewAudio with VHS_VideoCombine — outputs video+audio together
- Wire fps from FeatureExtractor to VideoCombine frame_rate
- Wire audio from Sampler into VideoCombine
- Clear hardcoded video filename
- Set filename_prefix to SelVA, save_output=true
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- nodes/__init__.py: fix [PrismAudio] leftover label in error print
- selva_feature_extractor: hash beginning, middle and end of video tensor
instead of just first 1MB, avoiding collisions on videos with same opening frames
- selva_sampler: derive SequenceConfig from model template via dataclasses.replace
instead of hardcoding sampling_rate/spectrogram_frame_rate per mode
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This branch registers only the three SelVA nodes. PrismAudio nodes stay
on master/feature/lora-trainer.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Newer hf_hub stopped passing proxies/resume_download/local_files_only/token
to _from_pretrained(). Give them defaults so the call doesn't fail when
these kwargs are omitted.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Actual filenames in jnwnlee/SelVA: generator_*_44khz_sup_5.pth.
download_utils.py had the wrong names so those MD5s are unverified — set to
None to skip MD5 check for 44k generators. All other files verified/unchanged.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously _ensure() trusted any existing file. Files downloaded by the
broken requests-based code (HTML error pages) would be silently reused.
Now checks MD5 on every load; deletes and re-downloads on mismatch.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
download_utils.py used requests without auth — jnwnlee/SelVA returned an
HTML error page which torch then failed to unpickle ('E' / opcode 69).
huggingface_hub.hf_hub_download() handles HF_TOKEN auth automatically,
validates downloads, and retries. Files are still copied to models/selva/.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
PyTorch 2.6 changed the default to weights_only=True. SelVA checkpoints
contain non-tensor types (numpy scalars etc.) that fail strict unpickling.
All weights come from trusted sources (jnwnlee/selva HF repo).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- SelvaSampler: multiline:false puts negative_prompt inline above sliders
- SelvaModelLoader: VAE filenames in download_utils are v1-16.pth/v1-44.pth,
not v1-{mode}.pth (mode includes the 'k' suffix)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Move negative_prompt to required inputs, right after prompt, so it appears
above duration/steps/cfg/seed in the ComfyUI node layout.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
find_pruneable_heads_and_indices and prune_linear_layer were removed from
both pytorch_utils and modeling_utils in some transformers builds. Provide
minimal inline implementations as final fallback — prune_heads() is never
called at inference time so correctness is only needed for completeness.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Users can now wire the prompt output directly to SelvaSampler's prompt input,
making the data flow explicit instead of relying on the implicit features fallback.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
ComfyUI renders required inputs above optional ones. Moving negative_prompt
to optional puts prompt first (natural order) and negative_prompt at the
bottom where it belongs as a power-user input. Also guards against
negative_prompt=None when not connected.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Uses selva_core/utils/download_utils.py (already has URLs + MD5s for all
weights). Models download to models/selva/ on first load. Synchformer reuses
models/prismaudio/synchformer_state_dict.pth if already present (no duplicate
download for PrismAudio users), otherwise downloads to models/selva/.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
SelvaFeatureExtractor now stores the prompt in SELVA_FEATURES (both in the
returned dict and the .npz cache). SelvaSampler's prompt is now optional —
when left empty it falls back to the prompt stored in features. A non-empty
override can still be passed when CLIP text should differ from the sync text.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Some transformers builds removed these from pytorch_utils. Fall back to
modeling_utils which exposes them in all known versions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- selva_feature_extractor: cache hash now includes resolved duration;
same video + different duration override no longer returns stale features
- selva_sampler: MPS-safe noise generation (torch.Generator on CPU then
move to device, same pattern as PrismAudioSampler)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Calls update_seq_lengths with actual feature dimensions (not seq_cfg) to
avoid rounding assertion mismatches. Progress bar tracks each Euler step.
Supports negative prompts for steering, normalizes output to [-1,1].
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
CLIP frames at 8fps→384px (normalize inside FeaturesUtils).
Sync frames at 25fps→224px, normalized to [-1,1] externally.
T5 text encoded via FeaturesUtils, sup tokens prepended, then text-conditioned
sync features extracted via TextSynch.encode_video_with_sync(). Results cached
as .npz keyed by hash(frames[:1MB] + prompt + fps + variant).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Pure PyTorch SelVA source for SelvaModelLoader/FeatureExtractor/Sampler nodes.
Imports rewritten from selva.* to selva_core.*. mel_converter.py: replaced
librosa.filters.mel with pure-numpy implementation to avoid librosa→numba→NumPy
version incompatibility in some ComfyUI environments.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The fps output was only returned on cache hits. Fresh extractions
returned only features, leaving fps null.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Point links to huggingface.co/FunAudioLLM/PrismAudio and use public
GitHub URL for install instructions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Setting duration to 0 in PrismAudioSampler now reads the duration
stored in the PRISMAUDIO_FEATURES dict (set by the feature extractor).
Default changed from 10.0 to 0.0 so V2A workflows are wired up
automatically.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Remove synchformer_ckpt input — always resolved from models/prismaudio/
(errors early with clear message if missing)
- Replace python_env string input with dropdown: managed_env (isolated
auto-created venv, default) or comfyui_env (current Python, with warning)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When the VHS LoadVideo video_info output is connected, loaded_fps is
used automatically instead of the manual fps input.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Saves frames as uint8 .npy instead of H.264 MP4, eliminating the
lossy codec roundtrip. extract_features.py loads .npy directly and
skips decord when given a numpy file. Passes --source_fps for
correct temporal sampling.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove debug_zero_video/debug_zero_sync inputs from PrismAudioSampler,
DIT velocity diagnostics, conditioner stats logging, and feature stats
prints from both sampler.py and text_only.py.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The wrong model (videoprism_public_v1_large, vision-only) was used,
causing V2A audio distortion. Switch to the LvT variant which has a
text tower, pass CoT captions for joint encoding, and extract per-frame
features from outputs['frame_embeddings'] (L2-normalized, [T, 1024])
instead of manually averaging spatial patches.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add per-key conditioning output stats (after Cond_MLP/Sync_MLP, after
_substitute_empty_features) to both sampler and text_only nodes. Also
add raw T5 text feature stats in T2A before conditioning.
This lets us directly compare:
- T2A vs V2A conditioning outputs to find which path differs
- T2A vs npz text feature ranges
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Match the diagnostic output already in text_only.py to compare
V2A vs T2A latent distributions and diagnose conditioning issues.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Zero features through bias-free Cond_MLP produce near-zero activations,
not the learned null signal the model was trained with. Use empty_clip_feat
(the learned null video embedding) just like empty_sync_feat for sync.
Also improve text_prompt tooltip to encourage detailed CoT descriptions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
AutoencoderPretransform.load_state_dict() doesn't return IncompatibleKeys.
Load into pretransform.model (AudioAutoencoder) to get the return value
and see actual missing/unexpected key counts.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Print key counts, missing/unexpected keys, and sample key names to
diagnose whether weights are actually loading correctly (strict=False
silently hides mismatches that would cause garbage audio output).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Allows isolating which feature set causes quality issues:
- debug_zero_video: zero video_features → text+sync only
- debug_zero_sync: zero sync_features → text+video only
Also logs mean/std/shape for all three feature tensors on every run.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Sync_MLP interpolates sync features based on video duration, but audio
latent length depends on the user-set audio duration. When video != audio
duration, the sequences diverge. Resample sync_cond to x's length before
the gated addition so any video/audio duration combo works.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Stream raw RGB bytes from tensor directly to ffmpeg stdin.
Eliminates all intermediate PNG file I/O — much faster for large frame counts.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Shows how long PIL+ffmpeg video export takes so we can see
if that's contributing to the gap before [extract] output appears.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Each step now prints elapsed seconds on completion.
Total time printed at the end to identify bottlenecks.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
prismaudio.json conditioner config requires:
- video_features: dim=1024 → switch videoprism_public_v1_base → large (ViT-L)
- sync_features: dim=768, length divisible by 8 → expand [num_seg,768] to
[num_seg*8,768] (per-frame) so Sync_MLP can reshape by groups of 8
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
T5-Gemma outputs BFloat16 which numpy does not support.
Cast all feature tensors with .float() before .numpy().
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When synchformer_ckpt input is empty, look for synchformer_state_dict.pth
in the ComfyUI prismaudio models directory automatically.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
videoprism/__init__.py is empty — API lives in videoprism.models.
Fix: from videoprism import models as vp (not import videoprism as vp).
Also add flax to managed venv packages (required by videoprism Flax model).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
RTX 6000 Pro (Blackwell SM 10.0) fully supports CUDA 13. Switch from
jax[cpu]+jaxlib to jax[cuda13] which bundles jaxlib and uses
pip-managed CUDA libraries. Delete _extract_env to force a rebuild.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
google/t5gemma-l-l-ul2-it is a gated HuggingFace model requiring auth.
Add optional hf_token input on the node; forward it (plus the legacy
HUGGING_FACE_HUB_TOKEN alias) to the subprocess env. Falls back to
HF_TOKEN from the host environment. Warn clearly when neither is set.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Empty string from clearing the node field caused subprocess to execute ''
which raises PermissionError. Now any blank or 'python' value uses the
auto-installed venv.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Creates data_utils/v2a_utils/feature_utils_288.py with FeaturesUtils:
- T5-Gemma text encoding via transformers
- VideoPrism video encoding via JAX videoprism package
- Synchformer visual encoder loading from checkpoint
Also fixes extract_features.py to add plugin root to sys.path so
data_utils is importable in the subprocess venv.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Exposes the video frame rate as an optional input (default 30).
Correct FPS ensures accurate temporal frame sampling in VideoPrism
and Synchformer feature extraction.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
write_video requires the optional 'av' (PyAV) package. Use PIL to save
frames as PNGs then combine with ffmpeg, which is always present in
ComfyUI Docker images.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Installs each package individually with [n/total] counters and
pip progress bars, so failures pinpoint the exact failing package.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- tensorflow-cpu==2.15.0 only supports Python <=3.11; relax to >=2.16.0
- capture_output=False so pip errors are visible in ComfyUI logs
- clean up incomplete venv dir before retrying install
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
vae.ckpt is a full training checkpoint containing discriminator, STFT
loss modules, and EMA wrappers that are absent from the inference
AudioAutoencoder. strict=False ignores these training-only keys while
still loading all encoder/decoder/bottleneck weights correctly.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
