nodes/olivv-cs-ComfyUI-FunPack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
2,481 Commits 3 Branches 11 Tags
3dbcc8a4e8fada32db8f7be62357ef385ce18dab
Commit Graph
471 Commits
Author SHA1 Message Date
digital-garbage 3dbcc8a4e8 Prompt enhancer: Dynamic style for more motion; counts-only motion check script 2026-11-05 23:29:27 +02:00
digital-garbageandClaude Sonnet 5.5 6249d5342b SLA/Veda: a run cancelled in its first step no longer leaves its stabilize-motion memory or kernel failure to the next run 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
 2026-11-05 22:12:45 +02:00
digital-garbage a5f35ea5b9 Veda: split the kernel call into index, quantize and attend timings; cheaper host prep with equal output 2026-11-05 22:10:34 +02:00
digital-garbageandClaude Sonnet 5.5 6a67c85a8e Test: a cached model gives the same answer after other runs and after a run cut short, for every modifier alone and all together, one stage and two 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
 2026-11-05 22:00:00 +02:00
digital-garbage affa3a863e Veda: opt-in per-phase timing, one line naming every size of a Two stages run, fewer device reads 
veda_timing (loader, method veda) logs the time in gather, score, select, kernel and scatter
for a run; off by default and it changes no number. The run line now names the size and tile
plan of each stage, not only the last. Cheap speed-ups that keep the numbers exactly: the
list width of kept tiles comes from the layout instead of a device read, the head groups and
the per-row pattern stay on the device, and the predictor's projections are copied from pinned
memory.
 2026-11-05 21:57:31 +02:00
digital-garbage 702cbe3960 Test: the Nth run of a model equals the first, for every modifier on the tiny H3 
Each modifier alone, same inputs four times: identical numbers, weights untouched, and feeding
the loader's output back into the loader installs nothing twice. Guards against state left on a
cached model making later generations different (the quality loss once suspected on LTX).
 2026-11-05 21:51:22 +02:00
digital-garbage fa3d4bd71a Veda predictor download row sits right under the predictor selector 
It was drawn at the bottom of the model page, far from the field it fills, and the
selector's hint pointed at 'the row below'. The row now follows the selector directly and
the hint no longer points anywhere.
 2026-11-05 21:46:44 +02:00
digital-garbage 1e2ac5ca67 Sparse attention method veda for MiniMax H3 (experimental), with an optional predictor download 
Learned tile-sparse attention (Veda, arXiv 2605.30325): tiles, tile plans, predictor
scoring and selection in plain torch (tested on CPU, including a real tiny H3 with a
stand-in kernel), an INT8 Triton kernel derived from SageAttention (untested on a GPU),
installed through the same override and per-step wrapper as SLA. Runs at the predictor's
trained budget; LoRA, MXFP8, Two stages, dense steps and the attention backend compose.
Settings > Models offers the official Hugging Face download of the predictor on a click,
with its licence beside it; taggers and this share one downloader (core/fetch.py).
 2026-11-05 21:32:39 +02:00
digital-garbageandClaude Sonnet 5.5 9b1cb52024 Post settings in Engine > Post: smoothing moved out of the node, plus auto contrast, auto brightness, colour strength 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
 2026-11-05 20:32:03 +02:00
digital-garbageandClaude Sonnet 5.5 0ed59f06d2 Two stages: Trilinear removed (migrated to Bicubic via renamed_options); Save Video: opt-in edge smoothing 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
 2026-11-05 20:19:14 +02:00
Claude 25216188ad Run meter: per-step grain and settling of the model's answer, compared with the previous run of the same seed (no extra model call) 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjXySkmNCh7j6UXrFYR
 2026-11-05 18:06:09 +00:00
digital-garbage 8c35c4b2ff Matrix: Block repeat, Block influence and Phrase probe join it (recording switched on, valid block range) 
PTD stays out: it replaces the sampler by its adapter's own hops.
 2026-11-05 16:55:38 +02:00
digital-garbage d792d7f5e9 Matrix: LTX file, LoRA slots across stages, stale KJNodes skips removed 
Every modifier alone, in pairs and all together on tiny H3 and tiny LTX; the H3 preset build tests run again.
 2026-11-05 16:01:20 +02:00
digital-garbage 5f6ed35929 Matrix: every modifier is listed or exempt with a reason, must report active, and must wire in; crossed with samplers and Two stages 
Brownian leaves the stock sampler alone when torchsde internals differ. LoRA: unwrapped refiner layout, absent blocks, LoKr.
 2026-11-05 15:58:29 +02:00
digital-garbage d510cf0116 Combination test: Turbo, Camera move and Seed search join the feature matrix (they were missing) 
Every feature alone, in every pair and all together, with and without Two stages: 218 runs.
 2026-11-05 15:49:41 +02:00
digital-garbage 4b306ae65f Round 3: Brownian noise equals stock noise at any sigma height; every LoRA wrapper reaches every Diffusers module 
Queries are limited as plain numbers inside the sampler call. Tests for 11 wrappers and sigma 14.6.
 2026-11-05 15:44:24 +02:00
digital-garbage 04e59550de Round 3: a stop during the encoder's last seconds still removes the file; a failed stop request is not a stop 
Encoder error text no longer says 'upscaled' for Interpolate.
 2026-11-05 15:36:42 +02:00
digital-garbage 23a89ca4c5 Test: noise merging follows the first noise's device (meta stand-in) 2026-11-05 15:30:20 +02:00
digital-garbage 478fb835dd Round 2: Brownian clamp survives the CPU tree's float32 cast; Stop per module; block map handles a paired input 
Stop only reaches jobs that check for it; a failure landing with Stop counts as a stop. Hint for imported workflows corrected.
 2026-11-05 15:29:32 +02:00
digital-garbage 3351fbf503 LoRA: a wrapped Diffusers file reaches the fused qkv rows; a wrong-width qkv pair is dropped, not merged 
README credits Karras et al. for the churn idea in Turbo. Tests for three wrappers and the width check.
 2026-11-05 15:23:43 +02:00
digital-garbage 9c4458307c Stop button also stops Upscale and Interpolate jobs 
Jobs end as stopped (kept for the poll, no half file, no restarted message); Stop shows while one runs.
Python and JS tests for the cancel path.
 2026-11-05 15:16:47 +02:00
digital-garbage 7d3baca3c6 Noise merging: the mix uses plain numbers and brings the other noise to the first one's device and dtype 
Test: the mixed noise keeps the dtype and device it meets.
 2026-11-05 15:13:29 +02:00
digital-garbage 2f1a0df7a0 Review: only H3 LoRAs are renamed to H3 names (LTX transformer_blocks LoRAs load again) 
Interpolate Always warns once, not per render. README credits the resize kernels; hint and example text corrected.
 2026-11-05 15:13:03 +02:00
digital-garbage 1b9f1cb6aa Review: Brownian noise is the stock noise again; block map profile survives audio+video blocks and cancel 
Queries are clamped into torchsde's span instead of moving the span. A cancelled profile saves nothing.
Attention explorer temperature stays above zero under the map; unmeasured groups are not given 3x.
 2026-11-05 15:10:18 +02:00
digital-garbage 51383b5370 Output review: stale last-frame picture removed, running jobs kept, encoder errors say why 
A failed Save Video keep no longer leaves an older video's picture for the next shot.
Job cap evicts finished jobs only; stream_file and probe report ffmpeg's reason, not Broken pipe.
 2026-11-05 15:05:48 +02:00
digital-garbageandClaude Sonnet 5.5 e3e2f5829a Test: Q steer's push is made visible in the tiny model by sharpening its attention (the video moved by rounding only) 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjV3UWsHHjJiAXix29B
 2026-11-05 14:48:36 +02:00
digital-garbageandClaude Sonnet 5.5 f7b1da518d Test: the VAE loader stub has a model, as the loader now requires 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjV3UWsHHjJiAXix29B
 2026-11-05 14:48:36 +02:00
digital-garbageandClaude Sonnet 5.5 f8f612277e Tests: the two spaCy-dependent camera tests skip without spaCy; test_env.sh installs spaCy and its model as an optional extra 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjV3UWsHHjJiAXix29B
 2026-11-05 14:48:36 +02:00
Claude 4579d31585 LoRA Loader: a slot's settings and the next slot's picker appear only once a LoRA is chosen 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjV2UVj2u3F3mbnYrQW
 2026-11-05 11:23:35 +00:00
Claude d6119b0545 Properties: Start shots removed (the start follows what the scene has), Continuity's switch moves to Project settings, fade-to-black guard always on 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjV2UVj2u3F3mbnYrQW
 2026-11-05 10:39:33 +00:00
Claude d42bcd269f Render: playing at the clips' real frame rate is an option (Settings), off by default 
The final video keeps the project's frame rate unless 'Keep the clips' own frame rate' is switched on.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjXAUznBpkWSZm9vtPL
 2026-11-04 07:09:45 +00:00
Claude 33c21c0bc3 Continuity: a shot after an interpolated or upscaled clip still continues from the sampler's own picture 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjYBYFyV8YG5YDfYPRb
 2026-11-02 14:44:07 +00:00
Claude cebfc72cee Render names are a hash, and Render keeps the clips' real frame rate 
Final render, selected export and media-bin saves no longer carry the project's name: the server names them with the same hash rule as saved scene clips. Render used the project's static FPS even when clips were interpolated; it now probes the files and uses the fastest real rate.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjVnGVrbKC3XMjBh7Te
 2026-11-02 14:37:47 +00:00
Claude 31871a5219 Continuity: carry the sampler's own last picture, not the encoded video's copy 
Save Video keeps the last decoded frame losslessly in temp; the carry uses it when the clip ends with the render.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjYBYFyV8YG5YDfYPRb
 2026-11-02 14:04:57 +00:00
Claude 208c92f5ca Continuity: a scene's aspect-source picture no longer blocks the carry; log extraction and slot instead of a toast 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjYBYFyV8YG5YDfYPRb
 2026-11-02 13:52:04 +00:00
Claude d8ce442c04 Last-frame pictures go to ComfyUI's temp folder, not the media bin 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjYBYFyV8YG5YDfYPRb
 2026-11-02 13:46:50 +00:00
Claude e2f14bd1a7 Continuity: H3 only for now, hint recommends R2V 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjYBYFyV8YG5YDfYPRb
 2026-11-02 13:37:55 +00:00
Claude 6fad1d4feb Continuity: carried frame takes the next free reference slot 
It was fed to the aspect-ratio picture slot only. Hooks may now return references, added to the scene's before wiring.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjYBYFyV8YG5YDfYPRb
 2026-11-02 13:27:13 +00:00
Claude 85fa1d607c Rating: a clip the Shot camera learned from is no longer reported as untaught 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjXtFiTrX7ixo9tG11y
 2026-11-02 13:09:07 +00:00
Claude 5eabca5839 LoRA loader: Apply to stage (First/Second/Both) for Two stages, and Diffusers-layout H3 LoRAs now load 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjVKTwSL4mCUi3hd811
 2026-11-02 12:56:50 +00:00
Claude f7d17cfb15 Diffusion Model Loader: read the stacked output head count from the weights when the file metadata lacks it 
A few-step H3 file whose metadata keeps explicit head sizes but not the head
count was built with one head and failed on the head biases. The count is now
taken from the head bias and patch projection shapes (shapes only, any device
or dtype), with a test through the real loader at every weight dtype.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjUsYDkyEtLCrzqv7jo
 2026-11-02 10:57:56 +00:00
Claude 7c26a30015 Combination test: every feature alone, in every pair and all together, with and without Two stages 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjU9jGrByQHYf6UYMVV
 2026-11-01 13:06:56 +00:00
Claude 37cdd70e75 Two stages: noise at the hand-over defaults to 0.5, and the schedule the model sees holds the backed-up start so step windows still find their step 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjU9jGrByQHYf6UYMVV
 2026-11-01 12:59:36 +00:00
Claude 12d41311d1 Two stages: optional noise added at the hand-over 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjU9jGrByQHYf6UYMVV
 2026-11-01 12:49:07 +00:00
Claude dc9e430cc1 Block map: the Log says when a run follows a saved map 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjW75aYB8UyVRZo6fCr
 2026-11-01 12:30:51 +00:00
Claude 8053b8bb90 Block map: the profile switch turns itself off once a map is saved 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjW75aYB8UyVRZo6fCr
 2026-11-01 12:16:04 +00:00
Claude f32aaac550 Block map: a profile of how much each block can take; REINS, Q steer, Attention temperature and Attention explorer nudge their strength by it, a big setting is kept up to what the block takes 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjW75aYB8UyVRZo6fCr
 2026-11-01 11:59:10 +00:00
Claude a5c74d0065 Turbo on top of a sampler: describe the churn honestly and test the real relation 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjW75aYB8UyVRZo6fCr
 2026-11-01 11:46:20 +00:00
Claude 143aa53cfe Save Video: round to 8 bit instead of cutting down 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjW75aYB8UyVRZo6fCr
 2026-11-01 11:37:36 +00:00
Claude 42d88d8f0b Turbo: 'Keep the chosen sampler' puts the fresh noise and the polish step on top of any sampler 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjW75aYB8UyVRZo6fCr
 2026-11-01 11:31:10 +00:00