Commit Graph
68 Commits
Author SHA1 Message Date
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 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-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
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 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 5620fcdc94 VAE loader: an unrecognised VAE file names its first keys in the error
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjWzZx5QzcqkheSuhsy
2026-10-30 17:41:26 +00:00
Claude 41559e188b VAE loader: a decode-only VAE this ComfyUI cannot read no longer stops the run; the main VAE decodes and the log says so
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjWzZx5QzcqkheSuhsy
2026-10-30 17:39:37 +00:00
Claude 3862f3ec36 LoRA Loader: up to eight LoRAs in one node, applied in order; slot 1 keeps its inputs
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjU1A5dzqoK4stWcmzk
2026-10-29 19:23:48 +00:00
Claude e439214b76 MXFP8: drop the claim that a forced compute dtype costs the speed (measured: same step times)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjWwkXVabCNSQ53xaW1
2026-10-29 11:21:09 +00:00
Claude 9278d17f8b MXFP8: also allowed with weight dtype bf16 (the usual setting for a bf16 file)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjWwkXVabCNSQ53xaW1
2026-10-29 11:10:57 +00:00
Claude 51acdb1c12 Diffusion Model Loader: MXFP8 result goes to the log, not only the status output
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjWwkXVabCNSQ53xaW1
2026-10-29 11:07:48 +00:00
Claude 3837d585e0 VAE Loader: optional decode-only VAE (Sol-H3-VAE, X2 Detail) while the main VAE encodes
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjW75aYB8UyVRZo6fCr
2026-10-29 10:59:06 +00:00
Claude b108d4f4f5 H3 loader: MXFP8 option for the middle blocks (Sol-H3 recipe), credit in README
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tw9fjW75aYB8UyVRZo6fCr
2026-10-29 09:44:49 +00:00
Claude f8955c2ef2 Keep ComfyUI's console on disk (Previous run in the Log window) and log memory at LoRA load and sampler start
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BzJLJWVu7RZnbRqm3nYAnX
2026-10-27 09:03:30 +00:00
Claude 9f8705dd63 Shot camera and SLA need H3-only traits: hidden on LTX (node fields can now require a trait)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BzJLJWVu7RZnbRqm3nYAnX
2026-10-27 08:29:22 +00:00
Claude a9f5c9f35d VAE loader: pass the file header to the VAE (audio VAEs need their config)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BzJLJWVu7RZnbRqm3nYAnX
2026-10-27 08:06:34 +00:00
Claude 2ce29a0998 SLA rising sparsity: note the first real H3 result
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0176SHndmQj7T1VtWAWZiNcq
2026-10-25 16:16:02 +00:00
Claude 5bc3ce3f95 SLA: test that the rising-share field is drawn only for the sla method
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0176SHndmQj7T1VtWAWZiNcq
2026-10-25 16:06:31 +00:00
Claude 527d368c82 SLA: optional rising sparsity (sla_sparsity_start), GSRS idea; CPU test env script
Skip fewer attention blocks on early steps and more on late ones, in a straight
line up to sla_sparsity. Off at 0, untested on H3. Credit GSRS (arXiv 2610.21129).
tools/test_env.sh and tools/cpuforce.py set up the suite on a GPU-less machine.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0176SHndmQj7T1VtWAWZiNcq
2026-10-25 16:05:33 +00:00
Claude 3adebb731f Rename FunPack to FunUI (5.0.3 Classy Chopper): /fun endpoint, wink logo, load-time migration of old names
Old project/workflow names and browser settings migrate on load; /funpack redirects to /fun.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016acRss69eH1235eRdFTZHY
2026-10-25 13:11:03 +00:00
Claude 156f86e618 fix: checkpoint loader reports a missing VAE; a start picture no pipeline slot takes is named as unused
Tests: VAE, LoRA (7 key dialects), CLIP and checkpoint loaders; every preset x every kind of picked file is used or named.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WP6HMLLhRYcghAEXK4KLwK
2026-10-21 04:36:20 +00:00
Claude c089fe3a9a test: every diffusion loader option loads a tiny H3 file; every reference kind reaches its H3 socket and passes ComfyUI validation
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WP6HMLLhRYcghAEXK4KLwK
2026-10-21 04:29:20 +00:00
Claude ab100f85a4 LoRA training add-on: sets, tagger, trainer node on ComfyUI's Train LoRA, H3 support
Backend only (window follows). Switched on in Settings > Modules. H3 repairs proven on a tiny H3 on CPU, UNTESTED on real weights.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MpnF7GwNGMbngU11wjym7P
2026-10-17 15:05:57 +00:00
digital-garbageandClaude Sonnet 5.5 eeaf72ba31 fix: the fp32-over-bf16 warning says it runs slowly
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-17 03:21:20 +03:00
digital-garbageandClaude Sonnet 5.5 ad6d6b54e6 fix: plainer loader warning for fp32 compute over bf16 weights
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-17 03:20:57 +03:00
digital-garbageandClaude Sonnet 5.5 69eaa15ce0 fix: the fp32-over-bf16 loader note no longer claims it does not run
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-17 03:20:01 +03:00
digital-garbageandClaude Sonnet 5.5 4549ffc2b3 feat: Diffusion Model Loader says when fp32 compute is picked over bf16 weights
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-17 03:19:36 +03:00
digital-garbageandClaude Sonnet 5.5 6e5035671a fix: provider errors say 'failed while' once; import guard covers class/for/match bodies; diffusion picker lists .gguf again so a saved pick gets the install message; stale-module wording true for broken modules (attacker round 17)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 22:58:01 +03:00
digital-garbageandClaude Sonnet 5.5 5c2588a9b2 fix: import guard sees every import form and is shown to fail; diffusion picker hides .gguf with no backend; provider errors are logged; stale module ids said at the modifiers node too (attacker round 16)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 22:36:30 +03:00
digital-garbageandClaude Sonnet 5.5 a88b814b28 fix: modules no longer import each other (shared taste helper is a plain file); UI gates match real ids and are tested; boot cannot stall; empty pipelines refused; truncated .gguf listing says so (attacker round 15)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 22:10:20 +03:00
digital-garbageandClaude Sonnet 5.5 0f86ffd909 fix: .gguf listing walks by folder identity (sibling cycles, root links, case); dangling links not listed (attacker round 14)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 20:26:47 +03:00
digital-garbageandClaude Sonnet 5.5 3042532c0c fix: a symlinked folder aliasing a sibling is listed under both names; an undequantizable gguf tensor type names the file (attacker round 13)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 19:28:02 +03:00
digital-garbageandClaude Sonnet 5.5 76d534adb7 fix: symlink loops no longer stall the .gguf listing; a name with .. is refused after a link too (attacker round 12)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 19:17:23 +03:00
digital-garbageandClaude Sonnet 5.5 6416c1b032 fix: a symlinked .gguf inside the model folder loads again (attacker round 11)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 19:09:22 +03:00
digital-garbageandClaude Sonnet 5.5 b08fba7d58 fix: .gguf loading stays inside the model folder, ignores a Manager-disabled pack, keeps big fp32 values finite; text-encoder picker hides .gguf without the pack; truthful refusal text (attacker round 10)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 19:03:32 +03:00
digital-garbageandClaude Sonnet 5.5 40c3366214 fix: a text encoder with vision tensors beside mapped text loads; clear messages for a pack missing its gguf library and for pack breakage (attacker round 9)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 18:53:45 +03:00
digital-garbageandClaude Sonnet 5.5 dfa8f8edf2 fix: a vision-projector .gguf is not accepted as a text encoder; a pack whose gguf library is missing says how to fix it (attacker round 8)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 18:41:18 +03:00
digital-garbageandClaude Sonnet 5.5 f37a647d85 fix: the ComfyUI-GGUF pack's (state dict, extras) result is unpacked; a text-encoder .gguf with no backend says the pack, not pip; the Enhance picker hides .gguf encoders without the pack; docs name the pack requirement, restart, thresholds (attacker round 7)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 18:33:07 +03:00
digital-garbageandClaude Sonnet 5.5 e5568ec02b fix: a text-encoder .gguf is refused when ComfyUI-GGUF is absent too; docs no longer overclaim Ready to generate, failure behaviour, rating thresholds, Export; Start shots hint shortened (attacker round 6)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 18:22:38 +03:00
digital-garbageandClaude Sonnet 5.5 08550e3569 fix: H3 second-pass preset held back; unsupported .gguf text encoders are refused instead of loading as the wrong model; Start shots hint is true for H3; LTX module titled UNTESTED; docs corrected (failure behaviour, Taste key location, install caution, gguf, second pass) (attacker round 5)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-15 18:10:57 +03:00
digital-garbage 41a75c3cd6 fix: server messages name the right Settings pages; effect, add-on, shot camera, LoRA, VAE, erase and SLA hints say what they do 2026-10-15 13:08:08 +03:00
digital-garbageandClaude Opus 5.5 8ce237dcdd feat: loader logs the file's real weight format (int8 file at bf16 says int8); warns when a cu12 torch on Blackwell halves int8 speed
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 18:34:58 +03:00
digital-garbageandClaude Haiku 5.5 77748b7a7e fix: dense steps '0' means none, as the tooltip and the earlier commit say (was flagged as a bad token)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
2026-10-10 16:15:51 +03:00
digital-garbageandClaude Haiku 5.5 78a418812b fix: SLA light references on comfy_kitchen pool in plain torch, so they run without Triton
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
2026-10-10 16:13:15 +03:00
digital-garbageandClaude Opus 5.5 f313ed6ef6 fix: Triton SLA no longer falls back to dense when K arrives in another layout than Q
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 01:52:37 +03:00
digital-garbageandClaude Opus 5.5 71892d05fd fix: SLA dense steps count from 1 like a person does ('1' is the first; 0 means none in both fields); the run log names the steps that ran at full attention
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 01:49:15 +03:00
digital-garbageandClaude Opus 5.5 8fabfe3b3f feat: the model loader's fields read in order: file, storage, attention, then SLA as its choices narrow it
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 01:40:01 +03:00
digital-garbageandClaude Opus 5.5 fc92e5869b feat: the model loader shows each SLA field only in the mode it belongs to (sparsity for sla/vsa, tau for sol-attn, block size and stabilize for Triton, tail for comfy_kitchen)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 01:38:45 +03:00
digital-garbageandClaude Opus 5.5 8eaf5be5f4 feat: SLA method choice (sla / sol-attn / vsa via ComfyUI's Model Sparse Attention); first step dense by default; dense steps read from the schedule, so a cancel or a two-call sampler no longer shifts them
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 01:30:59 +03:00