nodes/blepping-ComfyUI-bleh
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Start updating docs and example configuration files

Browse Source
This commit is contained in:
blepping
2025-05-22 04:25:55 -06:00
parent 54c6756c7e
commit daa97e4ac1
5 changed files with 42 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+15 -11
View File
@@ -6,17 +6,18 @@ For recent user-visible changes, please see the [ChangeLog](changelog.md).
## Features
1. Better TAESD previews (see below).
2. Allow setting seed, timestep range and step interval for HyperTile (look for the [`BlehHyperTile`](#blehhypertile) node).
3. Allow applying Kohya Deep Shrink to multiple blocks, also allow gradually fading out the downscale factor (look for the [`BlehDeepShrink`](#blehdeepshrink) node).
4. Allow discarding penultimate sigma (look for the `BlehDiscardPenultimateSigma` node). This can be useful if you find certain samplers are ruining your image by spewing a bunch of noise into it at the very end (usually only an issue with `dpm2 a` or SDE samplers).
5. Allow more conveniently switching between samplers during sampling (look for the [BlehInsaneChainSampler](#blehinsanechainsampler) node).
6. Apply arbitrary model patches at an interval and/or for a percentage of sampling (look for the [BlehModelPatchConditional](#blehmodelpatchconditional) node).
7. Ensure a seed is set even when `add_noise` is turned off in a sampler. Yes, that's right: if you don't have `add_noise` enabled _no_ seed gets set for samplers like `euler_a` and it's not possible to reproduce generations. (look for the [BlehForceSeedSampler](#blehforceseedsampler) node). For `SamplerCustomAdvanced` you can use `BlehDisableNoise` to accomplish the same thing.
8. Allows swapping to a refiner model at a predefined time (look for the [BlehRefinerAfter](#blehrefinerafter) node).
9. Allow defining arbitrary model patches (look for the [BlehBlockOps](#blehblockops) node).
10. Experimental blockwise CFG type effect (look for the [BlehBlockCFG](#blehblockcfg) node).
11. [SageAttention](https://github.com/thu-ml/SageAttention/) support either globally or as a sampler wrapper. Look for the [BlehSageAttentionSampler](#blehsageattentionsampler) and `BlehGlobalSageAttention` nodes.
* Better TAESD previews (see below).
* Visual previews for some audio models (currently only ACE-Steps).
* Allow setting seed, timestep range and step interval for HyperTile (look for the [`BlehHyperTile`](#blehhypertile) node).
* Allow applying Kohya Deep Shrink to multiple blocks, also allow gradually fading out the downscale factor (look for the [`BlehDeepShrink`](#blehdeepshrink) node).
* Allow discarding penultimate sigma (look for the `BlehDiscardPenultimateSigma` node). This can be useful if you find certain samplers are ruining your image by spewing a bunch of noise into it at the very end (usually only an issue with `dpm2 a` or SDE samplers).
* Allow more conveniently switching between samplers during sampling (look for the [BlehInsaneChainSampler](#blehinsanechainsampler) node).
* Apply arbitrary model patches at an interval and/or for a percentage of sampling (look for the [BlehModelPatchConditional](#blehmodelpatchconditional) node).
* Ensure a seed is set even when `add_noise` is turned off in a sampler. Yes, that's right: if you don't have `add_noise` enabled _no_ seed gets set for samplers like `euler_a` and it's not possible to reproduce generations. (look for the [BlehForceSeedSampler](#blehforceseedsampler) node). For `SamplerCustomAdvanced` you can use `BlehDisableNoise` to accomplish the same thing.
* Allows swapping to a refiner model at a predefined time (look for the [BlehRefinerAfter](#blehrefinerafter) node).
* Allow defining arbitrary model patches (look for the [BlehBlockOps](#blehblockops) node).
* Experimental blockwise CFG type effect (look for the [BlehBlockCFG](#blehblockcfg) node).
* [SageAttention](https://github.com/thu-ml/SageAttention/) support either globally or as a sampler wrapper. Look for the [BlehSageAttentionSampler](#blehsageattentionsampler) and `BlehGlobalSageAttention` nodes.
## Configuration
@@ -30,6 +31,9 @@ Restart ComfyUI to apply any new changes.
* Supports showing previews for more than the first latent in the batch.
* Supports throttling previews. Do you really need your expensive high quality preview to get updated 3 times a second?
The previewer can now show visual previews for ACE-Steps latents. If you want to disable that feature, you can add `aceaudio` to the
`blacklist_formats` list. For example if you are using a YAML configuration file you could do: `blacklist_formats: ["aceaudio"]`
**General settings defaults:**
|Key|Default|Description|
blehconfig.example.json
+6 -1
View File
@@ -10,6 +10,11 @@
"skip_upscale_layers": 0,
"compile_previewer": false,
"oom_fallback": "latent2rgb",
"oom_retry": true
"oom_retry": true,
"whitelist_formats": [],
"blacklist_formats": [],
"video_parallel": false,
"video_max_frames": -1,
"video_temporal_upscale_level": 0
}
}
blehconfig.example.yaml
+10
View File
@@ -21,6 +21,10 @@ betterTaesdPreviews:
# Minimum time between updating previews. The default will update the preview at most once per second.
throttle_secs: 1
# Can be set to use a different throttle time when using the fallback previewer
# (anything other than TAESD or TAEVID). If unset or null it will use throttle_secs.
throttle_secs_fallback: null
# When enabled and previewing batches, you will see previews spread across the batch. Otherwise it will be the first max_batch items.
maxed_batch_step_mode: false
@@ -28,6 +32,12 @@ betterTaesdPreviews:
# alone unless you know you need to change it. Previewing on CPU will likely be quite slow.
preview_device: null
# Can be set to override the previewer dtype (which probably defaults to float32).
# You may set it to a specific dtype: float32, float16, bfloat16
# Setting it to "keep" or null just leaves the dtype alone (which is probably float32).
# Setting it to "vae" will use whatever dtype ComfyUI is set to use for VAE.
preview_dtype: null
# Allows skipping upscale layers in the TAESD model, may increase performance when previewing large images or batches.
# May be set to -1 (conservative) or -2 (aggressive) to automatically calculate how many to skip. See README.md for details.
skip_upscale_layers: 0
changelog.md
+8
View File
@@ -2,6 +2,14 @@
Note, only relatively significant changes to user-visible functionality will be included here. Most recent changes at the top.
## 20250522
This set of changes involves refactoring parts of the previewer. Please create an issue if you experience problems.
* It's now possible to set the previewer dtype, see the `preview_dtype` setting. Note: Previews probably have been using float32 which is likely slower/more memory intensive than necessary. I'd recommend setting it to `vae`, `bfloat16` or `float16`.
* The previewer can show a visual representation of ACE-Steps latents (audio model). If you don't like it then you can add `aceaudio` to the `blacklist_formats` list in your configuration.
* You can set `throttle_secs_fallback` to use a different throttle setting for the fallback previewer (which includes stuff like the ACE-Steps previewer).
## 20250504
This is a fairly large set of changes. Please create an issue if you experience problems.
py/settings.py
+3 -4
View File
@@ -16,10 +16,9 @@ class Settings:
self.btp_max_batch = max(1, btp.get("max_batch", 4))
self.btp_max_batch_cols = max(1, btp.get("max_batch_cols", 2))
self.btp_throttle_secs = btp.get("throttle_secs", 1)
self.btp_throttle_secs_fallback = btp.get(
"throttle_secs_fallback",
self.btp_throttle_secs,
)
self.btp_throttle_secs_fallback = btp.get("throttle_secs_fallback")
if self.btp_throttle_secs_fallback is None:
self.btp_throttle_secs_fallback = self.btp_throttle_secs
self.btp_skip_upscale_layers = btp.get("skip_upscale_layers", 0)
self.btp_preview_device = btp.get("preview_device")
# default, keep, float32, float16, bfloat16