ComfyUI-bleh

BlehAdvancedAttentionSampler

TBD

Sampling/Custom Sampling/Samplers · 12 inputs · 0 parameters · 1 output

BlehBlendConditioning

Blends conditioning_1 with conditioning_2. Unlike the ComfyUI builtin node, this can handle multiple conditioning items on both side and respects time ranges but not other potential ranges (I.E. regional conditioning). Note however that conditionings are generally encoded as sequences of tokens, so something like 'a cute dog' blended with 'sketch of a cat' is likely to be doing something like blending 'a' with 'sketch', 'cute' with 'of', etc.

Conditioning · 1 output

BlehBlockCFG

Experimental model patch that attempts to guide either cond (positive prompt) or uncond (negative prompt) away from its opposite. In other words, when applied to cond it will try to push it further away from what uncond is doing and vice versa. Stronger effect when applied to cond or output blocks. The defaults are reasonable for SD 1.5 (or as reasonable as weird stuff like this can be). Enter comma separated blocks numbers starting with one of Iinput, Output or Middle like i4,m0,o4. You may also use rather than a block number to select all blocks in the category, for example i, o matches all input and all output blocks. The patch can be applied to the same model multiple times. Is it good, or even doing what I think? Who knows! Both positive and negative scales seem to have positive effect on the generation. Low negative scales applied to cond seem to make the generation bright and colorful. Note: Probably only works with SD 1.x and SDXL. Middle block patching will probably only work if you have FreeUAdvanced installed. Note: Doesn't work correctly with Tiled Diffusion when using tile batch sizes over 1.

BlehBlockOps

The top level YAML should consist of a list of objects with a condition if, a list of ops that run if the condition succeeds. Objects then and else also take the same form as the top level object and apply when the if condition matches (or not in the case of else). All object fields (if, then, else, ops) are optional. An empty object is valid, it just doesn't do anything. yaml - if: cond1: [value1, value2] cond2: value # Values may be specified as a list or single item. ops: [[opname1, oparg1, oparg2], [opname2, oparg1, oparg2]] then: if: [[opname1, oparg1, oparg2]] # Conditions may also be specified as a list. ops: [] # and so on else: ops: [] Note: Blocks match by default, conditions restrict them. So a block with no if matches everything. 1. bislerp: Interpolates between tensors a and b using normalized linear interpolation. 2. colorize: Supposedly transfers color. May or may not work that way. 3. cosinterp: Cosine interpolation. 4. cuberp 5. hslerp: Hybrid Spherical Linear Interporation, supposedly smooths transitions between orientations and colors. 6. inject: Inject just adds the value scaled by the ratio, so if ratio is 1.0 this simply adds it. 7. lerp: Linear interpolation. 8. lineardodge: Supposedly simulates a brightning effect. 1. none 2. bandpass 3. lowpass: Allows low frequencies and suppresses high frequencies. 4. highpass: Allows high frequencies and suppresses low frequencies. 5. passthrough: Maybe doesn't do anything? 6. gaussianblur: Blur. 7. edge: Edge enhance. 8. sharpen: Sharpens the target. 9. multilowpass: The multi versions apply to multiple bands.…

Bleh/Model Patches · 2 inputs · 1 parameter · 1 output

BlehCast

UNSAFE: This node allows casting its input to any type. NOTE: This does not actually change the data in any way, it just allows you to connect its output to any input. Only use if you know for sure the data is compatible.

Hacks · 1 output

BlehCFGInitSampler

Sampler wrapper that allows skipping some number of initial steps, similar to CFGZeroStar zero-init.

Sampling/Custom Sampling/Samplers · 2 inputs · 1 parameter · 1 output

BlehContrastiveOrthoCFG

Contrastive orthogonal CFG. CFG variant that allows you to control the scale of negative and positive features individually. For the most predictable effects, use positive scales and pre_cfg mode with use_noise disabled. This is just a slower version of CFG if you use the default parameters and positive/negative scales of 1.0.

Advanced/Guidance · 2 inputs · 8 parameters · 1 output

BlehDeepShrink

AKA PatchModelAddDownScale AKA Kohya Deep Shrink. Compared to the built-in Deep Shrink node this version has the following differences: 1. Instead of choosing a block to apply the downscale effect to, you can enter a comma-separated list of blocks. This may or not actually be useful but it seems like you can get interesting effects applying it to multiple blocks. Try 2,3 or 1,2,3. 2. Adds a startfadeoutpercent input. When this is less than endpercent the downscale will be scaled to end at endpercent. For example, if downscalefactor=2.0, startpercent=0.0, endpercent=0.5 and startfadeoutpercent=0.0 then at 25% you could expect downscalefactor to be around 1.5. This is because we are deep shrinking between 0 and 50% and we are halfway through the effect range. (downscalefactor=1.0 would of course be a no-op and values below 1 don't seem to work.) 3. Expands the options for upscale and downscale types, you can also turn on antialiasing for bicubic and bilinear modes. Notes: It seems like when shrinking multiple blocks, blocks downstream are also affected. So if you do x2 downscaling on 3 blocks, you are going to be applying x2 3 downscaling to the lowest block (and maybe downstream ones?). I am not 100% sure how it works, but the takeway is you want to reduce the downscale amount when you are downscaling multiple blocks. For example, using blocks 2,3,4 and a downscale factor of 2.0 or 2.5 generating at 3072x3072 seems to work pretty well. Another note is schedulers that move at a steady pace seem to produce better results when fading out the deep shrink effect. In other words,…

BlehDisableNoise

Allows setting a seed even when disabling noise. Used for SamplerCustomAdvanced or other nodes that take a NOISE input.

Sampling/Custom Sampling/Noise · 0 inputs · 2 parameters · 1 output

BlehDiscardPenultimateSigma

Compact description unavailable for this node.

BlehEnsurePreviewer

This node ensures Bleh is used for previews. Can be used if other custom nodes overwrite the Bleh previewer. It will pass through any value unchanged.

Hacks · 1 output

BlehFixGuiderPreviewing

Wraps a guider to give the Bleh previewing system a hint about the latent shapes. Only necessary for models like LTX-2 which use nested tensors.

Hacks · 2 inputs · 1 parameter · 1 output

BlehForceSeedSampler

ComfyUI has a bug where it will not set any seed if you have add_noise disabled in the sampler. This node is a workaround for that which ensures a seed alway gets set.

Sampling/Custom Sampling/Samplers · 1 input · 1 parameter · 1 output

BlehGlobalSageAttention

Deprecated: Prefer using BlehSageAttentionSampler if possible. This node allows globally replacing ComfyUI's attention with SageAtteniton (performance enhancement). Requires SageAttention to be installed into the ComfyUI Python environment. IMPORTANT: This is not a normal model patch. For settings to apply (including toggling on or off) the node must actually be run. If you toggle it on, run your workflow and then bypass or mute the node this will not actually disable SageAttention.

Hacks · 2 inputs · 1 parameter · 1 output

BlehHyperTile

Adds the ability to set a seed and timestep range that HyperTile gets applied for. Not well tested, and I just assumed the Inspire version works which may or may not be the case. It is also possible to set an interval for HyperTile steps, this time it is just normal sampling steps that match the timestep range. The first sampling step that matches the timestep range always applies HyperTile, after that the following behavior applies: If the interval is positive then you just get HyperTile every interval steps. It is also possible to set interval to a negative value, for example -3 would mean out of every three steps, the first two have HyperTile and the third doesn't. Note: Timesteps start from 999 and count down to 0 and also are not necessarily linear. Figuring out exactly which sampling step a timestep applies to is left as an exercise for you, dear node user. As an example, Karras and exponential samplers essentially rush to low timesteps and spend quite a bit of time there. HyperTile credits: The node was originally taken by Comfy from taken from: https://github.com/tfernd/HyperTile/ Then the Inspire node pack took it from the base ComfyUI node: https://github.com/ltdrdata/ComfyUI-Inspire-Pack Then I took it from the Inspire node pack. The original license was MIT so I assume yoinking it into this repo is probably okay.

BlehImageAsLatent

This node allows you to rearrange an IMAGE to look like a LATENT. Can be useful if you want to apply some latent operations to an IMAGE. Can be reversed with the BlehLatentAsImage node.

Latent/Advanced · 1 input · 1 parameter · 1 output

BlehInsaneChainSampler

Note: I'd recommend using my Overly Complicated Sampling node pack over this. It generally has better tools for scheduling samplers. A picture is worth a thousand words, so: This will use heunpp2 for the first five steps, eulerancestral for the next five, and dpmpp2m for however many remain. This is basically the same as chaining a bunch of samplers together and manually setting the start/end steps. Note: Even though the dpmpp2m insane chain sampler node has steps=1 it will run for five steps. This is because the requirement of fifteen total steps must be fulfilled and... you can't really sample stuff without a sampler. Also note progress might be a little weird splitting sampling up like this.

Sampling/Custom Sampling/Samplers · 2 inputs · 1 parameter · 2 outputs

BlehLatentAsImage

This node lets you rearrange a LATENT to look like an IMAGE. Note: It does not respect anything like masks or latent selection metadata (from nodes like LatentFromBatch) that might exist.

Latent/Advanced · 2 inputs · 1 parameter · 1 output

BlehLatentBlend

Allows blending latents using any of the blending modes available.

Latent · 1 output

BlehLatentOps

Basically the same as BlehBlockOps, except the condition type will be latent. Obviously stuff involving steps, percentages, etc do not apply. This node allows you to apply the blending/filtering/scaling operations to a latent.

Latent · 2 inputs · 1 parameter · 1 output

BlehLatentScaleBy

Like the builtin LatentScaleBy node, however it allows setting the horizontal and vertical scaling types and scales independently as well as allowing providing an extended list of scaling options. Can also be useful for testing what different types of scaling or enhancement effects look like.

Latent · 3 inputs · 3 parameters · 1 output

BlehModelPatchConditional

Note: Very experimental. This node takes a default model and a matched model. When the interval or start/end percentage match, the matched model will apply, otherwise the default one will. This can be used to apply something like HyperTile, Self Attention Guidance or other arbitrary model patches conditionally. The first sampling step that matches the timestep range always applies matched, after that the following behavior applies: If the interval is positive then you just get matched every interval steps. It is also possible to set interval to a negative value, for example -3 would mean out of every three steps, the first two use matched and the third doesn't. Notes and limitations: Not all types of model modifications/patches can be intercepted with a node like this. You also almost certainly can't use this to mix different models: both inputs should be instances of the same loaded model. It's also probably a bad idea to apply further patches on top of the BlehModelPatchConditional node output: it should most likely be the last thing before a sampler or something that actually uses the model.

BlehModelPatchFastTerminate

Patches a model to check if processing is interrupted at the start of every block. Makes interrupting generations more responsive on supported models (mainly useful for video models that might take 40+ second for a step). Should support most existing models.

Hacks · 1 input · 0 parameters · 1 output

BlehModelProcessLatentIn

Advanced node that can be used to scale a raw latent for model input. Generally only needed if you're doing something that bypasses the normal latent input mechanisms.

Latent/Advanced · 2 inputs · 0 parameters · 1 output

BlehModelProcessLatentOut

Advanced node that can be used to scale a latent to the correct range for output. Generally only needed if you're doing something that bypasses the normal latent output mechanisms.

Latent/Advanced · 2 inputs · 0 parameters · 1 output

BlehPlug

This node can be used to plug up an input but act like the input was not actually connected. Can be used to prevent something like Use Everywhere nodes from supplying an input without having to set up blacklists or other configuration.

Hacks · 1 output

BlehRefinerAfter

Allows switching to another model at a certain point in sampling. Only works with models that are closely related as the sampling type and conditioning must match. Can be used to switch to a refiner model near the end of sampling.

Bleh/Model Patches · 3 inputs · 1 parameter · 1 output

BlehSageAttentionSampler

Sampler wrapper that enables using SageAttention (performance enhancement) while sampling is in progress. Requires SageAttention to be installed into the ComfyUI Python environment.

Sampling/Custom Sampling/Samplers · 2 inputs · 2 parameters · 1 output

BlehSetSamplerPreset

This node allows setting a custom sampler as a preset that can be selected in nodes that don't support custom sampling (FaceDetailer for example). This node needs to run at least once with any preset changes before actual sampling begins. The any_input input acts as a passthrough so you can do something like pass your model or latent through before you start sampling to ensure the node runs. You can also connect something like an integer or string to the dummy_opt input and change it to force the node to run again. The number of presets can be adjusted (and the whole feature disabled if desired) by setting the environment variable COMFYUI_BLEH_SAMPLER_PRESET_COUNT. WARNING: Since the input and output are wildcards, this bypasses ComfyUI's normal type checking. Make sure you connect the output to something that actually accepts the input type.

Hacks · 1 output

BlehSetSigmas

Advanced node that allows manipulating SIGMAS. For example, you can manually enter a list of sigmas, insert some new sigmas into existing SIGMAS, etc.

Sampling/Custom Sampling/Sigmas · 4 inputs · 2 parameters · 1 output

BlehTAEVideoDecode

Compact description unavailable for this node.

BlehTAEVideoEncode

Compact description unavailable for this node.