Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > FunUI > FunUI Negative Erase

ComfyUI Node: FunUI Negative Erase

Class Name

FunUINegativeErase

Category
FunUI/Conditioning
Author
digital-garbage (Account age: 1276days)
Extension
FunUI
Latest Updated
2026-11-03
Github Stars
0.02K

How to Install FunUI

Install this extension via the ComfyUI Manager by searching for FunUI
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter FunUI in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

FunUI Negative Erase Description

Removes or subtracts negative-prompt directions from positive embeddings for CFG 1 workflows.

FunUI Negative Erase:

FunUINegativeErase is an experimental conditioning tool that gives negative prompts a useful role in workflows running at classifier-free guidance (CFG) 1, such as the supplied MiniMax H3 pipeline, where ordinary negative-prompt guidance has no effect. Instead of relying on the sampler to steer away from unwanted content, it changes the encoded representation of the positive prompt before sampling. You can think of encoded text as a set of directions describing visual ideas: this node uses the negative prompt to establish an unwanted direction, then removes or subtracts that direction from the positive prompt’s word representations. It offers projection-based removal and subtraction-based adjustment, with optional preservation of the original size of each word representation. This is an embedding-level experiment rather than literal word deletion, image editing, or guaranteed concept removal; a sentence’s meaning is not necessarily represented as one clean, separable direction, so visual results require testing.

FunUI Negative Erase Input Parameters:

positive

A required CONDITIONING connection containing the encoded positive prompt that you want to adjust. This is the prompt representation used to describe what you want to generate, not a plain-text field or an image. The node modifies its word representations according to the negative direction, selected method, and strength. At strength 0, this conditioning passes through unchanged. The supplied tests also show that, when compatible MiniMax token-layout metadata identifies image and text rows, image rows are preserved while text rows can be modified. Connect the positive-conditioning branch you intend to send onward for generation; in the provided H3 pipeline, this comes after prompt-markup application. No default value or numerical range is defined for this connection.

negative

A required CONDITIONING connection containing the encoded negative prompt, which supplies the direction to remove from the positive prompt’s word representations. Unlike ordinary negative guidance, this node uses that representation directly to modify positive conditioning, so it can have a role even when sampling at CFG 1. This connection expects encoded conditioning rather than raw negative-prompt text. Use conditioning from a compatible text-encoding setup: the supplied tests show that incompatible embedding widths prevent modification, while a zero-valued negative representation produces no usable direction. No default value or numerical range is defined for this connection.

strength

A required FLOAT controlling how strongly the negative direction changes the positive prompt’s word representations. The default is 0.0, the minimum is 0.0, the maximum is 2.0, and the adjustment step is 0.05. At 0.0, the node is disabled and returns the original positive conditioning with the status off. Increasing the value applies more of the selected operation. In project mode, 1.0 removes the component aligned with the negative direction; the supplied test confirms this with size restoration disabled. Values above 1.0 go beyond that full-removal setting rather than simply removing additional literal words. In subtract mode, strength controls the pull away from the negative direction. Treat this as an embedding adjustment, not a percentage guarantee that an unwanted object or style will disappear.

mode

A required COMBO selecting the adjustment method, with project as the default and project and subtract as the documented choices. The project method removes the part of each word representation that lies along the negative prompt’s direction, making it a targeted way to reduce that shared component. The supplied tests show that a representation already at right angles to that direction remains unchanged when size restoration is disabled. The subtract method instead pulls every word representation away from the negative direction by an amount related to that word representation’s own size. This means the methods can affect the same prompt differently: projection removes an existing aligned component, whereas subtraction applies a directional push. Neither method guarantees a specific visual outcome.

keep_size

A required BOOLEAN with a default of True and available values True and False. When True, the node attempts to restore each adjusted word representation to its original size, so the intended change is mainly to its direction rather than its overall magnitude. Here, “size” means the length of an encoded vector, not prompt length, image dimensions, or object size. When False, that restoration is disabled, leaving the magnitude changes caused by the selected operation in place. The tests include a gain-limit safeguard for size restoration, so you should understand this option as controlled restoration rather than an unconditional guarantee of exact original size in every case.

FunUI Negative Erase Output Parameters:

positive

The CONDITIONING result to connect to the downstream positive-conditioning path. When the operation can be applied, this contains positive conditioning adjusted using the negative prompt’s direction. It is not an edited image, a rewritten prompt, or a replacement negative prompt. At strength 0, it is the original positive conditioning passed through untouched. The supplied tests also confirm unchanged output when the negative representation has no usable direction or when incompatible embedding widths prevent modification. Check the accompanying status to distinguish a successful adjustment from a pass-through result.

status

A STRING reporting the operation’s outcome. The exact value off indicates that strength is 0 and no modification was requested. The supplied tests establish that diagnostic messages can include no usable direction when the negative representation cannot provide a direction, or could not be modified when modification cannot be applied, such as with mismatched embedding widths. A tested projection result includes project in its status, but the supplied context does not establish the complete wording of every possible message. The node also sends non-off notes through its logging mechanism, so a logged note alone does not necessarily indicate a failure.

FunUI Negative Erase Usage Tips:

  • Start with mode set to project, keep_size set to True, and strength set to 0.0 for a baseline. Keep your seed and other generation settings fixed, then increase strength in small increments to judge what the conditioning adjustment actually changes.
  • For CFG 1 workflows such as the provided H3 pipeline, place the node on the positive-conditioning path before sampling. Keep any negative connection required by your sampler in place; this node does not remove that interface requirement.
  • Encode your positive and negative prompts with a compatible text-encoding setup. Matching their representation format avoids the embedding-width mismatch demonstrated in the tests.
  • Compare project and subtract at the same strength. Projection removes the aligned component, while subtraction pushes each word representation away from the negative direction, so switching modes is a meaningful change rather than a cosmetic setting.
  • Treat strength 1.0 as the full projection-removal reference point, not a recommended setting for every prompt. Evaluate values above 1.0 cautiously because they go beyond that reference and may alter desired prompt information.
  • Compare keep_size True and False if the adjustment changes the overall prompt response in an unwanted way. Keeping it enabled aims to preserve representation magnitude; disabling it lets you evaluate the operation without size restoration.
  • Test one unwanted concept at a time when assessing behavior. This makes comparisons easier to interpret, especially because the module is experimental and does not establish that complex sentences map to cleanly separable semantic directions.
  • Inspect status whenever the result appears unchanged. An unchanged image alone cannot tell you whether the node was disabled, could not modify the conditioning, or made a conditioning change with little visible effect.

FunUI Negative Erase Common Errors and Solutions:

off

  • Explanation: This is a normal status, not an error. Strength is 0.0, so the node deliberately returns your positive conditioning unchanged.
  • Solution: Increase strength above 0.0 if you want an adjustment. Leave it at 0.0 when you need a disabled baseline for comparison.

no usable direction

  • Explanation: This diagnostic phrase means the negative conditioning did not provide a usable direction for the operation. The supplied test demonstrates this with an all-zero negative representation; the positive conditioning remains unchanged.
  • Solution: Check that negative is connected to correctly encoded conditioning containing your intended negative prompt. Avoid substituting zero-filled placeholder conditioning, then inspect the status again after encoding a meaningful prompt.

could not be modified

  • Explanation: This diagnostic phrase indicates that the operation could not modify the supplied positive conditioning. The supplied test demonstrates this when the positive and negative embeddings have different widths, meaning their encoded representations are incompatible.
  • Solution: Re-encode both prompts using a compatible encoder setup and check that neither branch comes from an unrelated model or conditioning format. Avoid combining conditioning solely because both sockets share the CONDITIONING type.

Little or no visible effect despite an active setting

  • Explanation: This is a troubleshooting symptom, not a documented error string. The node adjusts encoded prompt representations rather than deleting visual concepts directly, so a valid operation may have little visible effect. The module is explicitly experimental.
  • Solution: Confirm that the node’s positive output actually reaches the generation path, check status, and compare fixed-seed results at several strengths. Test both modes and simplify the negative prompt to make the comparison easier to interpret.

Desired content changes along with unwanted content

  • Explanation: This is a possible result to watch for, not a documented error message. The negative direction may overlap with representations that also help describe desired content, and stronger adjustments do not guarantee more selective removal.
  • Solution: Reduce strength, compare project against subtract, and test a more focused negative prompt. Return to strength 0.0 for a baseline before deciding whether the adjustment improves your result.

FunUI Negative Erase Related Nodes

Go back to the extension to check out more related nodes.
FunUI
RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

FunUI Negative Erase