This node modifies only the active text region to affect prompt adherence and edit behavior. The padding region is left untouched. Active region end is auto-detected from attention mask.

Installation

Navigate to your ComfyUI custom nodes folder:
```
cd ComfyUI/custom_nodes/
```

Clone this repository:

git clone https://github.com/capitan01R/ComfyUI-Flux2Klein-Enhancer.git

Restart ComfyUI

Nodes

New Nodes Update #2.1.0

FLUX.2 Klein Enhancer

General-purpose conditioning enhancement for both text-to-image and image editing workflows.

Parameters

Parameter	Default	Range	Description
`magnitude`	1.0	0.0 to 3.0	Direct scaling of active region embeddings. Values above 1.0 increase prompt influence, below 1.0 decrease it.
`contrast`	0.0	-1.0 to 2.0	Amplifies differences between tokens. Positive values sharpen concept separation, negative values blend them.
`normalize_strength`	0.0	0.0 to 1.0	Equalizes token magnitudes. Higher values balance emphasis across all tokens in the prompt.
`edit_text_weight`	1.0	0.0 to 3.0	Image edit mode only. Values below 1.0 preserve more of the original image, above 1.0 follows the prompt more strongly.
`active_end_override`	0	0 to 512	Manual override for active region end. 0 = auto-detect from attention mask.
`low_vram`	False	True/False	Use float16 computation on CUDA devices.
`device`	auto	auto/cpu/cuda:N	Compute device selection.
`debug`	False	True/False	Prints tensor statistics and modification details to console.

FLUX.2 Klein Detail Controller

Regional control over prompt conditioning. Divides active tokens into front/mid/end sections.

Parameters

Parameter	Default	Range	Description
`front_mult`	1.0	0.0 to 3.0	Multiplier for first 25% of active tokens (typically subject/main concept).
`mid_mult`	1.0	0.0 to 3.0	Multiplier for middle 50% of active tokens (typically details/modifiers).
`end_mult`	1.0	0.0 to 3.0	Multiplier for last 25% of active tokens (typically style/quality terms).
`emphasis_start`	0	0 to 200	Start position of custom emphasis region.
`emphasis_end`	0	0 to 200	End position of custom emphasis region. 0 = disabled.
`emphasis_mult`	1.0	0.0 to 3.0	Multiplier for the custom emphasis region.
`low_vram`	False	True/False	Use float16 computation on CUDA devices.
`device`	auto	auto/cpu/cuda:N	Compute device selection.
`debug`	False	True/False	Prints debug information to console.

Example of purely seperating the:

front(subject)/mid(details)/end(style) [results in these examples only meant to showcase the capability of pure seperation]

Prompt: turn only the ground into a mirror surface reflecting the sky, keep the full dog and its body unchanged and add the dog's reflection below

Added new support node to the FLUX.2 Detail Controller

Target: A user friendly node to help users comfortably section their prompts to add target mult to them using the FLUX.2 Detail Controller.
The node is sectioned to three parts front/mid/end just like the Detail Controller node + a combined box "Basically for Vanilla use".

Preserve Original - Solving FLUX Klein's Preservation Problem

FLUX Klein has a consistency problem. Sometimes it nails the preservation of subjects and objects. Sometimes it completely ignores what you're trying to keep and generates something else entirely. There was no native way to control this.

This node exposes preservation control that FLUX Klein doesn't provide. You can now control exactly how much original structure is maintained versus how much the prompt can modify the generation.

The Modes

dampen (Recommended)

Reduces modification strength before applying changes. This is the most reliable mode for precise preservation.

For consistent identity/object preservation: 1.20 to 1.30

This range provides solid preservation without killing your prompt's ability to make changes.

linear

Applies full modifications, then blends the result back with the original. Less predictable than dampen but useful when you want aggressive changes with some safety rails.

hybrid

Dampens parameters first, then blends the result. Combines both approaches. Can be overkill for most cases.

blend_after

Same as linear, just a different name.

Usage

The optimal value changes depending on your prompt and what you're preserving.

1.20-1.30 (dampen): Recommended starting point for solid preservation
1.40-1.50: Tighter control when needed, very prompt-dependent
0.0-1.0: Standard range from full enhancement to balanced preservation
Negative values: Experimental, not recommended for production

You might need 1.25 for one generation and 1.45 for another. That's normal. The node gives you the precision to dial in what each specific prompt needs.

Why It Matters

FLUX Klein doesn't expose preservation controls. This node creates that capability from scratch. Instead of rolling the dice on whether your subject's identity or object consistency survives the generation, you can lock it down while still letting the prompt do its work.

Addidtion:

Examples:

How It Works

Magnitude

Direct scaling of all embedding vectors in the active region:

active = active * magnitude

A value of 1.25 results in 25% stronger conditioning signal to the diffusion model. This directly affects cross-attention key magnitudes.

Contrast

Computes the mean embedding across the sequence, then amplifies deviations from that mean:

seq_mean = active.mean(dim=1, keepdim=True)
deviation = active - seq_mean
active = seq_mean + deviation * (1.0 + contrast)

This increases the distinctiveness of individual token embeddings, helping separate concepts in complex prompts.

Normalize Strength

Equalizes token magnitudes toward a uniform value:

token_norms = active.norm(dim=-1, keepdim=True)
mean_norm = token_norms.mean()
normalized = active / token_norms * mean_norm
active = active * (1.0 - normalize_strength) + normalized * normalize_strength

This prevents any single token from dominating the conditioning signal.

Image Edit Mode Detection

The node automatically detects image edit mode by checking for reference_latents in the conditioning metadata. When detected, edit_text_weight provides additional scaling.

Active Region Detection

The active region end is auto-detected from the attention mask in metadata:

attn_mask = meta.get("attention_mask", None)
nonzero = attn_mask[0].nonzero()
active_end = int(nonzero[-1].item()) + 1

This ensures only meaningful tokens are modified, regardless of prompt length.

Presets

Text-to-Image

              BASE   GENTLE   MOD   STRONG   AGG     MAX    CRAZY
              ----    ----    ----    ----    ----    ----    ----
magnitude:    1.20    1.15    1.25    1.35    1.50    1.75    2.50
contrast:     0.00    0.10    0.20    0.30    0.40    0.60    1.20
normalize:    0.00    0.00    0.00    0.15    0.25    0.35    0.60
edit_weight:  1.00    1.00    1.00    1.00    1.00    1.00    1.00

Image Edit

              PRESERVE   SUBTLE   BALANCED   FOLLOW   FORCE
              --------   ------   --------   ------   -----
magnitude:       0.85     1.00       1.10     1.20    1.35
contrast:        0.00     0.05       0.10     0.15    0.25
normalize:       0.00     0.00       0.10     0.10    0.15
edit_weight:     0.70     0.85       1.00     1.25    1.50

Usage Examples

Text-to-Image: Stronger Prompt Adherence

magnitude: 1.25
contrast: 0.20
normalize_strength: 0.00
edit_text_weight: 1.00

Text-to-Image: Complex Prompt with Multiple Concepts

magnitude: 1.35
contrast: 0.30
normalize_strength: 0.15
edit_text_weight: 1.00

Image Edit: Follow Prompt More

magnitude: 1.20
contrast: 0.15
normalize_strength: 0.10
edit_text_weight: 1.25

Image Edit: Preserve Original More

magnitude: 0.85
contrast: 0.00
normalize_strength: 0.00
edit_text_weight: 0.70

Image Edit: Force Prompt Adherence

magnitude: 1.35
contrast: 0.25
normalize_strength: 0.15
edit_text_weight: 1.50

Debugging

Enable debug: True to see console output:

==================================================
Flux2KleinEnhancer Item 0
==================================================
Shape: torch.Size([1, 512, 12288])
Active region: 0 to 71
Edit mode: True
Active std: 42.4100
Padding std: 2.3094

Before modifications:
  Active region mean norm: 893.77

Contrast (+0.20): deviation scaled by 1.20

Magnitude (1.25): all active tokens scaled

Edit text weight (1.15): applied for image edit mode

Final state:
  Active region mean norm: 893.77 -> 1284.56
  Output change: mean=42.5341, max=1506.23

The Output change line confirms the conditioning tensor was modified. If it shows 0.000000, no changes were applied.

Technical Details

Model: FLUX.2 Klein 9B
Text Encoder: Qwen3 8B (4096 hidden dim, 36 layers)
Conditioning Shape: [batch, 512, 12288]
Joint Attention Dim: 12288
Active Region: Dynamic, detected from attention mask (typically 0-77)
Guidance Embeds: False (step-distilled model, no CFG)

Visual Results: Vanilla vs. With Flux2Klein-Enhancer

Exact same workflow, seed and prompt - only difference is using the node or not.
Click images to view full size.

Source Photo

Comparison 1

Prompt: turn only the ground into a mirror surface reflecting the sky, keep the full dog and its body unchanged and add the dog's reflection below

Vanilla Flux.2 Klein	With Enhancer Node

Comparison 2

Prompt: replace the grass with shallow ocean water and add realistic water reflections of the dog, keep the sunny lighting

Vanilla	With Enhancer Node

Acknowledgments

Built through empirical analysis of FLUX.2 Klein's conditioning structure using diagnostic tools to inspect actual tensor shapes and statistics during inference.