facok
4b99468127
Remove claim of "precise spatial surgery" and "not affecting appearance" — DiversityBoost also relies on random noise phase. Reword to emphasize frequency-domain precision and mathematical safety bounds vs indirect token-level perturbation.
ComfyUI-DiversityBoost
Restore composition diversity for distilled diffusion models. Training-free, single-step frequency-domain phase injection.
The Problem
Step-distilled models (FLUX2.[Klein], z-image-turbo, etc.) generate high-quality images in few steps, but suffer from composition collapse: different seeds produce nearly identical layouts. A portrait prompt always puts the subject dead-center; a landscape prompt always uses the same horizon line.
Root cause: distillation freezes the spatial distribution of token norms across seeds, locking the model into a single "average" composition regardless of the initial noise.
The Fix
DiversityBoost injects seed-dependent low-frequency phase from the initial noise into the model's denoised prediction at step 0 only, via a post-CFG hook.
In the frequency domain, amplitude encodes energy distribution (naturalness), while phase encodes spatial arrangement (composition). By rotating only low-frequency phase — while attenuating high-frequency amplitude to prevent deformities — different seeds produce genuinely different compositions again.
Zero model modification. Zero training. One node.
How It Works
- FFT the model's step-0 prediction and the initial noise
- Compute a shared rotation field (channel-mean phasors) — preserves inter-channel phase exactly (no color fringing)
- Apply a 6th-order Butterworth low-pass filter — only composition-scale frequencies are touched; object-scale details are protected
- Tanh soft-cap per-bin rotation — decouples diversity strength from worst-case rotation risk
- Attenuate high-frequency amplitude — prevents "frequency shearing" (body moves but fingers stay pinned)
- IFFT back to spatial domain
Step 0 only. The model's own attractor handles the rest.
Installation
cd ComfyUI/custom_nodes
git clone https://github.com/facok/ComfyUI-DiversityBoost.git
No extra dependencies — only requires PyTorch (ships with ComfyUI).
Usage
MODEL → [Diversity Boost] → MODEL → KSampler
- Add the Diversity Boost node (category:
sampling) - Connect your model to the input, output to KSampler
- Generate with different seeds — compositions will vary
Default settings work well. No tuning needed for most use cases.
Parameters
| Parameter | Default | Range | Description |
|---|---|---|---|
| strength | 1.0 | 0.0 – 1.0 | Phase rotation strength. 1.0 is safe with the default max_rotation cap. |
| n_periods | 2 | 1 – 10 | Max spatial periods to affect. Resolution-independent. |
| max_rotation | 1.5708 (π/2) | 0.0 – 3.14 | Per-bin rotation budget (radians). Caps worst-case rotation via tanh. |
| energy_compensate | False | — | Rescale output RMS to match original. Off by default. |
n_periods Guide
| Value | Effect |
|---|---|
| 1 | Ultra-conservative (global balance only) |
| 2 | Conservative (default, recommended) |
| 3 | Balanced (moderate diversity) |
| 4 | Aggressive (more diversity, higher risk) |
The cutoff adapts to image resolution automatically: freq_cutoff = n_periods / max(H, W). This means n_periods=2 always affects the same composition-scale frequencies regardless of whether you're generating 512x512 or 2048x2048.
max_rotation Presets
| Value | Effect |
|---|---|
| 1.00 | Conservative — less diversity, very safe |
| 1.57 (π/2) | Balanced (default, recommended) |
| 2.00 | Aggressive — more diversity, some risk |
| 0.00 | Disabled — original uncapped behavior |
DiversityBoost vs Dummy Token
Both aim to restore diversity in distilled models, but they work at fundamentally different levels:
| DiversityBoost | Dummy Token | |
|---|---|---|
| Mechanism | Rotate low-frequency phase in frequency domain | Add/modify padding tokens to shift attention context |
| Target | Spatial arrangement (composition skeleton) | Global attention bias (indirect) |
| Composition change | Direct and controllable (precise rotation angles) | Indirect, random (butterfly effect) |
| Diversity source | Each seed's unique noise phase | Random padding token content |
| Safety | Amplitude exactly preserved; Butterworth + tanh double-bounded | No mathematical guarantees |
| Prompt adherence | Unaffected (operates on spatial structure, not semantics) | May degrade (alters attention distribution) |
In short: DiversityBoost operates precisely on composition-scale phase in the frequency domain, with mathematically bounded safety guarantees. Dummy token injects perturbation at the token level — the effect is indirect and unpredictable.
Tips
- Start with defaults — strength=1.0, n_periods=2, max_rotation=π/2 is a safe baseline
- Want more diversity? Raise
n_periodsfirst (more frequency bins affected), thenmax_rotation(higher per-bin ceiling) - Compatible with other model patches (ComfyUI-LCS color control, ControlNet, etc.) — operates on a different hook
Tested Models
| Model | Status |
|---|---|
| FLUX2.[Klein] 9B | Tested |
| z-image-turbo | Tested |
Feel free to report results with other distilled models.
License
MIT