The University of Texas at Austin LIVEYouTube

Rectified-CFG++

Geometry-Aware Guidance for Rectified Flow Models

Shreshth Saini

Laboratory for Image and Video Engineering (LIVE), The University of Texas at Austin

Advised by Prof. Alan C. Bovik

NeurIPS 2025 · November 2025

Saini, Gupta, Bovik. Rectified CFG++ for Flow Based Models. NeurIPS 2025 (arXiv:2510.07631)

The question

Why does guidance break on the best image generators?

Flux, SD3 and SD3.5 sample a rectified flow with a deterministic ODE. Classifier-free guidance (CFG) pushes their trajectories off the manifold, and nothing brings them back.

Why we need guidance at all

  • Without guidance, a flow model gives generic images that only loosely match the prompt.
  • Guidance should amplify the prompt: sharper detail, fuller color, better alignment.
  • CFG does that for diffusion models. On flow models it oversaturates.

Flux-dev, same prompt, same seed

No guidance

CFG, ω = 3.5

Ours, λ = 3.5

SD3.5, same prompt, same seed

No guidance

CFG, ω = 3.5

Ours, λ = 3

No guidance is dark and muddy. CFG oversaturates. Ours stays natural and faithful to the prompt.

Rectified-CFG++NeurIPS 2025

What CFG does to flow models

Standard CFG on random prompts. These are typical results.

Flux-dev with CFG, ω = 3 (paper Figure 6)

Plastic, cartoon look

Oversaturated

Flat, clipart look

Blown highlights

Washed out

SD3 with CFG, ω = 3 to 3.5

Oversaturated

Text garbled

Misspelled

Color blow-out

Color distortion

The same four failures every time: oversaturation, a plastic look, garbled text, flat texture.

Rectified-CFG++NeurIPS 2025

Why CFG works on diffusion but breaks on flows

Same extrapolation, different samplers

SDE (DDPM) +σε +σε +σε noise pulls back ODE (Flux) manifold drift no way back
SamplerUpdateAfter CFG pushes the sample off the manifold
SDE (diffusion)dx = [f(x,t) − g(t)2 ∇log pt(x)] dt + g(t) dw̄fresh noise at every step partly pulls it back
ODE (rectified flow)dx = vθ(x, t) dtnothing pulls it back, and the error compounds
Discrete DDPM step: xt−1 = μθ(xt, t) + σt z with z ~ N(0, I). The σt z term is the fresh noise.

So we need guidance built for deterministic ODE samplers.

Rectified-CFG++NeurIPS 2025

What good guidance for a flow model should do

  • D1. Stay on the manifold, so errors cannot pile up.
  • D2. Guide strongly early (layout, color), gently late (text, texture).
  • D3. Come with provable bounds.
  • D4. Drop into any pretrained model, no retraining.
MethodD1D2D3D4
CFGnononoyes
CFG-Zero*nononoyes
APGpartlynonoyes
Rectified-CFG++ (ours)yesyesyesyes
No existing method meets all four. APG is adaptive projected guidance.

We predict a midpoint on the manifold, evaluate guidance there, and add it as a bounded correction.

Rectified-CFG++NeurIPS 2025

Why CFG fails and how we fix it, in one picture

Three-panel geometric diagram

Ideal flow

The sample moves along the manifold.

CFG

Extrapolation steps off it, and the error stays.

Rectified-CFG++

Predict, then correct. The step lands back on the manifold.

Rectified-CFG++NeurIPS 2025

Our predictor-corrector stays on the manifold

Rectified-CFG++ predictor-corrector step on the manifold
Method legend
1. Predict
a half-step with vc alone
x̃mid = xt + (Δt/2) vct
2. Evaluate
guidance at the midpoint
Δvmid = vcmid − vumid
3. Correct
a bounded addition to vc
v̂ = vct + α(t)(vcmid − vumid)

α(t) = λmax(1−t)γ fades to 0 near the data, so fine detail survives.

The trajectory stays in a bounded tubular neighbourhood of Mt.

Rectified-CFG++NeurIPS 2025

Extrapolation versus interpolation

Where each rule puts the guided velocity

CFG extrapolates (ω > 1)

Mt xt vu vc v̂ω past vc off the manifold (ω−1)Δv, the extrapolation 0 (vu) 1 (vc) ω (> 1)

(1−ω)vu + ωvc with ω ∈ [3, 9]. ω = 3 means three times the distance from vu to vc, where the model never trained.

Ours interpolates (0 ≤ α ≤ 1)

Mt xt (Δt/2) vc x̃mid vc (base, on manifold) + α Δvmid x̂t−1 still on the manifold α(t)Δv, small and bounded vc (base) α ∈ [0, 1] vc + Δv never past this point

vc + α(t) Δvmid with α ∈ [0, 1]. The base is always vc; the correction is bounded and decays to 0 near the data.

Rectified-CFG++NeurIPS 2025

The two algorithms side by side

Highlighted lines are what changes

Standard CFG

for t = T down to 1:
  vc = vθ(xt, t, y)
  vu = vθ(xt, t, ∅)
  v̂ = vu + ω(vc − vu)        extrapolation
  xt−1 = xt + Δt v̂

2 network evaluations (NFE) per step

Rectified-CFG++

for t = T down to 1:
  vc = vθ(xt, t, y)
  x̃mid = xt + (Δt/2) vc           predictor
  vcmid = vθ(x̃mid, t − Δt/2, y)
  vumid = vθ(x̃mid, t − Δt/2, ∅)
  v̂ = vc + α(t)(vcmid − vumid)    interpolation
  xt−1 = xt + Δt v̂

3 NFE per step

  1. The base is always vc.
  2. Guidance is evaluated at the predicted midpoint.
  3. α(t) = λ(1−t)γ fades to 0 near the data.
Rectified-CFG++NeurIPS 2025

What we can prove

Four mild assumptions

A1vθ(x,t,y) and vθ(x,t,∅) are L-Lipschitz in xA2Bounded guidance direction, ‖Δvtθ(x)‖ ≤ B
A3α(t) is bounded and integrableA4Bounded conditional velocity, ‖vct‖ ≤ Vmax
Standard regularity conditions for a trained network.

Three bounds

Lemma 1
guidance stability
‖Δvθt−Δt/2 − Δvθt(xt)‖ ≤ L Vmax ΔtThe midpoint moves the guidance direction by only O(Δt)
Proposition 1
single-step perturbation
‖x̂t−1 − x̃t−1‖ ≤ α(t) B ΔtDeviation from the conditional path vanishes near the data
Proposition 2
distributional deviation
DKL(p̂0 ‖ p0) ≤ Cs (B + L Vmax Δt) ∫α(τ) dτTotal KL is set by the integrated guidance

No earlier guidance method for flow models has bounds like these.

Rectified-CFG++NeurIPS 2025

CFG versus ours, property by property

Lemma 2, manifold-faithful corrector

dist(x̂, Mt−Δt) ≤ C ε Δt

Only the training error ε moves us off the manifold.

Integrated guidance

CFG: ∫(ω−1)dt = ω−1 ≈ 2 to 8 Ours: ∫α(τ)dτ = λ/(γ+1) ≈ 0.5 0.5

4 to 16× lower KL divergence

PropertyCFGOurs
Per-step perturbation(ω−1) B Δt
constant, never vanishes
α(t) B Δt
goes to 0 near the data
Manifold distanceUnbounded
errors accumulate
O(ε Δt)
training error only
Integrated guidanceω−1 ≈ 2 to 8λ/(γ+1) ≈ 0.5
KL(p̂0 ‖ p0)O(ω−1)O(0.5)
Late-step behaviorSame ω everywhere,
destroys fine detail
α(t) goes to 0,
preserves fine detail
Guidance evaluationAt the current point
(possibly off the manifold)
At the predicted midpoint
(on the manifold)
Rectified-CFG++NeurIPS 2025

Watching the artifacts grow, or not

Standard CFG, ω = 2.5. Each strip shows 7 denoising steps, noise on the left, final image on the right.

CFG intermediates, prompt 27

Rectified-CFG++, λ = 0.5, same prompt

Rectified-CFG++ intermediates, prompt 27

CFG, another prompt. Drift starts at step 2 or 3.

CFG intermediates, prompt 21

Ours, same prompt. Every intermediate looks natural.

Rectified-CFG++ intermediates, prompt 21
Rectified-CFG++NeurIPS 2025

A 2D toy problem shows the same drift

CFG overshoots and needs sharp late corrections. Ours converges smoothly.

t = 1.0t = 0.8t = 0.6t = 0.4t = 0.2t = 0.0
CFG Rectified-CFG++
200 trajectories on a 2D mixed Gaussian

200 trajectories on a 2D mixed Gaussian. Top: CFG (blue) drifts off the support. Bottom: ours (green) stays on the manifold.

Rectified-CFG++NeurIPS 2025

Baseline, CFG and Rectified-CFG++ on three Flux prompts

3 by 3 comparison on Flux

Rows from the top: Flux with no guidance, CFG, Rectified-CFG++.

  • Cactus. CFG draws a cartoon with an emoji face.
  • Dog and moon gate. CFG loses the stone arch.
  • Parrot. CFG turns the bird psychedelic.

Ours keeps the realism and the structure in all three.

Rectified-CFG++NeurIPS 2025

The same on SD3 and Lumina-Next

SD3, CFG

SD3, ours

SD3, CFG

SD3, ours

Lumina, CFG

Lumina, ours

Lumina, CFG

Lumina, ours

A drop-in replacement across rectified flow models. No retraining, no architecture changes.

Rectified-CFG++NeurIPS 2025

Text survives because guidance fades at the end

CFG

CFG stop sign

Rectified-CFG++

Rectified-CFG++ stop sign

Prompt: "A stop sign with 'ALL WAY' written below it."

α(t) is 0 in the last steps, so the conditional flow alone draws the strokes.

Rectified-CFG++NeurIPS 2025

Results on MS-COCO 10K

ModelMethodFID ↓CLIP ↑ImgRwd ↑Pick ↑HPS ↑
LuminaCFG26.930.2610.54721.030.253
LuminaOurs22.490.2680.62121.190.259
SD3CFG26.330.2730.68421.480.264
SD3Ours24.680.2780.75321.620.269
FluxCFG37.860.2850.89222.040.279
FluxOurs32.230.2890.96122.180.283
CFG against Rectified-CFG++ on three flow models. Bold is best.
Guidance (SD3.5)FID ↓ImgRwd ↑CLIP ↑HPSv2 ↑
CFG67.711.0530.3520.294
CFG-Zero*68.390.9950.3460.288
APG67.231.0750.3510.294
Rectified-CFG++67.151.0850.3510.296
Against the other guidance methods on SD3.5.

Best or tied on every metric, for every model.

−5.6FID on Flux
−4.4FID on Lumina
4models tested
0extra training
Rectified-CFG++NeurIPS 2025

Experts prefer Rectified-CFG++ on every model

30 experts × 32 prompts × 4 models = 15,360 responses

SD3SD3.5Lumina-NextFlux
User study preference per model and criterion
OverallDetailColorTextOverallDetailColorTextOverallDetailColorTextOverallDetailColorText

CFGAPGCFG-Zero*Rectified-CFG++

Four-way forced choice on four criteria. CFG dark blue, APG medium blue, CFG-Zero* light blue, Rectified-CFG++ green.

34 to 36%overall share, against 25% chance (p < 0.001 against APG)
28 to 32%text legibility, against 22 to 25% for the others
κ = 0.61Fleiss' kappa, substantial agreement
Rectified-CFG++NeurIPS 2025

CFG collapses at high guidance, ours stays flat

Metrics against guidance scale

x-axis: guidance scale (λ for ours, ω for CFG). y-axis: FID ↓, CLIP ↑, ImageReward ↑, Aesthetic ↑. Blue is CFG, green is ours.

CFG

FID goes from 77 to 148 at ω = 10. ImageReward falls from 1.0 to −1.5, CLIP by 30%, Aesthetic by 35%.

Rectified-CFG++

FID stays at 75. ImageReward goes from 1.0 to 0.95.

One setting holds across a 50× range of λ.

Rectified-CFG++NeurIPS 2025

The same quality in about half the steps

Metrics against number of sampling steps

x-axis: sampling steps. Blue is CFG, green is ours. Ours converges faster on every metric.

  • At 5 steps, FID 71 against 178 (2.5× better), ImageReward 1.0 against −1.5.
  • 15 of our steps match 28 of CFG's, about 2× faster.
ConfigurationFID ↓CLIP ↑HPSv2 ↑
Unconditional only91.120.1440.187
Without predictor73.700.3410.297
Without corrector74.650.3410.298
Full72.970.3450.300
Component ablation on SD3.5. Both the predictor and the corrector are needed.
Rectified-CFG++NeurIPS 2025

Takeaways

  1. On-manifold guidance for rectified flows, a predictor-corrector with decaying strength.
  2. Three propositions, and bounds 4 to 16× tighter than CFG (λ/(γ+1) against ω−1).
  3. Drop-in for Flux, SD3, SD3.5 and Lumina. 15 of our steps match 28 of CFG's.
  4. 43.5% preference across 15K expert responses, p < 0.001.

Related work of mine

This talk: Saini, Gupta, Bovik. Rectified CFG++ for Flow Based Models. NeurIPS 2025. arXiv:2510.07631

Saini, Liao, Ye, Bovik. LGDM: Latent Guidance in Diffusion Models for Perceptual Evaluations. ICML 2025. arXiv:2506.00327

Saini, Gedik, Birkbeck, Wang, Adsumilli, Bovik. LumaFlux: Structure-Aware SDR-to-HDR Conversion with Monotone Luminance Flows. Preprint, 2026. arXiv:2604.02787

Chen, Saini, Adsumilli, Bovik. LumaGuide: Distribution Shaping for Training-Free HDR Generation in Diffusion Models. Preprint, 2026. arXiv:2607.26237

Saini, Birkbeck, Wang, Adsumilli, Bovik. CachedSearch: Training-Free Cached Exploration for Test-Time Search in Video Diffusion. Preprint, 2026. arXiv:2607.23159

Thank you.

The University of Texas at Austin LIVE YouTube

Backup

Backup slides

The ideal flow on the manifold

Conditional, CFG and Rectified-CFG++ paths on the manifold
Method legend Flow legend
  • Three manifolds: Mt now (the blue and gold band), Mt−1 at the next step, and M0, the data (the orange curve). Red dots are the ideal positions.
  • The ideal trajectory moves along the surface. Its velocity stays in the tangent space.
  • Blue is the Flux conditional velocity vc, red is CFG, green is Rectified-CFG++. Green and blue stay near the manifold, red diverges.
Rectified-CFG++NeurIPS 2025

CFG extrapolates off the manifold

CFG steps off the manifold
Method legend
  • From the black dot on Mt, CFG computes v̂ω at the current point. With ω > 1 it extrapolates past vc.
v̂ω = vu + ω(vc − vu), ω > 1 overshoots
  • v̂ω is not in the tangent space TxtMt. The Euler step lands off Mt−1, and a deterministic ODE has no way back.
  • The next step evaluates guidance at that off-manifold point. The error compounds into oversaturation, color blow-out and distorted structure.
Rectified-CFG++NeurIPS 2025

Four guidance rules for the same model

Only ours avoids extrapolation altogether

Standard CFG (Ho and Salimans, 2022)

v̂ = (1 − ω) vu(xt) + ω vc(xt)

ω > 1 extrapolates past vc. Built for SDEs with renoising. Same strength at every timestep.

CFG-Zero* (Fan et al., 2025)

v̂ = (1 − ω) st* vu(xt) + ω vc(xt)

The rescaling factor st* aligns the two norms. Less initial drift, but it still extrapolates (ω > 1).

APG (Sadat et al., 2024)

v̂ = vc(xt) + Δvt(η, r, β)

Projects the guidance direction to shrink its parallel component. Still evaluated at the current point, with no decay schedule.

Rectified-CFG++ (ours)

v̂ = vc(xt) + α(t) (vcmid − vumid)

The base vc and the midpoint x̃mid are both on the manifold. α(t) = λ(1−t)γ goes to 0 near the data.

PropertyCFGCFG-Zero*APGOurs
Guidance typeextrapolationextrapolationprojectioninterpolation
Evaluated atcurrent xtcurrent xtcurrent xtpredicted x̃mid
Temporal schedulenone (constant)nonenoneα(t) decays to 0
Formal guaranteesnononoyes (3 propositions)
NFE per step2223
Rectified-CFG++NeurIPS 2025

Same prompt, same seed, three guidance strategies on Flux

Baseline, CFG and Rectified-CFG++ on Flux-dev

Flux-dev, same prompt and seed. Left: no guidance. Center: CFG, ω = 3.5, with rainbow artifacts. Right: Rectified-CFG++, λ = 0.5.

Rectified-CFG++NeurIPS 2025

Rectified-CFG++ on random prompts

Flux-dev, random prompts from MS-COCO and Pick-a-Pic, no cherry-picking

Gallery of Rectified-CFG++ samples
  • Photoreal scenes. Natural lighting, correct shadows, no oversaturation.
  • Artistic and stylized. Strong early guidance sets the global style, and fine brush strokes survive the late steps.
  • Text and composition. α(t) going to 0 keeps signage, titles and multi-object layouts intact.
Rectified-CFG++NeurIPS 2025

Four guidance methods on SD3.5, same prompt and seed

CFG, ω = 3.5

CFG-Zero*

APG

Ours, λ = 6

CFG

Ours

CFG

Ours

Rectified-CFG++NeurIPS 2025

CFG versus ours on Flux-dev, eight pairs

CFG

Ours

CFG

Ours

CFG

Ours

CFG

Ours

Rectified-CFG++NeurIPS 2025

Text rendering, a neon street sign

CFG

CFG neon sign

Rectified-CFG++

Rectified-CFG++ neon sign

Prompt: "A neon street sign that says 'CyberCore Cafe', glowing in magenta and blue."

CFG garbles the letters into "Cyberre Cidie Cafe". Ours keeps every letter and glow separate.

Rectified-CFG++NeurIPS 2025

Text rendering, headlines and inscriptions

CFG

Ours

CFG

Ours

Prompt: "A crow detective reading a paper titled 'Feathered Conspiracies', headline in bold noir font."

Prompt: "A magical sword embedded in stone, with the name 'SOLARFANG' etched along its blade in glowing runes."

CFG writes "Feathhrad Conspiracies" and turns SOLARFANG into garbled runes. Ours gets both right, on aged paper and on glowing stone.

Rectified-CFG++NeurIPS 2025

Text rendering, a slide and a scroll

CFG

Ours

CFG

Ours

Prompt: "A fox giving a TED talk, slide behind reads 'Entropy and the Soul'."

Prompt: "A burning scroll with the title 'Soulbound by Nightfall' in ornate calligraphy."

CFG writes "Entφy amd the Soul" and leaves the scroll title barely readable. Ours gives clean titles on both the screen and the scroll.

Rectified-CFG++NeurIPS 2025

Text rendering, carved and printed

CFG

Ours

CFG

Ours

Prompt: "A desert gate carved in sandstone reading 'CITY OF WHISPERS'."

Prompt: "A Japanese fish stamp print with 'WAVES OF JUSTICE' in red ink."

Neon, carved stone, ink stamps, calligraphy, digital displays: the text stays legible because α(t) goes to 0 while fine detail is resolved.

Rectified-CFG++NeurIPS 2025

Text legibility on SD3.5, the full gallery

Each pair: CFG on the left, Rectified-CFG++ on the right

CFG

Ours

CFG

Ours

CFG

Ours

CFG

Ours

CFG

Ours

CFG

Ours

CFG

Ours

CFG

Ours

Legible text on every prompt. The adaptive schedule α(t) is what does it.

Rectified-CFG++NeurIPS 2025