Practitioners building continuous-latent text diffusion models have long observed an uncomfortable asymmetry: the same few-step deterministic generation techniques that produce excellent images in one to four network evaluations produce incoherent, token-flipping garbage when applied to text. The standard explanations invoke training budget, model scale, or scheduling choices. This paper argues those explanations are wrong, and proves it. The failure is geometric and irreducible within the deterministic-continuous class. Full paper: arXiv:2606.30705.
The Core Problem
Image decoders map continuous latents to pixel values through smooth functions. Text decoders, by contrast, terminate in an argmax over a vocabulary embedding matrix, partitioning the latent space into sharp Voronoi-like cells. A deterministic transport map, however accurate, delivers its output only to within some residual blur. For images, that blur is absorbed gracefully because the decoder is smooth. For text, the same blur, if it lands near a decision boundary, flips a token. The question is not whether the transport is accurate on average, but whether it can guarantee placement within the correct argmax cell for every token position simultaneously.
The paper formalises this intuition through Theorem 3, which states that the token flip rate at the posterior-mean terminal step scales as the latent mass within an O(s(t)) tube around decision boundaries, where s(t) is the transport residual magnitude. Critically, this rate is set by decoder sharpness, not by how well the flow was trained. You cannot train your way out of it while remaining in the deterministic-continuous class.
Key Contributions
- Non-commitment theorem (Theorem 3): The token flip probability is asymptotically proportional to the probability that the clean latent lies within O(s(t)) of a decision boundary. Oracle roll-in experiments on the ELF model confirm this, with 12-41% terminal flip rates compounding to roughly 90% error, and all step counts collapsing onto a single s(t) curve.
- DABI diagnostic: Decoder Amplification of Boundary-aligned Inputs measures the ratio of cross-entropy increase from a boundary-aligned perturbation versus a norm-matched isotropic one. Four published continuous-text decoders show DABI between 500 and above 10^5. Four published image VAEs, tested across five NFE settings, show DABI between 0.85 and 1.94. There is no overlap between the two distributions across orders of magnitude.
- CCI diagnostic: Categorical Commitment Index identifies whether a generator commits to discrete tokens before the final readout. Autoregressive and masked diffusion models have high DABI (sharper readouts than ELF) yet succeed precisely because their generators are discontinuous by construction, committing categorically at each step.
- Sharp transport laws (Theorems 5-7): In an idealised separated-mode regime, the deterministic stiffness required to separate M modes grows as the square root of log M once the latent dimension exceeds C log M, and as M^(1/n) in fixed dimension. A depth-B hierarchy reduces the per-step peak by a factor of the square root of B. These are matched upper and lower bounds, constituting a genuine accuracy-depth-stiffness tradeoff.
- Escape mechanisms identified: Stochastic re-injection (SDE sampling) and categorical commitment are the two exits from the bound. On the same ELF-B teacher model, ODE sampling at K=4 gives perplexity 294 versus SDE perplexity 50. Removing categorical commitment from an autoregressive decoder causes a fivefold perplexity jump.
Methodology and Empirical Validation
The theoretical framework distinguishes two regimes: an overlapping regime where text autoencoder latents from different tokens genuinely intermingle, and an idealised separated regime used to derive clean scaling laws. Theorem 3 operates in the overlapping regime under smooth density assumptions and exact decoding, with Proposition 16 extending coverage to learned denoisers whose terminal active-normal residual is stable and O(s).
The empirical work is careful about what it is and is not claiming. DABI is measured using a single margin-normal probe applied identically to all codecs, with 95% bootstrap confidence intervals. The image-text gap is not a marginal statistical finding; boundary-aligned perturbations flip 47-77% of tokens in text decoders while isotropic perturbations of identical norm flip essentially zero. The authors also handle potential counterexamples explicitly.
FMLM, a one-step deterministic flow model for text, appears at first glance to refute the entire argument. The paper addresses this directly: FMLM interpolates toward one-hot simplex targets and applies argmax decoding, which is categorical commitment by definition (CCI=1). The authors measure DABI on the official FMLM checkpoint and find the same sharp-readout signature as autoregressive models. FMLM succeeds because one-hot targets provide large margins, not because it achieves smooth continuous transport. Similarly, FastDiSS's training-time perturbation regularisers and Loopholing's deterministic feature pathway are correctly classified as orthogonal to or consistent with the taxonomy rather than counterexamples.
The survey of 2025-2026 continuous-latent text generators finds that the deterministic-ODE, smooth-continuous, few-NFE cell of the taxonomy is empty across all surveyed literature. Every few-step system uses either SDE sampling or categorical commitment. This is a strong empirical regularity, and the theory explains why it should hold.
Limitations and Open Questions
The formal results assume smooth densities and exact clean decoding. Learned generators only approximate posterior-mean transport, and the bounds apply strictly to deterministic samplers. The authors note that label noise, finite-perimeter irregularities, and codimension-2 boundary effects change constants but not the linear-in-s scaling, which gives the first-order law reasonable robustness to these idealisations.
A more substantive open problem is a Wasserstein theory for overlapping latents. The current sharp transport laws (Theorems 5-7) are proved in the separated regime, and the coarea identity of Theorem 17 connects them to the overlapping tube, but a full within-core W_p theory remains open. This matters practically because real text autoencoder latents are not well-separated.
The paper is also explicit that its results constrain the deterministic-continuous class and say nothing about the quality ceiling achievable by SDE or categorical generators. The finding is a lower bound on what deterministic transport cannot do, not an upper bound on what alternatives can achieve. Whether SDE sampling or categorical commitment introduces its own quality costs at very low NFE is a separate question left open.
For practitioners, the implications are fairly direct. Continuous-latent text diffusion models requiring fewer than roughly 16 steps need either stochastic sampling or a categorical commitment mechanism. Training longer or scaling up will not resolve the geometric constraint. The DABI and CCI diagnostics are cheap to compute on published checkpoints and provide a principled basis for classifying new systems before committing to a generation strategy.