SoL-Refiner: A high resolution video refiner
- SoL-Refiner is a model that converts low resolution or low fidelity video output into high resolution, including 4K. Doing so reduces computational cost by performing semantic and motion tasks at low resolution before refining high-resolution specifics.
SoL-Refiner is a one-step latent-video refinement model that converts low-resolution or low-fidelity video outputs into high-resolution video, including 4K outputs, using a single denoising evaluation. It addresses the computational cost of high-resolution video generation by shifting semantic composition and motion formation to a lower-resolution base generator and reserving high-resolution computation for targeted refinement. Its training combines high-resolution continual training, reinforcement-learning post-training, and one-step distillation. At approximately 2K resolution, SoL-Refiner outperforms the evaluated external refiners on VBench and UniPercept averages; at , it improves both metrics over the three-step LTX-2.3 Refiner. With the complete acceleration stack, it achieves an refinement-latency speedup over that baseline (Liu et al., 29 Sep 2026).
1. Computational motivation and design objectives
High-resolution video generation is expensive because the number of spatiotemporal latent tokens increases with frame count and spatial resolution. If denotes the number of frames, and the spatial dimensions, and and the temporal and spatial compression factors, the latent token count is approximately
For a diffusion or flow-based model requiring denoising steps, computational cost scales approximately with the product of the number of denoising evaluations, model cost, and spatiotemporal token count. Direct high-resolution generation therefore incurs both large per-step cost and repeated high-resolution model evaluations.
SoL-Refiner adopts a cascaded formulation:
where a relatively inexpensive base generator produces a low-resolution video and the refiner reconstructs high-resolution detail. The base stage establishes scene identity, composition, motion, and semantic content; the refinement stage targets local texture, edges, facial details, fine geometry, imaging quality, and artifacts introduced by low-resolution generation.
Conventional cascaded refiners can nevertheless require three, eight, or more high-resolution denoising steps. SoL-Refiner treats the refiner itself as a one-step generation problem. Its objective is not restricted to spatial upsampling: the model can also improve artifacts at an existing resolution, although resolution enhancement is a principal application.
The method is initialized from the pretrained LTX-2.3 video model. Its conditioning includes the low-resolution or low-fidelity video latent, text conditioning, and reference-image features. Reference-image tokens guide generation but are excluded from the refinement reconstruction loss.
2. Latent formulation and flow-matching refinement path
SoL-Refiner operates in latent video space. An encoder 0 maps a video 1 to a clean latent:
2
Let 3 denote the low-quality or low-resolution latent and 4 the high-quality target latent. Rather than initiating refinement from independent Gaussian noise, the method constructs a noisy source endpoint that retains information from the low-resolution input:
5
Stage I uses
6
A flow-matching path connects the high-quality target to the noisy low-resolution-conditioned source:
7
The target velocity is
8
The refiner predicts a velocity field 9, where 0 includes the low-resolution video, text, and reference-image conditioning. At the target endpoint 1, the state equals 2; at 3, it equals the noisy source 4. This source-conditioned path distinguishes refinement from unconstrained generation from noise.
The model uses a multiscale residual video architecture with base channel width 64, channel multipliers 5, and two residual blocks per stage. Inference and end-to-end training use an eight-step Euler sampler in the multi-step configuration:
6
3. Three-stage training procedure
Stage I: high-resolution continual training
The first stage teaches a pretrained high-capacity video model to transform a low-fidelity video into a high-fidelity target while preserving motion and semantic content. The primary data source is an internal real-video dataset. The original high-quality video provides 7, while spatial downsampling followed by upsampling produces the aligned degraded input 8.
A smaller synthetic-pair dataset is used for warm-up. A supported base generator creates a low-resolution video, and the LTX-2.3 Refiner produces the corresponding high-quality target. The training objective is the flow-matching loss
9
The noise level is sampled from a shifted logit-normal distribution truncated to 0. This truncation ensures that training states remain connected to the low-resolution input rather than requiring the model to solve an unconstrained generation problem. Stage I yields a multi-step high-resolution refiner and does not yet produce the final one-step model.
Stage II: reinforcement-learning post-training
Flow-matching supervision does not directly optimize perceptual qualities such as human preference, facial detail, or image degradation. Stage II therefore applies Reward Feedback Learning (ReFL) to the multi-step refiner.
The final training configuration uses 1 effective Euler updates. Reward supervision is applied only at later, lower-noise states. A timestep is sampled as
2
Earlier steps are executed without retaining gradients. At the selected state, the model estimates the clean latent using
3
The decoded video is divided into beginning, middle, and end segments, with one frame sampled uniformly from each segment. The reward models are HPSv3++, which measures prompt-conditioned human preference and perceptual quality, and DeQA, which measures degradation-aware visual quality. If 4 and 5 are their scores, the combined reward is
6
The clipping terms limit reward exploitation. The multi-step model reaches a VBench average of 7 and a UniPercept average of 8 after this stage.
Stage III: one-step distillation
Stage III compresses the multi-step Stage II model into a one-step model through few-step DMD distillation followed by one-step DMD-GAN distillation. The intermediate three-step model stabilizes the final one-step optimization.
In the three-step stage, the student generator, frozen real-score teacher, and trainable fake-score model are initialized from the Stage II checkpoint, with the real-score teacher frozen. The student noise levels are
9
The student predicts the high-quality latent 0. The DMD gradient is formed from the normalized difference between fake and real score-model predictions:
1
The student is optimized with the surrogate DMD loss
2
where 3 denotes stop-gradient. The fake-score model is updated from detached student samples, and implicit distribution alignment uses
4
For the final one-step stage,
5
and inference uses the schedule 6. The model directly maps the noisy low-resolution-conditioned latent to the high-quality latent in one generator evaluation.
The final student also receives adversarial supervision from a projected DiT discriminator. The discriminator uses features from blocks 7, 8, and 9 of the frozen real-score DiT, together with noise- and text-conditioned query heads. Approximate R1 regularization is used:
0
This regularization encourages local discriminator smoothness without requiring second-order gradients through the DiT feature extractor.
4. Inference acceleration and implementation stack
SoL-Refiner’s reported latency improvement combines three mechanisms:
- One-step distillation, which reduces high-resolution denoising evaluations.
- Tiny Autoencoder (TAE), which replaces the full video variational autoencoder and reduces encoding and decoding latency.
- Sol-Engine, which applies kernel fusion and sparse Sol-Attn attention to accelerate transformer computation.
For multi-step refinement, Sol-Engine can reuse cached intermediate results across denoising steps.
The principal 2K latency experiment uses one H100 GPU and 241-frame, 24-fps videos at 1:
| Configuration | Latency |
|---|---|
| Three-step LTX-2.3 Refiner | 2 s |
| One-step SoL-Refiner | Baseline divided by 3 |
| One-step plus TAE | Previous latency divided by 4 |
| One-step plus TAE and Sol-Engine | 5 s |
The complete speedup is
6
The adjacent acceleration factors are 7 from one-step distillation, 8 from TAE, and 9 from Sol-Engine.
For longer videos, Sol-Engine reduces multi-step refinement latency from 0 to 1 seconds for 81 frames at 2K and from 2 to 3 seconds for 161 frames. At 4K, the corresponding reductions are from 4 to 5 seconds for 81 frames and from 6 to 7 seconds for 161 frames.
A separate end-to-end example uses distilled four-step MiniMax H3 generation followed by one-step SoL-Refiner. A 8 five-second, 24-fps output takes 9 seconds on one GB200, compared with 0 seconds for published full-resolution 50-step H3 generation, a 1 end-to-end improvement. This result is distinct from the 2 refiner-only comparison.
5. Refiner-Bench and evaluation methodology
Refiner-Bench contains 150 videos: 50 generated by WAN 2.1 1.3B, 50 by SANA-Video 2B, and 50 by LTX-2 Stage 1. The use of multiple base generators tests whether refinement quality depends on a particular upstream model. The benchmark includes 12 content groups, three motion levels, and six camera-motion types.
Its content categories include human daily activity, animals, nature and landscape, sports and dance, cinematic portraits, urban and architecture, vehicles and machines, multi-subject interaction, camera-motion stress, fantasy and science fiction, food and cooking, and product or object close-ups. The benchmark contains 50 low-motion, 64 medium-motion, and 36 high-motion videos. Camera-motion labels include dolly, orbit, pan, handheld, tilt, and static.
A shared-input protocol aligns the upstream outputs:
| Source | Native resolution | Aligned crop | Shared input |
|---|---|---|---|
| WAN 2.1 1.3B | 3 | 4 | 5 |
| SANA-Video 2B | 6 | 7 | 8 |
| LTX-2 Stage 1 | 9 | 0 | 1 |
The evaluated refiners are LingBot Stage-2 Refiner, LTX-2.3 Refiner, LTX-2.0 Refiner, SEEDVR2, one-step SoL-Refiner, and the 23-step SoL-Refiner.
VBench measures subject consistency, background consistency, motion smoothness, dynamic degree, aesthetic quality, and imaging quality. UniPercept measures image aesthetic assessment, image quality assessment, and image structure or temporal assessment. Its scores use eight uniformly sampled frames per video.
At approximately 2K resolution, the main results are:
| Method | Steps | VBench average | UniPercept average |
|---|---|---|---|
| Resized input | — | 2 | 3 |
| LingBot Stage-2 | 8 | 4 | 5 |
| LTX-2.3 | 3 | 6 | 7 |
| LTX-2.0 | 3 | 8 | 9 |
| SEEDVR2 | 1 | 0 | 1 |
| SoL-Refiner, 23-step | 23 | 2 | 3 |
| SoL-Refiner, one-step | 1 | 4 | 5 |
The one-step model obtains the best external-refiner result on both reported averages. Its VBench aesthetic-quality and imaging-quality scores are 6 and 7, respectively. The 23-step model remains stronger, demonstrating a quality–speed trade-off.
At 8, the one-step model improves over the three-step LTX-2.3 Refiner by 9 on VBench average and 00 on UniPercept average.
6. Transfer, ablations, and limitations
Transfer across base generators
SoL-Refiner can be applied without modifying or retraining the base generator. The experiments use WAN-5B, WAN-1.3B, and Cosmos-Nano:
| Base generator | Latency reduction |
|---|---|
| WAN-5B | 01 |
| WAN-1.3B | 02 |
| Cosmos-Nano | 03 |
For WAN-5B, direct high-resolution generation scores 04 on VBench and 05 on UniPercept with 06 seconds of latency; the low-resolution-plus-refiner pipeline scores 07 and 08 with 09 seconds. For WAN-1.3B, scores increase from 10 to 11 on VBench and from 12 to 13 on UniPercept, while latency decreases from 14 to 15 seconds. For Cosmos-Nano, scores increase from 16 to 17 and from 18 to 19, while latency decreases from 20 to 21 seconds.
Training-stage ablations
| Stage | VBench average | UniPercept average |
|---|---|---|
| Stage I: continual multi-step training | 22 | 23 |
| Stage II: RL post-training | 24 | 25 |
| Stage III: one-step DMD-GAN | 26 | 27 |
Stage I establishes the refinement mapping, Stage II provides the largest perceptual improvement, and Stage III sacrifices some quality to obtain one-step inference while remaining superior to the Stage I baseline.
The complete reward recipe reaches a VBench average of 28. Removing the regularization recipe lowers this to 29; using only HPSv3++ lowers it to 30. The two-reward configuration balances prompt-conditioned preference and degradation-aware quality.
Sequential RL followed by DMD is also superior to jointly optimizing RL and DMD: the DMD-R strategy reaches 31, whereas RL followed by DMD reaches 32.
Quality–speed trade-offs
The one-step model is faster but remains below the 23-step teacher in perceptual quality. Its principal advantage is operational: it allows a low-resolution base generator to establish semantic and temporal structure while restricting high-resolution computation to one denoising evaluation.
The method is primarily designed for local visual refinement. Large semantic, geometric, or motion errors in the base video may persist. Stage II evaluates three sampled frames with image-quality and preference rewards rather than a direct long-range temporal-consistency objective. Consequently, temporal consistency is handled through the joint video latent representation and temporally aligned training data, but is not directly optimized by a dedicated temporal reward.
The benchmark consists of AI-generated videos from three source generators and does not establish performance on broad camera-captured degradation, arbitrary video restoration, or general video editing. Source fidelity also constrains the correction process: SoL-Refiner is trained to enhance rather than substantially reinterpret the input video.
SoL-Refiner therefore represents a refinement-aware generative architecture in which the low-resolution generator and high-resolution corrector are separated computationally but coupled through training. Its defining contribution is the combination of a low-resolution-conditioned flow path, reward-optimized multi-step refinement, and one-step DMD-GAN distillation. The result is a high-resolution video refiner that retains the quality advantages of a learned multi-step corrector while substantially reducing the second-stage sampling bottleneck.