SoL-Engine · Research Blog
High-Resolution Video Refinement with Few-Step Diffusion
The H3 base model drafts a video at low resolution, and the official H3-Regenerate-2K API regenerates it at 2K, but that model is not open. Open-H3-Regenerate-2K is an open replacement: a three-step H3 diffusion pass that redraws the draft’s detail at full resolution.
Sol-EngineVideo Inference
Abstract
Open-H3-Regenerate-2K keeps the H3 transformer and video VAE frozen and adds one refiner LoRA, trained like Stage I of SoL-Refiner on real high-resolution clips paired with degraded copies of themselves. At inference the draft is encoded once; H3-Latent-Upscaler, a community 3D-convolutional network, enlarges the latent straight onto the target grid, with no decode and re-encode; the enlarged latent is re-noised to σ = 0.909 and denoised in three Euler steps (0.909 → 0.857 → 0.667 → 0) with classifier-free guidance off, so each step is one transformer call at full resolution. For these few steps the refiner is merged with LightX2V’s FL2VA Turbo and the LMS Sharp adapter into a single LoRA.
The official API only accepts unmodified H3 768P output and returns 2K. Open-H3-Regenerate-2K takes a draft at any low resolution, such as 384P, 512P, 576P or 768P, made by H3 or by another video model, and regenerates it natively at 768P, 1080P, 2K (2560 × 1440) or 4K (3840 × 2160), for clips of 5 to 15 seconds in one pass. With SoL-Engine (Sol-Attn sparse attention, MXFP8 linear layers and CUDA graphs) and the fast H3 VAE decoder of StreamRender-H3, latent upscaling, the three steps and VAE decoding for a 5-second 2K clip take 22.45 s on one B200 instead of 87.26 s.
On RefinerBench-150, with both systems at 2560 × 1440, ours scores higher than the official API on all three UniPercept axes and on VBench aesthetic quality, imaging quality and subject consistency. In a blinded user study, ours was preferred in 87% of 200 judgments.
Why an Open Regenerate Model?
H3 makes a video in two steps. The base model settles content and motion at low resolution, where sampling is cheap. H3-Regenerate-2K then redraws the draft at 2K, so faces, fabric, fur and foliage hold up at full size.
Only the first step is open. H3-Regenerate-2K is a hosted API without released weights, so people building on the open H3 base model cannot run it on their own GPUs, adapt it, or go beyond what the service offers.
Open-H3-Regenerate-2K fills that gap with an open model for the same step: it starts from the H3 draft, keeps its content and motion, and regenerates the detail at full resolution.




From a Low-Resolution Draft to 2K
Open-H3-Regenerate-2K wraps a frozen H3 transformer and video VAE. It takes the draft and its prompt and returns the same scene at the target resolution.
- Encode. The H3 VAE encodes the draft once.
- Enlarge. H3-Latent-Upscaler maps the latent onto the target grid.
- Regenerate. The enlarged latent is partly re-noised and denoised in three steps.
- Decode. The H3 VAE decoder turns the result back into frames; the fast path uses the 24-layer fast H3 VAE decoder from StreamRender-H3.
Upscaling alone adds no detail. It carries layout, color and motion onto the larger grid; the three diffusion steps draw the surfaces at that size.
2.1Visual Results
The input on the left; the official H3-Regenerate-2K output and ours, both 2560 × 1440, in the middle, with ours always below; on the right, the region under the magnifier in each output. The first group starts from drafts made by the H3 base model, the second from drafts made by other video models (LTX-2 and SANA-Video), prepared at 1024 × 576.
H3 Input
Hover over a video to move the magnifier
Other-Model Input
Hover over a video to move the magnifier
2.2From Any Resolution to 4K
Open-H3-Regenerate-2K is not tied to one input or output size. It takes a draft at any low resolution, such as 384P, 512P or 576P, and regenerates it natively at 768P, 1080P, 2K or 4K: the latent upscaler enlarges the draft latent to the target grid, and the three diffusion steps run at that size. Below, a dance draft resized to 4K and the same draft regenerated natively at 4K.
Hover over a video to move the magnifier
2.35 to 15 Seconds
Clips from 5 to 15 seconds are refined in a single pass. This 15-second clip, 360 frames, is regenerated at 2K in one run.
Hover over a video to move the magnifier
Key Technologies
3.1Training
We train the refiner following Stage I of SoL-Refiner, its high-resolution continual training, with the same recipe and data but with H3 in place of SoL-Refiner’s original base model. Each training pair is built from a real, high-quality clip at 1280 × 704 or 2560 × 1440. The target is the clip itself; the condition is a degraded copy, shrunk by 8× to 64× with an antialiased bilinear filter and resized back. Both are encoded by the H3 VAE.
The H3 transformer stays frozen and only a refiner LoRA is trained, with a flow-matching velocity loss on noise levels from σ = 0.91 down, the same level the three-step inference starts from (Section 3.3). SoL-Refiner’s later stages, RL post-training with ReFL and DMD distillation to a one-step refiner, are not used yet. For inference the refiner LoRA is merged with a community distillation LoRA.
3.2Latent Upscaling
The refiner works on the target grid, so the draft first has to reach the output size. Done in pixels, that is a round trip: decode the low-resolution latent with the VAE, resize every frame, and encode the frames again at 2K or 4K. At these sizes the VAE is slow. H3-Latent-Upscaler, a community model, skips the round trip by enlarging the latent directly with a learned 3D convolutional network. We use it as released. At 2K it maps a 36 × 64 latent grid to 90 × 160, 2.5× per side, leaving the 24 channels and 37 latent frames unchanged; for 4K it is applied twice, each time within its trained range.
3.3Three-Step Regeneration
For speed we build on another community work: LightX2V’s FL2VA Turbo, a LoRA that lets H3 generate in four steps instead of dozens. We merge it as released, together with the community LMS Sharp adapter by Alissonerdx. A four-step run starts from pure noise, but our refiner LoRA (rank 384, 5,000-step EMA) is trained to start from the upscaled draft partly re-noised at σ₀ = 0.909. The draft already fixes layout and motion, so three steps are enough: 0.909 → 0.857 → 0.667 → 0. Classifier-free guidance is off, so each step is a single transformer call at the target resolution.
3.4Inference Optimization
Starting from a 1344 × 768 H3 latent, latent upscaling to 2560 × 1440, the three refine steps and VAE decoding take 87.26 s without and 22.45 s with SoL-Engine and the fast H3 VAE decoder of StreamRender-H3 on a single B200, 3.9× faster.
Sparse attention saves the most: Sol-Attn removes 39.99 s. The fast H3 VAE decoder, the released 36-layer decoder pruned to 24 layers and distilled from it, cuts decoding from 15.0 s to 1.6 s. Lossless kernels, MXFP8 linear layers with a fused SwiGLU quantizer and per-step CUDA graphs account for the rest.
Warm inference on one GPU, median of three runs. Generating the 768P draft, model loading, text encoding and MP4 writing are not included. The DiT modules are measured one at a time and scaled to the measured total; the VAE row is measured directly.
Experiments
We compare against the official H3-Regenerate-2K API with two automatic metrics on RefinerBench-150 and with a user study. RefinerBench-150, from the SoL-Refiner benchmark, has 150 videos covering everyday scenes, sports and dance, animals, vehicles, landscapes, portraits and fast camera motion. Both systems receive the same input and output 2560 × 1440.
4.1VBench
VBench rates six dimensions on a 0–1 scale. Subject and background consistency check that identity and scene stay stable over time, motion smoothness and dynamic degree describe the motion, and aesthetic and imaging quality judge how each frame looks. Following SoL-Refiner, the average is the unweighted mean of the six, not the leaderboard total.
Ours is higher on aesthetic quality, imaging quality and subject consistency. The official API is slightly ahead on motion smoothness and dynamic degree, and background consistency is level.
4.2UniPercept
UniPercept scores perceived quality from 0 to 100 along three axes: aesthetics (IAA), technical image quality (IQA), and structure and texture (ISTA). The average is the mean of the three.
Ours is higher on all three axes, by about eight points on aesthetics and image quality and three points on structure and texture.
4.3User Study
The blinded pairwise study uses 50 cases: 25 RefinerBench-150 videos and 25 prompts shared online, mostly on X. Each trial shows the input as a reference and the two 2560 × 1440 results as A and B in random order. Participants pick the one that looks better, weighing clarity, detail, naturalness, artifacts and temporal stability. Each session covers ten cases; 20 completed sessions give 200 judgments.
SoL-Engine · Research Blog · IC Version
Consistent High-Resolution Video Refinement via In-Context Learning
Open-H3-Regenerate-2K redraws a re-noised draft in three steps; it adds detail but can drift on faces, mouths and lettering. The IC version keeps the draft in the transformer’s context instead: the draft stays clean, the high-resolution video is generated from noise beside it, and every step attends to it. Faces, lip sync, text and layout follow the draft while the detail is drawn at 2K or 4K.
Sol-EngineVideo Inference
Training
H3 ships a reference-to-video transformer, Ref2VA, that reads reference videos as extra tokens in the same sequence as the video it generates. We keep it frozen and train a LoRA on it. In each training pair the low-resolution copy of a clip is the reference and the clip itself is the target: the reference latent stays clean, the target latent is noised along the flow-matching path, and the loss is taken on the target tokens only.
Each training pair starts from a real clip at 768P up to 4K. Resizing it down gives the draft-like low-resolution copy, and some copies are blurred further so the model also sees drafts softer than a clean resize. Both are encoded with the H3 VAE at their own resolutions. In the sequence the copy’s latent sits in front of the target, and the target latent is mixed with Gaussian noise at a level σ drawn over the whole path, from pure noise (σ = 1) down to 0, so the LoRA learns to generate the high-resolution clip from noise next to the draft instead of polishing a re-noised copy of it. HyperFlow and LMS Sharp are not used in training.
Inference
At inference the 768P H3 draft takes the place of the low-resolution copy. It is encoded once at its own resolution and kept in front of the target as clean reference rows, and its frames and the original H3 prompt go to the Qwen3-VL context. The 8-step HyperFlow LoRA and LMS Sharp at strength 1.0 are merged into the transformer together with the IC LoRA; with CFG 1 every step is one transformer call. The samplers differ only in where the 2K or 4K latent starts.
- From noise (
hf8). The target starts as pure noise and runs all eight HyperFlow steps, σ 1 → 0. - From the latent upscaler (
up3,up4). The latent upscaler, a 3D convolution network, enlarges the draft latent straight to the 2K or 4K latent grid without decoding it. That latent is mixed with noise to σ 0.835 (up3) or 0.923 (up4) and finished with the last three or four steps of the same HyperFlow grid. The upscaled latent sets layout and colour, the remaining steps draw the detail, and the draft stays in context at every step.up3is the default: three transformer calls per clip.
The 2K result is decoded with the fast H3 VAE and 4K with the H3 VAE; the draft’s audio track is copied unchanged.
hf8 starts the 2K latent from pure noise, up3 and up4 from the draft enlarged by the latent upscaler. The crops show the same window from the 768P draft and the 2K output.up3 is the default.Against Open-H3-Regenerate-2K
Drafts made by the H3 base model at 1344 × 768, with the previous open model and the IC version at 2560 × 1440, playing in sync; the row underneath magnifies the same window from each. The previous model draws more texture but moves eyes, wrinkles and mouth shapes away from the draft; the IC version keeps them. The magnifier follows the pointer.
Hover over a video to move the magnifier
Against H3-Regenerate-2K
The same drafts through the official H3-Regenerate-2K API and the IC version, both at 2560 × 1440. On the car park, the official output turns the sign into unreadable marks while the IC version keeps “PRIVATE CAR PARK”; on the kingfisher and the glass mushrooms, the IC version draws crisper water spray and moss.
Hover over a video to move the magnifier
More 4K Results
768P drafts regenerated natively at 4096 × 2304.
Hover over a video to move the magnifier
From 384P to 768P
The IC version also starts from smaller drafts. Each video below was generated by H3 at 672 × 384 and regenerated at 1344 × 768 in eight HyperFlow steps.
Drag across the video to compare
Credits
Acknowledgements
Authors
Sensen Gao*Haozhe Liu*Yitong LiHaopeng LiTian YeJia-Wang BianSong HanEnze Xie
* Equal contribution
- HyperFlow8-step distillation LoRA used by Open-H3-Regenerate-2K-IC.
- SoL-H3Speed-of-light H3 inference.
- SoL-EngineInference engine and the acceleration in Section 3.4.
- Sol-AttnSparse attention in SoL-Engine, the largest single speedup in Section 3.4.
- StreamRender-H3Fast H3 VAE decoder (its non-causal variant) for VAE decoding.
- MiniMax H3Base model and VAE; the official H3-Regenerate-2K API is the baseline.
- H3-Latent-UpscalerLatent upscaler.
- SoL-RefinerRefinement recipe and the VBench protocol.
- LightX2VFL2VA Turbo adapter.
- LMS SharpSharpening LoRA by Alissonerdx, merged into our LoRA.
- VBenchVideo quality metrics, with UniPercept.
How to cite
Citations
If Open-H3-Regenerate-2K helps your work, please cite the SoL-Refiner paper together with this blog.
@misc{liu2026solrefiner,
title = {{SoL-Refiner}: Speed-of-Light One-Step Refinement for High-Resolution Video},
author = {Haozhe Liu and Tian Ye and Shuchen Xue and Sensen Gao and
Yitong Li and Junsong Chen and Haopeng Li and Jincheng Yu and
Duomin Wang and Ruihua Zhang and Lei Zhu and Jia-Wang Bian and
Song Han and Enze Xie},
year = {2026},
eprint = {2609.37969},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2609.37969}
}
@misc{gao2026openh3regenerate2k,
title = {Open-H3-Regenerate-2K: High-Resolution Video Refinement with Few-Step Diffusion},
author = {Sensen Gao and Haozhe Liu and Yitong Li and Haopeng Li and Tian Ye and Jia-Wang Bian and Song Han and Enze Xie},
year = {2026},
howpublished = {\url{https://nvlabs.github.io/Sana/Open-H3-Regenerate-2K/}},
note = {Project page}
}