/ Paper arXiv 2611.14033
Curator's notes
APEX removes the fake-score networkfake-score networkA separate network that predicts scores for generated samples to guide few-step generators. for few-step streaming video generation by shifting conditions within the generator.
APEX makes fast video generation cheaper by removing an extra network and using a clever trick inside the model.APEX removes the fake-score networkfake-score networkA separate network that predicts scores for generated samples to guide few-step generators. for few-step streaming video generation by shifting conditions within the generator.APEX applies condition shiftingcondition shiftingAltering the input condition to estimate the velocity field of the generated distribution using the same model. within the generator to replace DMD's fake-score networkfake-score networkA separate network that predicts scores for generated samples to guide few-step generators., achieving better instruction compliance and aesthetics with lower GPU memory and iteration time.Curator's notesWritten by AI from the paperEvery number here appears in the abstractReport a mistake
What's new
It replaces the separate fake network with condition shiftingcondition shiftingAltering the input condition to estimate the velocity field of the generated distribution using the same model., reducing memory and compute while keeping other regularizations. It also adds a one-step spatial Refiner.
Key result
APEX improves instruction fulfillment (3.856 vs 3.831) and visual aesthetics (0.641898 vs 0.632606) over S2-Avatar, with 30% net human preference for instruction following.
Why it matters
Builders can generate high‑quality streaming video faster and with less GPU memory.
Key numbers
From the paperCopied from the abstract, not written by AI.
Why this ranked
MeasuredPublic numbers, not AIHow ranking works
24 / 100#26 of 198 papers that day, from GitHub, Hugging Face, Hacker News and Semantic Scholar, looked up the morning it was ranked.
Builder evidence0 / 40
- No public code found
Community attention18.2 / 30
- ▲ 66 upvotes on Hugging Face
- 0 points on Hacker News
Author track record5 / 15
- From Tsinghua University (a known lab)
What the paper shows0.4 / 15
- Less like recent papers than 7% of that day's candidates
Abstract
From the paperThe authors' own words.
Few-step samplingfew-step samplingGenerating video using only a small number of diffusion steps to reduce computation. reduces the cost of streaming video generation. Distribution Matching Distillation (DMD) supervises few-step generators with a separate fake-score network, which adds training memory and computation and must track the evolving generator. We extend APEX to streaming video by using condition shifting to estimate the velocity field of the generated distribution within the generator. The original text condition selects generation, while a shifted condition selects fake estimation using the same parameters. This removes the independent fake network while retaining S2-Avatar's frozen reference teacher, Self-Replay Forcing, and perceptual regularization. We evaluate APEX on 180-second videos using three backbone evaluations per block and a one-step spatial Refiner, with a 70-case human study and a 320-case StreamAV-Bench comparison. Compared with S2-Avatar, APEX achieves higher video instruction fulfillment on the Progressive track (3.856 versus 3.831) and higher visual aesthetics across both tracks (0.641898 versus 0.632606). Human raters favor APEX most strongly on instruction following, with +30.0% net preference, followed by consistency and video quality at +4.3% each. Motion quality favors S2-Avatar, audiovisual synchronization is predominantly tied, and usability is unchanged. Separate measurements under matched training settings show 4.4% lower peak allocated GPU memory per GPU and 9.2% shorter mean iteration time than DMD.
From the full paper
From the paperCopied from the paper itself by our code.
Reading the paper on arXiv…
What you could build
Written by AIIntermediateRequires adapting existing video diffusion models and integrating condition shifting.
- Create a real‑time text‑to‑video chatbot using APEX.
- Develop a low‑cost video summarization tool for long lectures.
- Implement an interactive game cutscene generator with few‑step sampling.
Ask this paper
/ Ask a questionAnswers come from the full paper and name the section
/ Quiz yourself
What component does APEX eliminate compared to Distribution Matching Distillation?
APEX eliminates the independent fake-score network, using the same generator parameters with a shifted condition for fake estimation.
/ Your notes
Select text in the abstract to highlight it or add a note. They stay in this browser.
