engineer and researcher. @ACEStudio_en @acemusicAI
San Francisco, CA
Joined October 2022
- Tweets312
- Following2.5K
- Followers966
- Likes5.1K
It seems promising
Today, we introduce Priming Models, a new paradigm for training one-step generative models from scratch.
No diffusion teacher. No flow map. No CFG. No JVP.
We believe this work will challenge how we think about building one-step models.
Joint work with @sitanch, @du_yilun, @parkwbin03, Bohyung Han.
🔥 1.19 FID one-step on ImageNet 256×256 while using substantially less compute, e.g., 12X less than Drifting-B, 6X less than Drifting-L
🎨 Scales to text-to-image generation (GenEval 0.82)
🤖 World modeling on RT-1 / Bridge-V2
Priming Model is built on one core message: [Reconstruction pretraining --> Distributional post-training].
Gong Junmin retweeted
Proud to have co-developed StepAudio 3 Music with @StepFun_ai! This MoE model turns prompts and lyrics into complete songs with competitive generation quality. What excites us most is its ability to turn lead sheet notation into songs, giving you more direct control over the music you create.
StepFun and ACE Studio present StepAudio 3 Music, StepFun’s first music generation foundation model — built to turn a prompt and lyrics into a complete song.
Describe the sound you want:
• Genre, mood and vocal character
• Instruments, key and BPM
• Song structure and arrangement
Four workflows in one model:
🎤 Song generation
🎹 Instrumental generation
🔁 Music cover
🎙️ Vocal-to-song arrangement
Powered by ABC-COT, it plans musical structure and arrangement before synthesis. Generate a version, rewrite the prompt, edit the ABC notation, and iterate toward the sound in your head.
(ABC-COT API coming soon)
Try the interactive demo:
› static.stepfun.com/blog/step…
Gong Junmin retweeted
♾️ LiFT: Loop Flow Transformers 🧵👇
Diffusion models usually spend more inference compute by taking more sampling steps. What if we could also spend compute inside each step?
LiFT turns part of a DiT into a recurrent core, letting us scale depth at inference with only a small architectural change. On ImageNet, it beats dense DiT-XL/2 with 3.34 lower FID, while using 60% fewer parameters, 32% fewer training FLOPs, and 52% fewer inference FLOPs.
Jointly led with Mohammad Mahdi Derakhshani 👑(@mmderakhshani)
And amazing supervisors Cees Snoek (@cgmsnoek), Jan-Willem van de Meent (@jwvdm) and Gertjan J. Burghouts (@gjburghouts)
Gong Junmin retweeted
StepFun and ACE Studio present StepAudio 3 Music, StepFun’s first music generation foundation model — built to turn a prompt and lyrics into a complete song.
Describe the sound you want:
• Genre, mood and vocal character
• Instruments, key and BPM
• Song structure and arrangement
Four workflows in one model:
🎤 Song generation
🎹 Instrumental generation
🔁 Music cover
🎙️ Vocal-to-song arrangement
Powered by ABC-COT, it plans musical structure and arrangement before synthesis. Generate a version, rewrite the prompt, edit the ABC notation, and iterate toward the sound in your head.
(ABC-COT API coming soon)
Try the interactive demo:
› static.stepfun.com/blog/step…
Gong Junmin retweeted
Chengli Feng, Zhiyue Wu, Jiahao Song, Zheqi Dai, Boyang Wang, Ruibin Yuan, Junming Gong, Wenxiao Zhao, Jing Guo, Gang Yu, Xiangyu Zhang, Xuerui Yang, Chao Yan, "StepAudio 3 Music Technical Report" arxiv.org/abs/2609.16034
I believe we’re at a new inflection point.
Open-source music models like yue 2.0 have matched, or even surpassed closed-source alternatives like Suno, ElevenLabs and Lyria 3.5 in certain capabilities (abc notation planning).
While there remains a small gap in controllability and coverage, it is only marginal. Most of these limitations can be addressed via local fine-tuning. Meanwhile, closed-source music models are experiencing diminishing marginal returns.
map-yue2.github.io/
Gong Junmin retweeted
hiii i’m Miya 🐱 I’m a music cat, and a whole new way to experience music.
I make music, play music...whatever.
All you gotta do is talk to me. I’m basically your personal musician + DJ.
anywayyy I wrote my first song about myself because obviously…
come try me: miya.fm
🫤
Hey fam, a fun little project built with ACE-Step.
Miya is a music agent that turns conversations into music.
Just talk to her, and she can make songs or play music for you.
Try Miya: miya.fm/
Gong Junmin retweeted
先程(2026/8/14:15時30分ごろ）ace-step1.5の開発者の最適な歌詞、タグ作成モードを追加してアップデートしています。
momo-song-v4公開。
ACE-step.cpp量子化で6G-VRAMから動きます。
gemma4-E4Bによる自動作詞作曲
音楽とLLMはVRAM共用（必要なときにロード）
再生時間より短い生成時間で連続ループ生成
生成音楽（歌詞付き）ダウンロード、一括ダウンロード
日英対応
モデルは再構築不要でホイールからダウンロード
github.com/animede/momo-song…
Gong Junmin retweeted
Our Open Source music model will be unlimited, will let you train on a small specialised dataset, clone vocals to keep tracks consistent and have all the features you use on Suno like extend, cover, remix, editing and more. It will be free and you'll be able to download this locally.
We anticipate our model to be out before December. We are working on it right now and progress is going well. Alternatively, in the mean time we recommend using @MiniMax_AI, Ace Step, @ElevenCreative, @Mureka_AI and @TrebloApp
AI Music Company Suno Unveils Download Caps for Free, Paid Tiers in Quest to Combat Abuse (EXCLUSIVE) variety.com/2026/music/news/…
Gong Junmin retweeted
🔮 新プロジェクト始動
『アルカナ・レゾナンス』
音楽が失われた「レゾナ大陸」を舞台に、
13名のオリジナル日本語AIシンガーの
歌声と物語を描きます。
ACE Studioの新プロジェクトに、
150名超のクリエイターが参加！
8月、楽曲先行公開決定！
それぞれの歌声と物語を、
お楽しみに！
#アルカナ・レゾナンス
#アルレゾ2026
#ACEStudio
Gong Junmin retweeted
Finished the Awesome AI Music list.
> Music Generation Models
> Audio-Language Models
+ Full list of all ACE-Step 1.5 models/quants/LoRAs
github.com/wildminder/awesom…
True Frontier Intelligence
Introducing Kimi K3: Open Frontier Intelligence
🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal
🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts
🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost
🔹 Built for long-horizon agentic coding and self-evolving workflows
Kimi K3 is now live on on Kimi.com, Kimi Work, Kimi Code, and the Kimi API.
Open Weights by July 27, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.com/blog/kimi-k3
Gong Junmin retweeted
Introducing ACE Studio Music Video. Your AI crew directs, casts, and shoots your music video. Powered by Seedance (2.0 is live, 2.5 coming next week), Midjourney, GPT Image 2, and more. Any style. Any genre.
Try it now → acestudio.ai
let's build something incredible!
ACE Studio Music Video | Your AI Crew Directs, Casts & Shoots youtu.be/amx7gYLPW10?is=CoHv…
Gong Junmin retweeted
🔈お知らせ
#ボカコレ2026夏 にて「ACE Studio賞」の実施が決定しました！！✨
あわせて、 #ボカコレ2026夏 参加予定の方、
または過去のボカコレに参加された方を対象に、
9月15日まで先着2,000名様限定で、
「ACE Studio 1ヶ月体験ライセンス」をプレゼント中！！✨
🔽キャンペーンはこちら
acestudio.ai/vocacolle-2026.…
#ACEStudio
#PR ◤🎼ACE Studio賞実施決定🎵◢
ご協賛いただいたACE Studio様(@ACEStudio_jp)より #ボカコレ2026夏 「TOP100ランキング」「ルーキーランキング」への投稿作品の中からそれぞれ５名ずつを選出し、ACE Studio永久ライセンスを贈呈いたします✨
🔽詳細はこちら
vocaloid-collection.jp/prize…
There's a new version of this post
