@atomtanstudio

Shipped Sound / Vision an open source music and music video creation app. Get it here: https://xitter.cf/t.co/xnrvSuFWy9

 Internet-native studio
atomtan.studio
Joined December 2025
  • Tweets4.7K
  • Following68
  • Followers388
936Media
I built Sound / Vision: songs, album covers and music videos, made on your own GPU. One sentence becomes a song, then a cover, then a video. Open source. Run local models, or the frontier ones you already pay for. github.com/atomtanstudio/sou…
5
2
1
11
1,502
Here is a remake of the classic arcade game NBA Jam! using deepseek-v4.1-flash-1017 and MiniMax-M3.1-Flash. The Deepseek version was generated inside ZCode using my @alibaba_cloud account and the Minimax version was using Minimax Code. As you can see, Minimax did not complete the task. I used Matt Schumer's gauntlet loop for both. I noticed that for both games I tried with Minimax 3.1 that it threw in the towel and said that it couldn't close the gap. Odd to see that from a model, but I suppose it is appreciated rather than wasting tokens.
6
222
I'm incredibly grateful for the labs that are releasing open models as they are getting better and better by the day, but as I've been testing Minimax 3.1 Flash and Deepseek 4.1 1017 Flash today, I'm reminded of just how far ahead Opus 5.5 is at the moment. It's a tremendous gulf.
33
Dear @bot please create a browser extension called xfoliate that will keep politics off my timeline and the what's happening section of X.
4
8
2,912
Holy shit you're doing it!
1
1
56
So far MiniMax-M3.1-Flash is not wowing me on the game development front. ui7738qyy6ip9.space.minimax.…
34
John McCoy
@johnmccoyx16h
Both of these are 100% made with AI Left is Seedance 2.5 at $1.44 Right is Wan 3.0 at $0.80 not only is Wan 50% cheaper but it's way more realistic and it's also a clean 60fps unlike Seedance at a rough 24fps
38
Here is MiniMax-M3.1-Flash
Rich · Atom Tan Studio
@atomtanstudioOct 17
Anyone who thinks that DeepSeek 4.1 1017 is Opus 5.5 level, here is your answer. At least with regard to SVG's. Create a beautiful SVG animation of a rhino riding a giraffe across the savanna.
6
198
With the unlimited access to Minimax 3.1 Flash in Minimax Code this week, I'm going to be doing all the gauntlet loop games at 1M Max. Let me know if you have any requests.
39
Is there a model that is even in the same stratosphere as Gemini when it comes to analyzing music? Every frontier model is really bad by comparison.
1
26
deepseek-v4.1-flash-1017 is live on @alibaba_cloud if you are a token plan subscriber.
3
155
Here are a couple of MOSS vs. KROMOSS examples. KROMOSS is a brand new fine tune of the fantastic MOSS image model. The model for each example is in the ALT.
MOSS Example 1

ALT MOSS Example 1

KROMOSS Example 1

ALT KROMOSS Example 1

MOSS Example 2

ALT MOSS Example 2

KROMOSS Example 2

ALT KROMOSS Example 2

1
80
Song: Miike Snow - Black Tin Box Video created and upscaled in H3LIX using the video agent director using the visual poem preset. Images: GPT Image 2.5 Video: Minimax H3 upscaled with Starlight Precise 2.6
1
115
Grok @bot created this video in 6 hours and 18 minutes with me attaching the song and giving it this prompt. Can you take a listen to this? It's an instrumental song and create a music video like yesterday but use a different set up and scenario given the complete difference musically? Grok Imagine video can beautifully cinematic, but it had do redo nearly the entire video and it still has issues. Hopefully there will be a better video release from their team sometime soon.
1
167
Just an FYI to any of you on the @alibaba_cloud token plan. Deepseek 4.1 1017 Flash is not being hosted yet. I did some testing against the official model from DSH and the difference is pretty staggering.
71
Rich · Atom Tan Studio retweeted
Ting Chen Liang
@ting_Oct 17
we just open-sourced our image model MOSS-Image!! @MosiAI_Official @Open_MOSS 🔥 a 5B Diffusion Transformer with strong Chinese & English text rendering > JSON layouts for text placement & object arrangement > @ComfyUI support > Base & SFT weights > Apache-2.0 come try it and build with us!
MOSS
@MOSS_GlobalOct 16
🤗 MOSS-Image is now open source on @huggingface . Built on a 5B Diffusion Transformer (DiT), MOSS-Image brings Chinese & English text rendering and controllable layouts to image generation — from posters and infographics to creative visuals.@Open_MOSS What it offers: 📝 Long prompts & text rendering — up to 4K-token prompts, including multi-line text 📐 Experimental JSON layout control — specify text, objects, and their positions using bounding boxes 🖼️ Flexible generation — multiple aspect ratios, resolutions, and customizable inference settings 🛠️ Developer workflows — Python (Diffusers), Gradio, ComfyUI, and a Prompt Enhancer Skill 📦 Base & SFT checkpoints are both available, with open model weights and Apache-2.0 licensed inference code. 🙌 Huge thanks to @sgl_project @vllm_project @ModelScope2022 for the Day-0 support!
9
3
84
4,503
I have to give Grok @bot credit for its candor. Hopefully there is a better Grok Imagine video in the pipeline.
29
Here is a quick look at my latest project H3LIX. One of the options is Projects. This can be just about anything either image, video or sound related. You pick options and then collaborate with your LLM of choice on the outcome you want.
32
Ok Grok @bot I am starting to like you more. I wanted to find a specific prompt a certain influencer uses in his videos and I just pointed Grok Bot at his YouTube page and it went through the transcripts and verified with screen shots of the videos the precise prompt.
37
Song: Obsidian Temple (Suno 6.0) Images: @MosiAI_Official in H3LIX app Video: @MiniMax_AI H3 in H3LIX app Directed by Opus 5.5 as a director also in H3LIX app. All image and video generation was local using an RTX5090. Total project time was 2 hours and 10 minutes with most of that time rendering video.
1
4
338
There are some seam issues that I am fixing now. This is a challenge for all video models when you merge two clips together. Even when you line up the frames properly, there's generally some sort of camera shift that's pretty jarring. In my sound and vision app, I was able to smooth these transitions out and they look good, and so I just need to take the code and move it over.
19
I refuse to engage with people who make dumb statements like this. Large companies like OpenAI have multiple teams working on their various products all at the same time. Just because Anthropic is clearly the leader at the moment doesn't mean they just stop all of their other work. These kinds of statements tell me that the poster knows nothing about how any of this works.
23
Load more