Thread

Show HN: RxFilm Studio–Create and edit your product videos with AI agent

19 points · 15 comments · zilue

Loading the complete thread in the background. This saved snapshot is available now. Refresh

cs1996 · 1790167801 · focus · HN ↗
nice, i really like how it just knew to use my default codex!
1. zilue · 1790175997 · focus · HN ↗
  Thanks for your feedback!
  Streaming agent's workflow is a good suggestion! We actually already have this feature - if you click the chat bubble button at the top-right corner, just below “Preview,” you’ll be able to see it. That said, the UI definitely needs to improve. We will try to make the AI agent's workflow more visible in our next update.
  I’m currently looking into the issue that’s causing it to get stuck. It should work if you would like to try another way: connect the film studio MCP to your codex and let codex call the film studio mcp to create a video for your product.
  1. cs1996 · 1790177068 · focus · HN ↗
    <a href="https://github.com/andrewarrow/film-workflow/commit/60581de37bd41e0a8d5f36c599dbca53998794f5" rel="nofollow">https://github.com/andrewarrow/film-workflow/commit/60581de3...
    <a href="https://chatgpt.com/share/6ab3eee3-fcdc-83e8-a04f-9635e93508e0" rel="nofollow">https://chatgpt.com/share/6ab3eee3-fcdc-83e8-a04f-9635e93508...
    I took a crack at why the MCP tools needed but didn't get very far. When I see MCP I think streaming tool to streaming tool and that's why the MCP part is needed but I wonder if it gets over used for things that could just be skills or prompts. Why is MCP even needed here?
    zilue · 1790179337 · focus · HN ↗
    Hi andrew, MCPs are required here because the only way that our app can communicate with the agents like Claude Code and Codex is through MCP protocol. Skills and prompt are also the key part in our app to provide the best in class experience for different agents
    cs1996 · 1790186057 · focus · HN ↗
    I'm using Sign in with ChatGPT and then Codex App Server protocol for my desktop app:
    <a href="https://github.com/cubacadabra/studio" rel="nofollow">https://github.com/cubacadabra/studio
    <a href="https://github.com/cubacadabra/studio/blob/main/src/codex.rs#L249" rel="nofollow">https://github.com/cubacadabra/studio/blob/main/src/codex.rs...
    <a href="https://www.youtube.com/watch?v=bTawLd5r2mw" rel="nofollow">https://www.youtube.com/watch?v=bTawLd5r2mw
    <a href="https://learn.chatgpt.com/docs/app-server" rel="nofollow">https://learn.chatgpt.com/docs/app-server
xnx · 1790168317 · focus · HN ↗
Kind of like Pommeli (<a href="https://labs.google.com/pomelli" rel="nofollow">https://labs.google.com/pomelli) and Google Veo.
1. sandworm101 · 1790168748 · focus · HN ↗
  Its kinda like a dozen UI projects that are little more than a pretty layer above common tools. Music/voice composition, image generation and first-frame/last-frame video creation... this can all be done in comfy using free/oss tools.
2. zilue · 1790179526 · focus · HN ↗
  Our product uses different popular models including the one from Google but provide so much more functionalities～
rkeswick · 1790169438 · focus · HN ↗
My last product demo video took forever to edit. Hoping your AI agent can actually make that process much less painful.
1. zilue · 1790179974 · focus · HN ↗
  Hi rkeswick, thanks for your interest! If u have any issues or problems when using our app, please feel free to submit your issues on GitHub, we will address it as soon as possible.
jakaroo · 1790169702 · focus · HN ↗
Congrats on the launch. I like the voice used in the demo marketing video, what name does it have in the app?
1. zilue · 1790177554 · focus · HN ↗
  [dead]
boobsbr · 1790170954 · focus · HN ↗
Stop hijacking the scroll!
It's been over a decade that we've been complaining about this and yet smart, intelligent people still think it's cool to do it.
1. 16bitvoid · 1790174196 · focus · HN ↗
  To be fair, in this case it wasn't a person at all. It was clearly Claude or some other agent.
hn45e7pbij · 1790175467 · focus · HN ↗
[dead]
DougN7 · 1790180651 · focus · HN ↗
I have such a hard time imagining this working well. I used an AI service to make a picture of an archer in a different position from where the image originally showed him and it consistently insisted on swapping the arrow tip and the feathers, so feathers at the front of the arrow. Just because you give these things instructions does not mean they get followed precisely.
1. zilue · 1790230426 · focus · HN ↗
  That’s a fair concern DougN7. Generative image/video models are probabilistic, so they can easily cause mistakes like the arrow example you mentioned.
  Film Studio works differently for many edits: It is not a generative image/video model. Instead, the instruction will be translated into deterministic code/parameters, which are then rendered with Remotion. So “move this 20px left” or “start this animation at frame 60” becomes an exact operation, rather than a suggestion to a generative model.
  It’s more like an AI-powered version of Adobe After Effects + Premiere
sofiarossi98 · 1790286792 · focus · HN ↗
[dead]
assouma950 · 1790464637 · focus · HN ↗
[dead]

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.