This is pretty dang cool but I guess I was hoping to see some very clever use of vision and maybe waveform analysis to give the model "eyes and ears" to edit audio and video. V2
Cool concept, will try it out! I've had decent results with computer use operating conventional editing tools, but being able to directly edit JSON project files is a solid optimization and opens up a lot of opportunities with things like modular templating.
I couldn't find a skill in your project that would help integrate mediabunny. The docs and examples look like they could really help. Having good example coverage allows agents to pickup libraries easily.
Of course, it's impossible to know for sure what was LLM processed or not, but some of your posts (like this one) have been getting classified that way.
I apologise for the activity as some of my replies were AI-assisted. As I wanted to be precise and frame my idea accordingly. I'll keep the guidelines in mind and maintain the legitimacy of my comments and post.
its a real session, not staged: one prompt ("edit a reel from these assets"), claude code plans the cut for ~6 minutes, writes the entire timeline as one json document, and the editor is already live with it. the 24.5s reel it produced is also in the release, untouched: https://github.com/ronak-create/FableCut/releases/tag/v1.3.0
fittingly, the demo video itself was also edited in fablecut.
I have no use for this, however I am stunned by the simple and effective solution of using json+sse instead of an API to allow agent interaction
Amazing work
This looks great, will try it out. I had a crazy idea of doing a full agentic demo video narration and cutting, so might try your tool with some tweaks.
Thanks! That sounds like a really interesting use case. I'd love to see how it turns out. If you end up making those tweaks for full agentic narration and video cutting, feel free to share your experience or ideas, I'd be happy to learn from them and see if they could be incorporated into the tool.
I have little experience with video editing; a tool like this might be an ideal path to learn more. Thanks for creating and publishing it and sharing here!
almost! the editor/compositor is fully browser side, but the server is doing real work: project.json lives on disk (thats the whole point, any agent or script can edit it and fs.watch + sse hot reloads the ui), plus ffmpeg for the final export encode. so it needs the local server for the actual workflow.
a hosted demo mode is a good idea though, project in localstorage + sample footage + mediarecorder export would let people feel the editor without cloning. the one thing it cant show is the main trick, watching an external agent rebuild your timeline live, since theres no local file for it to edit. might do a stripped demo build if theres interest
Since people like an unrelated product, you thought it would be good marketing to jack the brand? Maybe I should start calling my B2B SaaS products Salesforce too!
Sure, but we can't handwave away every observation about LLM output with "(some of) the original training corpora came from humans". Qualitatively, the vibes are off™.
https://github.com/Vanilagy/mediabunny
Of course, it's impossible to know for sure what was LLM processed or not, but some of your posts (like this one) have been getting classified that way.
its a real session, not staged: one prompt ("edit a reel from these assets"), claude code plans the cut for ~6 minutes, writes the entire timeline as one json document, and the editor is already live with it. the 24.5s reel it produced is also in the release, untouched: https://github.com/ronak-create/FableCut/releases/tag/v1.3.0
fittingly, the demo video itself was also edited in fablecut.
can this be used to make these style of videos from prompt?
Style 1: https://www.youtube.com/shorts/vGjrv2zjfsM
Style 2: https://www.youtube.com/shorts/6KoRHpAZ4zk
a hosted demo mode is a good idea though, project in localstorage + sample footage + mediarecorder export would let people feel the editor without cloning. the one thing it cant show is the main trick, watching an external agent rebuild your timeline live, since theres no local file for it to edit. might do a stripped demo build if theres interest
Sure, but we can't handwave away every observation about LLM output with "(some of) the original training corpora came from humans". Qualitatively, the vibes are off™.
https://old.reddit.com/r/ClaudeAI/comments/1uok58g/claudes_s...
https://x.com/emollick/status/2049894109318459798
Remember the gremlin thing?
https://news.ycombinator.com/item?id=47319285
has clawdbot taught us nothing?
Unfortunately anthropic is actively eroding community goodwill in regards to potential IP conflicts:
https://abnormal.ai/blog/abnormal-response-to-anthropic-laws...
https://x.com/steipete/status/2016068265391354181