RenderBob by TraynMe
We make generative video ready for the edit.
We're building toward a timeline where editors generate, fix, and swap shots without leaving the cut, on compute that's priced before it runs.
Generative video, the compute under it, and the tools that bring it into the edit.
Generative video. We test the models that make moving pictures: text and image to video, reference control, and sound generated with the picture. We write down what holds up, what drifts, and what each second costs.
The compute layer. RenderBob 2 ran GPU rendering in the cloud for 3D and VFX teams. We know what a job costs, where it fails, and when a local machine is enough.
AI in the edit. A generated clip is half the job. The rest happens on the timeline: finding the shot, keeping a cast consistent, and swapping one take without breaking the cut. That is where we build now.
RenderBob 2 was our cloud rendering service for 3D and VFX projects. Frames rendered: 50,000+. Countries: 40+. Job success rate: 99.2%. Average render time: ~12 min. GPU hours delivered: 1,875.
“I must say this is, hands down, my best render farm experience so far. It's fast, easy to use, and refreshingly different!” (Nok Animation Studios, RenderBob 2 customer)
Price comes first. Teams wanted the cost before the render started. RenderBob 2 quoted every job up front, and the quote did not move after submit.
Catch failures before the GPU starts. Automated checks for missing textures and dependencies ran before a job was billed. 99.2% of jobs completed.
Local hardware covers more than you think. Creator machines already handle much of traditional 3D. The cloud earns its place on heavy, short bursts of work.
Plain words beat a settings manual. Bob AI, our render copilot, answered questions about errors, settings, and estimates in plain language, inside the tool.
Bob Labs (https://labs.renderbob.ai/) is our open bench of concepts for generating a shot and cutting it.
From the blog
- YouTube Now Counts Estimated TV Co-Viewers in Public View Counts: What Changes for Creators and What Stays the Same
YouTube announced on 16 November that public view counts now include estimated co-viewers on TV, so one TV playback can register as more than one view.
- "Not AI's To Take": Australian Media and Industry Bodies Launch a Campaign Against Weakening Copyright for AI Training
Australian media companies and industry bodies launched Not AI's To Take on 18 November, asking the government to reject any weakening of copyright for AI training.
- Frankensplat Trains Gaussian Splats Inside a Browser Tab: What WebGPU Makes Possible and What the Benchmark Does Not Show
Rafael Spring's Frankensplat trains Gaussian splats from COLMAP reconstructions in a browser tab with WebGPU. The speed and quality figures are the developer's.