01. Analysis
Of the 1,945 new stars added across the five most-starred AI video repositories on GitHub this week, 1,816 — 93.4% — went to HeyGen Hyperframes. The other four tools split the remaining 129. That ratio is the sharpest signal six weeks of tracking this ecosystem has produced, and it points in a direction that matters for anyone making a build-or-integrate decision in AI video right now.
At 28,780 stars, Hyperframes' lead over second-place LiveKit Agents (11,050) has grown to 17,730 — a gap larger than LiveKit Agents' entire accumulated star count. Six weeks ago, on May 8, the gap between first and second was approximately 11,775 (22,541 vs. 10,766). Since then, nearly 6,000 stars of additional distance have opened. The field isn't losing ground slowly; it's being lapped.
This roundup exists because the June 5 routing guide told you what each tool does and who it's for, and the June 12 creator-tier deep dive gave backgroundremover and ShortGPT the extended treatment they deserved. This one tells you what the six-week trajectory means — and whether the tools behind Hyperframes still deserve a place in your decision-making.
How we researched this
Our research pipeline queried the GitHub AI-for-video topic on June 19, 2026, surfacing the same five repositories we've tracked since May 8. Reddit threads and Hacker News stories returned zero results for AI video tools — the same null result on every research run for this hub, across six consecutive queries. Official pricing pages for Runway, Pika, HeyGen, and Synthesia were fetched but returned no parseable structured data.
What we have is six weeks of precise GitHub star counts and last-updated timestamps for these five repositories. This article synthesizes that longitudinal data against today's June 19 snapshot. All star counts are from our June 19, 2026 research run unless otherwise noted with explicit prior-article dates.
Prior articles I'm drawing from for historical data:
- May 8, 2026: LiveKit Agents vs. TEN Framework comparison — May 8 star counts for both live AI frameworks
- May 15, 2026: Hyperframes vs. ShortGPT comparison — Hyperframes at 22,541, ShortGPT at 7,375
- May 22, 2026: Hyperframes vs. LiveKit Agents — Hyperframes at 23,135
- June 5, 2026: All-five routing roundup — Hyperframes at 24,577, LiveKit at 10,846, TEN at 10,649
- June 12, 2026: Creator tier roundup — Hyperframes at 26,964, LiveKit at 10,953, TEN at 10,668, backgroundremover at 7,918, ShortGPT at 7,401
Six weeks of data
The table below is every star-count data point collected since May 8. Dashes mean that tool wasn't measured in that article's snapshot.
| Date | Hyperframes | LiveKit Agents | TEN Framework | backgroundremover | ShortGPT |
|---|---|---|---|---|---|
| May 8, 2026 | — | 10,766 | 10,629 | — | — |
| May 15, 2026 | 22,541 | — | — | — | 7,375 |
| May 22, 2026 | 23,135 | 10,766 | — | — | — |
| June 5, 2026 | 24,577 | 10,846 | 10,649 | — | — |
| June 12, 2026 | 26,964 | 10,953 | 10,668 | 7,918 | 7,401 |
| June 19, 2026 | 28,780 | 11,050 | 10,685 | 7,925 | 7,409 |
The gradient across those rows is the story. Hyperframes moves in hundreds per week. Everything else moves in single digits.
Week-over-week velocity, June 12 → June 19:
| Tool | Stars June 12 | Stars June 19 | New stars | Stars/day |
|---|---|---|---|---|
| Hyperframes | 26,964 | 28,780 | +1,816 | 259 |
| LiveKit Agents | 10,953 | 11,050 | +97 | 14 |
| TEN Framework | 10,668 | 10,685 | +17 | 2.4 |
| backgroundremover | 7,918 | 7,925 | +7 | 1.0 |
| ShortGPT | 7,401 | 7,409 | +8 | 1.1 |
| Total | 63,904 | 65,849 | +1,945 |
Hyperframes' 259 stars per day is itself decelerating slightly — the June 5 roundup recorded 339 stars per day for the May 30 → June 5 window, and June 12's roundup clocked 341. The pace has dropped by about 24%. But even at the current rate, Hyperframes is adding more stars every single day than LiveKit Agents adds in a week.
What concentration looks like in open-source video
I want to be precise about what these numbers do and don't mean.
GitHub stars measure developer attention — specifically, the moment when a developer decides a repository is interesting enough to bookmark. Star acceleration tells you when a tool is actively circulating: appearing on trending lists, being referenced in newsletters, getting shared in developer communities. A tool adding 259 stars per day isn't doing anything magically different than it was at 200 stars per day. The code hasn't changed. But 259 stars per day means thousands of developers are actively discovering and choosing to track this project right now. That's demand signal.
What stars don't tell you: production adoption, revenue, actual user count, code quality, or whether the tool works reliably at scale. A starred repository is a repository someone found interesting. A deployed repository is one they built with. Those numbers aren't public.
With that framing: the concentration ratio matters, but not because it proves Hyperframes has won. It matters because developer attention is a leading indicator for ecosystem depth — documentation improvements, third-party integrations, Stack Overflow answers, tutorials, and hiring availability. A team betting on Hyperframes in June 2026 is betting that this attention converts to ecosystem support over the next 12 months. The trajectory makes that a reasonable bet. It's still a bet.
The 93.4% concentration ratio is also partly self-reinforcing. When a repository is visibly trending, discovery platforms push it further. When more developers star it, more "awesome" lists and comparison articles reference it (including, obviously, this one). The gap between Hyperframes and the field may continue widening not because Hyperframes is adding capabilities faster, but because it has crossed a visibility threshold that perpetuates itself.
The honest assessment: Hyperframes' design genuinely deserves much of its momentum. Its "Write HTML. Render video. Built for agents." framing landed at exactly the right moment — when AI agent workflows shifted from research curiosity to mainstream software pattern. An HTML-based video template is something a language model can actually author; a timeline-based video editor is not. The tool's core architecture is well-reasoned. Some of the velocity is earned. Some is snowball. You can't fully separate them, and you don't need to.
The case for each runner-up
Given Hyperframes' dominance, I want to be direct about when I'd choose each of the other four tools over it. Not as consolation picks — as the genuinely correct choice for specific jobs.
LiveKit Agents — 11,050 stars, +97 this week
If your use case is an AI that participates in a live video or voice conversation — an interview coach, a virtual customer support agent, a language tutor that responds to what you actually said — Hyperframes is completely irrelevant. It generates video files asynchronously from HTML templates. It cannot see or hear a user in real time.
LiveKit Agents remains the production-hardened choice for live AI voice and video applications. Its 11,050 stars reflect steady accumulation since early 2025 across a community building exactly those applications. The parent LiveKit WebRTC infrastructure has been in production since 2021, which means the debugging community and edge-case documentation exist at a level a newer framework cannot replicate. When something breaks at 2am in a live stream, that inheritance is the difference between a three-hour fix and a three-day one.
The +97 stars this week is slow by Hyperframes' standards. It's steady by any other standard. LiveKit Agents isn't competing for the same developer attention as Hyperframes. It's accumulating the right developers — the ones who have a specific real-time AI application to build and have researched the options carefully. Those populations move at different speeds.
When to choose LiveKit Agents over Hyperframes: Any application requiring real-time, bidirectional AI interaction over voice or video. No contest.
TEN Framework — 10,685 stars, +17 this week
Seventeen stars in a week is TEN Framework's flattest result in our tracking period. LiveKit Agents accumulated 97 in the same span — nearly six times more. The repository was last updated June 19, 2026 (today, per our pipeline), confirming active maintenance. But the star velocity is genuinely slow, which means the developer community is currently not actively discovering this project.
TEN's core argument — AI-native architecture without the inheritance of WebRTC infrastructure complexity — remains valid for greenfield teams. If you're building a conversational AI product from scratch and have no existing LiveKit investment, TEN's cleaner abstractions may reduce conceptual overhead early. That argument hasn't changed since May 8.
What has changed is the compounding gap behind LiveKit Agents. On May 8, 137 stars separated them (10,766 vs. 10,629). Today that gap is 365 stars (11,050 vs. 10,685). The direction of travel matters: LiveKit is accumulating ecosystem resources — plugins, community answers, integrations — faster, and that advantage compounds. In six more months, the gap may become meaningful for teams starting from scratch today.
When to choose TEN Framework: Greenfield conversational AI products where you have no existing LiveKit infrastructure. Run a proof-of-concept sprint with both frameworks before committing to either. The gap is still close enough to evaluate seriously.
backgroundremover — 7,925 stars, +7 this week
Seven stars in a week. That's not a trajectory — it's rounding error. And yet backgroundremover was updated June 19, 2026 per our pipeline (within hours of our research run), does exactly one thing with zero external API dependencies, and runs entirely locally. For any workflow requiring AI background removal without uploading footage to a cloud service, backgroundremover is still the only free, open-source, locally-running option in this dataset.
The slow star velocity isn't decline. It's saturation. The audience for "background removal via command line" is real but finite and has probably already found this tool. You don't need to watch the star count. The update timestamp — same-day as our research run — is the signal that matters: someone is keeping the underlying model current as segmentation architectures improve.
When to choose backgroundremover: Any content workflow where footage privacy matters, where a commercial background removal SaaS is too expensive, or where you simply need a free, local, scriptable option. Star velocity is irrelevant to the tool's utility.
ShortGPT — 7,409 stars, +8 this week
ShortGPT's +8 this week matches its +8 from the June 12 → June 19 window exactly. The repository still carries the "experimental" label in its own description. I'd take that at face value: this is an orchestration framework for developers who want to understand and customize short-form video production automation, not a production-ready pipeline for content teams with publishing deadlines.
The value ShortGPT offers is pipeline transparency. Commercial short-form video generators make opinionated choices at every step — voiceover provider, image generation source, assembly logic — and you accept those choices as part of the product. ShortGPT lets you swap in your own API keys and control every step. For experimental and educational use, that matters. For a team trying to publish 20 videos per week reliably, the experimental label and flat star trajectory are risks worth weighing.
When to choose ShortGPT: Short-form automation experiments with well-defined test criteria and a tolerance for debugging. Not for production publishing operations with real deadlines.
What I'd use and why
| Tool | Best for | Stars (June 19) | Stars/day | Cost floor |
|---|---|---|---|---|
| Hyperframes | Programmatic video generation for agent pipelines | 28,780 | 259 | HeyGen API (usage-based) |
| LiveKit Agents | Live AI voice/video agents | 11,050 | 14 | Open source + optional cloud |
| TEN Framework | Greenfield conversational AI (evaluate vs. LiveKit) | 10,685 | 2.4 | Open source |
| backgroundremover | Local background removal, no cloud upload | 7,925 | 1.0 | Free |
| ShortGPT | Short-form video pipeline experimentation | 7,409 | 1.1 | API costs vary |
If I were starting a new project today where video is an output of an AI agent workflow — something that renders videos from structured data without a human in the production loop — I'd use Hyperframes without deliberation. The star velocity tells me documentation and community support are accumulating fast. HeyGen's backing tells me the project is unlikely to be abandoned. The HTML/CSS rendering model is genuinely clever: language models can write HTML. They cannot reliably operate a video editor's timeline. Hyperframes sidesteps the hardest problem in AI video generation by choosing the right format.
If my project required an AI that participates in live conversations, LiveKit Agents wins — not TEN Framework, despite TEN's architectural merits. At this stage in the trajectory, LiveKit's WebRTC inheritance and faster-compounding community are worth the tradeoff. I'd revisit that call if TEN's weekly velocity recovers above 10 stars.
If I needed background removal with no cloud upload, backgroundremover gets the call immediately. Nothing else in this dataset does that job locally for free.
If I were experimenting with short-form automation and had API budget to test with, I'd run ShortGPT through a defined test project before deciding whether to invest more time.
Limitations
Three structural limits on this data are worth naming clearly.
First, GitHub stars are attention signals, not production adoption metrics. We have no visibility into which tools are actually deployed at scale, how many teams have shipped products with them, or what the real user base looks like.
Second, Reddit and Hacker News continue returning zero results for AI video tool discussions across six consecutive research runs. Whether this reflects genuinely sparse public discussion or private community activity we can't access (Discord, Slack, closed forums) is indeterminate. I've reported the null result every time; I won't paper over it.
Third, commercial pricing for Runway, Pika, HeyGen's platform, and Synthesia remains unavailable from our pipeline. Any open-source vs. commercial comparison on cost is incomplete by design.
Bottom line
Hyperframes is pulling away. Its gap over LiveKit Agents — 17,730 stars — is now larger than LiveKit Agents' entire star count. Ninety-three percent of new stars in the AI video GitHub ecosystem this week went to one tool. Six weeks of data show a single consistent direction: programmatic, agent-driven video generation is where developer attention is converging.
That doesn't make the other four tools irrelevant. LiveKit Agents is still the correct call for live AI interactions; backgroundremover is still the correct call for local background removal; ShortGPT is still viable for automation experiments. The routing logic from June 5 still holds. But the market signal is clear: if you're building in AI video and haven't yet evaluated whether your use case maps to agent-driven programmatic generation, that evaluation is now overdue.
The next data point will tell us whether Hyperframes' deceleration from 339 stars/day to 259 stars/day represents a fading trending effect or a new, lower-but-sustained baseline. Either way, the category has found its center of gravity. The question for builders now isn't which tool is winning — it's whether the winning tool solves the problem you actually have.