01. Analysis
The brief deceleration in week six was not a trend reversal. It was a pause.
HeyGen Hyperframes added 2,645 GitHub stars in the seven days ending June 26, 2026 — 378 stars per day, the highest daily rate this series has recorded across seven consecutive weeks of tracking. For reference: the June 5 roundup described a 339-star-per-day pace as "the kind of velocity that means this project is actively surfacing on developer feeds, 'awesome' lists, and shared links." This week's rate is 11.5% higher than that. The June 19 roundup noted a deceleration to 259 stars per day and flagged it as a signal worth watching. The answer to what it signaled is now clear: it was a dip, not a structural change.
The four other tools in our dataset — LiveKit Agents, TEN Framework, backgroundremover, and ShortGPT — added a combined 152 stars this week. Hyperframes captured 94.6% of new developer attention across these five repositories. Last week's concentration ratio was 93.4%. The pattern did not soften. It sharpened.
At 31,425 stars, Hyperframes is now 20,279 ahead of LiveKit Agents, the second-place tool. That gap is larger than LiveKit Agents' entire accumulated star count.
How we researched this
Our research pipeline queried the GitHub AI-for-video topic on June 26, 2026, returning the same five repositories tracked since May 8. Reddit and Hacker News returned zero results for AI video tools — the seventh consecutive week with null results from both platforms. Official pricing pages for Runway, Pika, HeyGen, and Synthesia were fetched but returned no parseable structured data. This article's analysis is built entirely from GitHub metadata: star counts, descriptions, and last-updated timestamps.
Prior data points I'm drawing from for historical reference:
- May 8, 2026: LiveKit Agents vs. TEN Framework comparison — first star counts for both live-agent frameworks
- May 15, 2026: Hyperframes vs. ShortGPT — Hyperframes at 22,541, ShortGPT at 7,375
- May 22, 2026: Hyperframes vs. LiveKit — Hyperframes at 23,135, LiveKit at 10,766
- June 5, 2026: All-five routing roundup — Hyperframes 24,577, LiveKit 10,846, TEN 10,649
- June 12, 2026: Creator tier roundup — Hyperframes 26,964, LiveKit 10,953, TEN 10,668, backgroundremover 7,918, ShortGPT 7,401
- June 19, 2026: Concentration roundup — Hyperframes 28,780, LiveKit 11,050, TEN 10,685, backgroundremover 7,925, ShortGPT 7,409
All star counts in this article are from the June 26, 2026 research run unless a prior-article date is explicitly stated.
The seven-week data table
| Date | Hyperframes | LiveKit Agents | TEN Framework | backgroundremover | ShortGPT | Total |
|---|---|---|---|---|---|---|
| May 8 | — | 10,766 | 10,629 | — | — | — |
| May 15 | 22,541 | — | — | — | 7,375 | — |
| May 22 | 23,135 | 10,766 | — | — | — | — |
| June 5 | 24,577 | 10,846 | 10,649 | — | — | ~61,000 |
| June 12 | 26,964 | 10,953 | 10,668 | 7,918 | 7,401 | 63,904 |
| June 19 | 28,780 | 11,050 | 10,685 | 7,925 | 7,409 | 65,849 |
| June 26 | 31,425 | 11,146 | 10,707 | 7,940 | 7,428 | 68,646 |
The bottom row is today's research output. Everything above it is from previously published ToolSift articles in this series.
Week-over-week velocity: June 19 → June 26
| Tool | June 19 | June 26 | New stars | Stars/day |
|---|---|---|---|---|
| Hyperframes | 28,780 | 31,425 | +2,645 | 378 |
| LiveKit Agents | 11,050 | 11,146 | +96 | 13.7 |
| TEN Framework | 10,685 | 10,707 | +22 | 3.1 |
| backgroundremover | 7,925 | 7,940 | +15 | 2.1 |
| ShortGPT | 7,409 | 7,428 | +19 | 2.7 |
| Total | 65,849 | 68,646 | +2,797 | — |
Hyperframes' share of new stars: 2,645 / 2,797 = 94.6%
The velocity curve, reconstructed
Seven weeks of data now let me reconstruct Hyperframes' stars-per-day trajectory with reasonable precision:
- ~May 30 → June 5: 339 stars/day (stated in the June 5 article)
- June 5 → June 12: 341 stars/day (26,964 − 24,577 = 2,387 / 7)
- June 12 → June 19: 259 stars/day (28,780 − 26,964 = 1,816 / 7)
- June 19 → June 26: 378 stars/day (31,425 − 28,780 = 2,645 / 7)
The week ending June 19 was the first and only significant deceleration in this series. It showed up, I wrote about it, and one week later it reversed with authority. The 378-star-per-day rate is 39 points higher than the previous ceiling. Whatever caused the June 12-19 slowdown — a quieter week on developer newsletters, a competing announcement that briefly pulled attention elsewhere, normal weekly variance — it was transient. The underlying demand signal has not changed direction.
For further context: Hyperframes has added 8,884 stars in the six weeks since we first recorded it at 22,541 on May 15. That's 1,481 stars per week on average. This week's 2,645-star gain is 78% above that average. This is not a tool hitting its natural ceiling.
What 31,425 means relative to the field
Numbers need reference points to be meaningful.
Hyperframes (31,425) is now larger than LiveKit Agents and TEN Framework combined: 11,146 + 10,707 = 21,853. The first-place tool has more stars than the second through third place tools added together.
The gap between Hyperframes and LiveKit Agents — 20,279 — has grown by 2,549 in a single week (from 17,730 on June 19). At current trajectories, Hyperframes would cross 50,000 stars before LiveKit crosses 12,000.
The gap between Hyperframes and the entire rest of the field (all four other tools): 31,425 vs. 37,221. Hyperframes represents 45.8% of total stars across five repositories. In week one of this series (approximately May 8), it was not even measured — LiveKit and TEN were near parity at roughly 10,766 and 10,629. The landscape has been reshaped.
The self-reinforcing structure
I've resisted calling this a winner-take-all market in prior roundups because that framing overstates what GitHub stars measure. I want to push on the mechanism more directly now that we have seven data points.
Developer attention compounds in open-source ecosystems. A repository adding 378 stars per day is being discovered by ~378 developers every day who decide it's worth bookmarking. Some fraction of those developers will write tutorials, which surface on Google. Some will ask or answer questions on Stack Overflow, which surfaces in searches. Some will add Hyperframes to "awesome" lists and comparison tables — a category I am, obviously, directly contributing to right now. Each of those downstream artifacts creates new discovery surfaces that funnel more developers toward Hyperframes, which produces more stars, which increases its visibility rankings.
The theoretical endpoint of this loop is a repository with so much ecosystem gravity that alternatives functionally cannot compete for developer attention, even if they are technically superior for specific tasks. I don't think we're there yet for AI video. But the gap is now wide enough that I'd start asking: what would it take for LiveKit Agents or TEN Framework to close it? The answer probably isn't "write better code." It's "get a step-function visibility event" — a major platform announcement, a viral developer thread, a high-profile deployment case study. Organic weekly accumulation at current rates won't close a 20,000-star gap.
This is not an argument that Hyperframes is better software than its alternatives for every use case. It's an observation that developer attention has its own momentum, and the trailing tools are competing against that momentum in addition to competing on technical merits.
The five tools, current assessment
HeyGen Hyperframes — 31,425 stars
Description from the repository: "Write HTML. Render video. Built for agents."
This framing is the key to understanding Hyperframes' velocity. The "built for agents" positioning hit the market at exactly the moment when AI agent workflows stopped being a research category and became a mainstream software pattern. HTML is a templating language language models can actually author — a video generation API that takes structured HTML inputs is natively compatible with agent pipelines in a way that timeline-based video editors are not. The tool's architecture earned its attention.
Last updated: June 26, 2026. Actively maintained.
Who should use it: Any developer generating video programmatically at scale — product demos, data-driven reports, personalized marketing content, agent-authored explainer videos. There is no meaningful alternative in this dataset for this job.
LiveKit Agents — 11,146 stars, +96 this week
Description: "A framework for building realtime voice AI agents."
Thirteen days ago I would have said LiveKit was holding its position while Hyperframes ran away. That's still roughly true — +96 stars this week is slow, but it represents continuous organic interest, not abandonment. LiveKit's stability at roughly 13-14 new stars per day suggests a baseline of genuine user demand rather than viral-driven spikes.
Last updated: June 26, 2026. Actively maintained.
Who should use it: Any developer building a live AI assistant that needs to participate in audio or video calls — interview prep tools, real-time coaching, AI meeting participants, voice-driven customer service. Hyperframes cannot do this. These tools don't compete.
TEN Framework — 10,707 stars, +22 this week
Description: "Open-source framework for conversational voice AI agents."
TEN and LiveKit are solving the same problem and have converged to near-identical star counts — a 439-star gap after seven weeks of separate tracking. Three stars per day is slower than LiveKit's pace, and TEN's last update was June 25 (one day before LiveKit's June 26). Neither of these details is disqualifying; both tools are actively maintained and credible.
Last updated: June 25, 2026. Actively maintained.
Who should use it: Developers building conversational voice AI who have a specific architectural reason to prefer TEN's framework design over LiveKit's. The 439-star gap matters less than which framework's documentation and community answers your first ten questions.
backgroundremover — 7,940 stars, +15 this week
Description: "Background Remover lets you Remove Background from images and video using AI with a simple command line interface that is free and open source."
Fifteen new stars this week. Two stars per day. The tool does exactly one thing — removes backgrounds from images and video using AI inference, locally, via a command-line interface — and it does it without sending your footage to any third-party API. For a developer building an automated pipeline where data privacy matters or per-frame API costs are prohibitive, this specific combination of capabilities has no equivalent in this dataset.
Last updated: June 26, 2026. Actively maintained.
Who should use it: Automated pipelines requiring background removal from video with local inference, no API calls, no data leaving the machine. This is a narrow but real use case, and backgroundremover is the right tool for it.
ShortGPT — 7,428 stars, +19 this week
Description: "Experimental AI framework for youtube shorts / tiktok channel automation."
"Experimental" in the repository's own description is the operative word. ShortGPT gained 19 stars this week, which is marginally faster than backgroundremover despite being 512 stars behind it in total. It targets a workflow — automating short-form video creation at volume for platforms like YouTube Shorts and TikTok — that none of the other four tools address. The fact that a project framed as experimental has accumulated 7,428 stars suggests real demand exists for this category, even if ShortGPT itself isn't production-ready.
Last updated: June 26, 2026. Actively maintained.
Who should use it: Developers experimenting with automated short-form video pipelines. Not for production systems with reliability requirements. Treat it as a research artifact and starting point for a custom implementation, not a dependency you ship.
What we'd use and why
Seven weeks in, my position on tooling recommendations hasn't changed, but the confidence levels have shifted.
For programmatic video from structured data, Hyperframes is the obvious choice and the gap between it and any alternative has only grown. I would use it without hesitation for an agent-driven video generation workflow.
For live AI voice or video participation, I'd choose LiveKit Agents. Not because TEN Framework is worse — they're genuinely comparable — but because three additional weeks of tracking confirm LiveKit's slightly higher weekly velocity, and in a market where ecosystem support matters, the marginally faster-growing community is the better bet for a production dependency. If TEN's architecture better fits your use case after reading both docs, that's a legitimate reason to pick it; the star gap alone is not disqualifying.
For automated background removal from video at scale, backgroundremover. Its zero-API, local-inference design is the deciding factor for privacy-sensitive or cost-sensitive pipelines.
For short-form video automation experimentation, ShortGPT with the understanding that "experimental" means exactly that. Use it to scope what's possible before building something custom.
For commercial AI video tools — Runway, Pika, HeyGen's paid platform, Synthesia, Kling, Sora — this dataset doesn't cover them. Our research pipeline attempted to fetch their pricing pages and returned nothing parseable. Any recommendation there would require separate primary research.
Limitations
The same constraints that applied to weeks one through six apply now.
Reddit and Hacker News have returned zero results across seven consecutive research runs for the AI-for-video hub. I have no community discourse, no user failure reports, no "what broke in production" testimonials. The analysis is built on GitHub signals only. If the important conversations about these tools are happening in private Slack groups, Discord servers, or internal engineering blogs, they are invisible to this series.
GitHub stars measure discovery and interest, not production adoption. All five tools in this dataset could have zero production deployments or ten thousand. The data doesn't distinguish.
This dataset covers five repositories. AI video as a market is broader than these five projects — commercial SaaS tools, closed-source models, and the dozens of open-source projects below the top-five threshold are not represented. Conclusions about "the AI video market" that extrapolate from this five-tool sample are overstated.
Weekly snapshots create temporal blind spots. A single viral event mid-week could inflate or deflate a given data point. Seven snapshots reduce but don't eliminate this variance.
Bottom line
The June 19 deceleration to 259 stars per day was the only break in Hyperframes' trajectory across seven weeks of tracking. This week it's gone — replaced by a new high of 378 stars per day and a concentration ratio of 94.6%, the highest this series has recorded.
The more consequential question isn't whether Hyperframes is dominant — at this point, that's settled. It's whether the compounding attention effect I described above has reached the threshold where it structurally disadvantages the trailing tools in ways that extend beyond star counts: documentation lag, reduced contributor pull, slower integration support. We don't yet have data that answers that question. What we have is seven consecutive weeks of a gap widening at roughly 2,500 stars per week, with every tool in the dataset actively maintained and updated.
The stars are asymmetric. The maintenance is not. That tension is worth watching.
I'll have week eight numbers on July 3.