WTF is with the likes count?
The likes count of 10.8k as of now is more than Comfy-Org/MiniMax-H3 (2.53k) and MiniMaxAI/MiniMax-H3 (6.62k) combined.
Yet, it absolutely doesn't represent the real usage of this model, which has faded away into obscurity in any real AI community.
You may think you can fake the likes, however you cannot fake the downloads of the real users, which speak for themselves: 38 mil downloads last month for MiniMax (32+6 mil) vs. pathetic 2,561,845 for LTX-2.5.
If you are buying the likes to make this model show in the "Trending" tab, I feel very bad for you.
yeah, even I'm entered here to check that thing because it wasn't reasonable.
this's just weird
Ran the numbers on this. The likers list is public (/api/models/<repo>/likers), and each liker's _id encodes when their account was created, so you can see who is liking, not just how many.
Compared against MiniMaxAI/MiniMax-H3, which launched in the same window:
| LTX-2.5 | MiniMax-H3 | |
|---|---|---|
| Total likers | 10,864 | 6,634 |
| From accounts created before Aug 2026 | 4,608 | 5,248 |
| From accounts created Aug–Nov 2026 | 6,256 (57.6%) | 1,386 (20.9%) |
| New-account likers by month (Aug → Sep → Oct) | 1108 → 2101 → 2610 | 690 → 410 → 240 |
| Nov 1–6 | 437 | 46 |
H3 looks like a normal launch: fresh accounts spike in the release month, then decay. LTX-2.5's new-account likes kept climbing for three months after release while downloads stayed flat at ~2.6M/month, and they're still running at roughly 10x H3's pace this month.
Among accounts that existed before August, H3 actually has more likers than LTX-2.5. The entire like lead comes from ~4,900 surplus likes from brand-new accounts, and that's what put this repo on Trending.
I can't tell from this who's behind it, but it isn't organic usage, and it pushes real models down the Trending tab. Would be good to hear from Lightricks or HF on this.
Reproduce it yourself (PowerShell):
$d = Invoke-RestMethod https://huggingface.co/api/models/Lightricks/LTX-2.5/likers; "$($d.Count) likers"; $d | % { [DateTimeOffset]::FromUnixTimeSeconds([Convert]::ToInt64($_._id.Substring(0,8),16)).UtcDateTime.ToString('yyyy-MM') } | Group-Object | Sort-Object Name | ft Name,Count
Swap the repo name to compare.
Even if he bots his likes up to a million, I still wouldn't use it. On the Model Arena, 2.5 scores at least 30% lower than H3.
Ran the numbers on this. The likers list is public (
/api/models/<repo>/likers), and each liker's_idencodes when their account was created, so you can see who is liking, not just how many.Compared against MiniMaxAI/MiniMax-H3, which launched in the same window:
LTX-2.5 MiniMax-H3 Total likers 10,864 6,634 From accounts created before Aug 2026 4,608 5,248 From accounts created Aug–Nov 2026 6,256 (57.6%) 1,386 (20.9%) New-account likers by month (Aug → Sep → Oct) 1108 → 2101 → 2610 690 → 410 → 240 Nov 1–6 437 46 H3 looks like a normal launch: fresh accounts spike in the release month, then decay. LTX-2.5's new-account likes kept climbing for three months after release while downloads stayed flat at ~2.6M/month, and they're still running at roughly 10x H3's pace this month.
Among accounts that existed before August, H3 actually has more likers than LTX-2.5. The entire like lead comes from ~4,900 surplus likes from brand-new accounts, and that's what put this repo on Trending.
I can't tell from this who's behind it, but it isn't organic usage, and it pushes real models down the Trending tab. Would be good to hear from Lightricks or HF on this.
Reproduce it yourself (PowerShell):
$d = Invoke-RestMethod https://huggingface.co/api/models/Lightricks/LTX-2.5/likers; "$($d.Count) likers"; $d | % { [DateTimeOffset]::FromUnixTimeSeconds([Convert]::ToInt64($_._id.Substring(0,8),16)).UtcDateTime.ToString('yyyy-MM') } | Group-Object | Sort-Object Name | ft Name,CountSwap the repo name to compare.
https://huggingface.co/Comfy-Org/MiniMax-H3
H3's download count on the ComfyUI repository has increased by 10 million this month, going from 20 million to over 30 million, while 2.5 has only gained a few hundred thousand downloads.
In terms of like ratios, LTX 2.5 is currently far ahead of other projects as well. However, the problem is that LTX 2.5 is much less popular than H3 on Civitai right now. So your suspicion is justified: the like activity from new LTX 2.5 users is abnormal.
But none of that really matters. H3 has already become an essential production tool for short-form dramas. Unless LTX comes up with a better model, no amount of artificially boosting likes will make a difference.