Linum
Linum trains text-to-video models from scratch and ships the weights under Apache 2.0 — download Linum v2, no render queue.
If you need a text-to-video model you can inspect, fine-tune, and ship commercially without a per-render meter, Linum v2 is one of the few real options at 2B params and Apache 2.0. It is a research artifact rather than a product, so treat v2 as a base to build on, not something to deploy and forget. Teams that want a hosted pipeline with support should look at Runway or Pika instead.
Verified 4d ago · liveness 63/100 · cite: rightaichoice.com/tools/linum
- ML engineers who want a fine-tunable text-to-video base model
- Research groups needing a permissively licensed model to publish with
- Creative technologists embedding video generation into custom pipelines
- Startups that prefer self-hosting over per-clip hosted APIs
- Non-technical creators who want to type a prompt and download a finished clip
- Teams that need a hosted API, SLA, or commercial support contract
- Anyone without GPU hardware capable of running a 2B-parameter video model
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Linum if you want to type a prompt into a browser and download a finished clip — v2 is weights you self-host, so you bring the GPUs, the inference pipeline, and the serving layer yourself.
Downloading the weights is free, but running a 2B-parameter video model means paying for GPU compute — either owned hardware or a rented instance — for every clip you generate.
Linum v2 is free under Apache 2.0 — you pay for GPUs, not for the model. That makes it dramatically cheaper than per-clip hosted video generators at any real volume, and dramatically more expensive in engineering time if nobody on your team can run a 2B model. For a research group or a startup with existing GPU capacity, that trade is clearly favorable; for a marketing team with no ML staff, it is not a saving at all.
In short
Linum — Linum trains text-to-video models from scratch and ships the weights under Apache 2.0 — download Linum v2, no render queue. Best for ML engineers who want a fine-tunable text-to-video base model, Research groups needing a permissively licensed model to publish with, Creative technologists embedding video generation into custom pipelines. Free to use.
What's new in Linum
Checked 4 days agoAcross the latest 3 updates: 3 news mentions.
Road to Linum v3 · Issue 02 — Training Text-to-Image Models 3.6× Faster
Field Notes issue arguing that pixel-space encoder-decoders beat latent diffusion models, reporting 3.6× faster text-to-image training.
Data Filtering for Generative Video Pre-training
Field Notes Issue 01 on how data filtering helps generative video models learn better and faster during pre-training.
Building a $8/Month GPU-Cluster Monitor
Tools write-up on the cheap monitoring setup the lab used to keep a training cluster running over the holidays.
What people actually say about Linum — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
40 mentions across 4 sources (Hacker News, YouTube, Product Hunt, Lemmy) · researched Aug 27, 2026.
Average across the 4 sources that answered — each source counts once, not each post.
- +Apache 2.0 license ensures permissive use and modification freedom.
- +Self-hostable and fine-tunable, giving full control to engineers.
- +2B parameter model is sizeable yet still runnable on consumer GPUs.
- +Strong transparency with Field Notes newsletters and detailed writeups.
- +Founders actively engage on Hacker News and answer technical questions.
- −Requires substantial GPU VRAM (16GB+) for smooth operation.
- −No user-friendly interface; setup demands technical expertise.
- −Output resolution limited to 720p, insufficient for high-end production.
- −Quality inconsistencies reported, with some 'bad' sample generations.
- −Model is young and evolving; may have rough edges in stability.
- • Requires significant GPU hardware investment (16GB+ VRAM recommended).
- • Time and expertise needed for setup and fine-tuning.
Viability Score
How well maintained and how widely used is Linum? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Text-to-video generation from natural-language prompts
- Linum v2 model with 2 billion parameters
- Apache 2.0 open-source license for commercial use
- Model weights downloadable from the vendor site
- Self-hosting on your own GPU infrastructure
- Fine-tuning on custom video data
- Consistent subjects and characters maintained across a shot
- Prompt-directed camera framing within generated clips
- Prompt-directed motion control
- Trained on Crusoe GPU infrastructure
- Field Notes research newsletter with dated issues
- Road to Linum v3 research series covering the next model
- Published data-filtering methodology for video pre-training
- Pixel-space encoder-decoder research for faster text-to-image training
About Linum
Linum is a two-person research lab that trains generative video models from scratch and publishes the weights instead of selling a hosted editor. Its current release is Linum v2: a 2-billion-parameter text-to-video model distributed under Apache 2.0. You download the weights, run them on your own GPU hardware, and there is no render queue, no credit balance, and no UI in the loop. That makes it a base model for people who want to adapt or embed video generation, not a service that hands you a finished clip. The model generates short clips from natural-language prompts, and the demo prompts on linum.ai show the intended register: long, specific scene descriptions that hold subject, lighting, and camera framing across a shot rather than drifting. A shaggy baby goat strums a cherry-red guitar; honey pours over a stack of pancakes; a panda chews bamboo. Prompt-directed motion, character consistency across the shot, and deliberate camera framing are the product surface. The lab documents its own work through Field Notes, its research newsletter. Recent issues include "Data Filtering for Generative Video Pre-training" (Aug 26, 2026), "Training Text-to-Image Models 3.6× Faster" using pixel-space encoder-decoders instead of latent diffusion (Sep 16, 2026), and "Pyramid-JiT: Low-Res Drafts, High-Res Images" (Oct 6, 2026). A Crusoe partner post from Jun 22, 2026 covers the infrastructure Linum trained v2 on, and the running "Road to Linum v3" series makes clear the lab is iterating past v2. This puts Linum at the opposite end of the market from hosted generators like Runway or Pika. You get weights, a permissive license, fine-tuning latitude, and training write-ups; you give up the UI, the support contract, and any expectation of plug-and-play. It fits ML engineers, research groups, and creative technologists who want a base model to adapt or embed.
Behind the Verdict
We'd reach for Linum when the deliverable is a model, not a clip. Two people training video models from scratch and publishing the weights is rare, and the Apache 2.0 license is the part that actually matters: you can fine-tune v2 on your own footage, embed it in a pipeline, and ship without a per-render meter or a vendor account in the middle. The research surface is real too. Field Notes issues are dated and specific — data filtering for video pre-training, pixel-space encoder-decoders beating latent diffusion at 3.6× faster text-to-image training, Pyramid-JiT multiresolution prediction — and they tell you how the team thinks, which matters if you plan to build on their roadmap. The catch is that v2 is a checkpoint, not a product. You bring the GPUs, the serving stack, and the evaluation. There is no editor, no managed API, and no support contract. "Road to Linum v3" is openly in progress, so anyone signing a production timeline against v2 is building on a moving target — plan for version drift and keep your own evals. Compared to hosted options like Runway or Pika, the trade is control for convenience. They give you a UI, a queue, and someone to call; Linum gives you weights and latitude. If your team has the ML bench to fine-tune a 2B video model, that swap is worth it. If not, it isn't. Where it bites: prompt adherence and motion quality still depend on your own tuning, and the demo prompts on the site are long and specific by design — short lazy prompts will not get you the same results. Budget real engineering time before you assume you can beat a hosted tool on convenience.
Researching Linum? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Linum actually fits — and what changes day-one when you adopt it.
Pull the Linum v2 weights from linum.ai, stand up an inference script on a rented GPU instance, and wire it into an internal storyboard tool that turns product copy into rough animated clips.
Outcome: A working text-to-video path inside your own pipeline, with no per-clip billing and full control over the prompts and output resolution.
Reproduce the results in Field Notes Issue 01 on data filtering for generative video pre-training, then fine-tune v2 on a domain-specific clip dataset for a paper.
Outcome: An Apache 2.0-licensed base model you can cite and publish against, plus a documented pre-training methodology to compare your own filtering experiments to.
Fine-tune Linum v2 on a visual style the client owns, then run generation offline on local hardware inside the venue where no hosted API would be acceptable.
Outcome: A self-contained generative video setup that runs without an internet dependency or a vendor account, using footage the client actually owns.
Use Cases
- Generate short animated clips from text descriptions for prototyping or storyboarding
- Fine-tune the model on custom footage for a specific visual style or subject
- Embed text-to-video generation into a self-hosted production pipeline
- Research controllable motion and camera framing in generative video
- Inspect and modify the training and inference code for publication or coursework
- Own inference costs on GPU hardware instead of paying per generated clip
Models Under the Hood
as of 2026-10-17
Limitations
- Linum v2 is a 2-billion-parameter text-to-video model released under Apache 2.0, with weights available for download from the vendor site.
- The evidence describes a research-stage, self-hostable model from a two-person lab, with the next version (v3) still on the publicly documented "Road to Linum v3" series — so expect a moving target rather than a frozen release.
- Practical use requires downloading and running the weights on your own GPU infrastructure, and you own inference, serving, and failure handling.
- If you cannot run a 2B video model yourself, this model is not usable by you today.
as of 2026-10-04
Verification history
We have re-verified Linum 10 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 10 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Linum tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Free
$0
Ideal for
ML engineers, research groups, and creative technologists with their own GPU capacity who want to fine-tune or embed a text-to-video base model.
What this tier adds
Starting tier — the Linum v2 weights themselves, 2B params under Apache 2.0, free to download, self-host, fine-tune, and use commercially.
Where the pricing makes sense
The company stage and team size where Linum's pricing actually pencils out — and where peers do it cheaper.
Linum v2 is free under Apache 2.0 — you pay for GPUs, not for the model. That makes it dramatically cheaper than per-clip hosted video generators at any real volume, and dramatically more expensive in engineering time if nobody on your team can run a 2B model. For a research group or a startup with existing GPU capacity, that trade is clearly favorable; for a marketing team with no ML staff, it is not a saving at all.
Setup time & first value
How long it actually takes to get something useful out of Linum — broken out by persona, not the marketing-page minute.
ML engineer with an existing GPU box and PyTorch environment: an afternoon to download weights and get a first clip out. Research group reproducing the Field Notes pre-training work: days, mostly spent assembling and filtering a video dataset. Creative technologist with no prior diffusion experience: a week or more before output is usable, since you are building the inference and prompt-handling
Switching to or from Linum
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From Runway or Pika: move from per-clip hosted generation to downloading Linum v2 weights and re-building your prompt-to-clip flow as a local inference script.
- →From another open-source video model: swap in the Linum v2 weights, keep your existing inference harness, and re-check output quality against your own prompt set.
- →From a manual storyboard workflow: generate rough animated clips with v2 for pre-viz, then hand the selected shots to your existing edit pipeline.
- ↗To Runway or Pika: if you need a hosted editor, managed rendering, and a support relationship, move generation to a hosted provider and keep Linum only for offline or fine-tuned experiments.
- ↗To another Apache-licensed video base model: if v3's release cadence outruns your ability to re-test, port your fine-tuning dataset to a model with a settled version history.
- ↗To a commercial video API: when your team no longer wants to own GPU serving, replace local inference with an API call and treat Linum as a research reference.
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Linum”, and we withheld 6: 6 could not be judged, because “Linum” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Linum.
Official links
Tools that pair well with Linum
Common stack mates teams adopt alongside Linum, with the specific reason each pairing earns its keep.
PixelMotion
Turn product photos into professional marketing videos with 57 AI models, from $29/mo.
Anakin.ai
Anakin.ai is a no-code AI platform for text, image, video and voice generation, workflows and batch automation.
Writingmate
Writingmate bundles 350+ chat, image, and video AI models behind one $19.99/month credit pool.
Featured Head-to-Head Comparisons
Linum vs Splice
Choose Splice if you're a music producer needing a massive library of royalty-free samples and rent-to-own plugins like Serum 2—it's a no-brainer for under $5/month. Pick Linum only if you're a technically skilled creator or researcher who wants to experiment with open-source text-to-video generation and has GPU horsepower to run it. These tools serve completely different purposes; there's no overlap.
Linum vs Storyfile
If you're a technical creator aiming to generate or fine-tune synthetic animation, Linum's free open-source model with controlled camera and character motion is unmatched. If you need a ready-to-deploy interactive exhibit featuring a real person's authentic responses—like the museums and CNN have done—StoryFile's recorded-video conversational AI is the only credible choice. They serve fundamentally different needs: synthetic originality vs. human authenticity.
Linum vs Landr Mastering
Choose Lanum if you're a machine learning engineer or technical animator needing open-source, customizable text-to-video generation with character consistency. Choose LANDR Mastering if you're a musician or producer seeking instant, professional AI mastering without technical complexity.
Alternatives to Linum
View all
PixelMotion
Turn product photos into professional marketing videos with 57 AI models, from $29/mo.
Anakin.ai
Anakin.ai is a no-code AI platform for text, image, video and voice generation, workflows and batch automation.
Writingmate
Writingmate bundles 350+ chat, image, and video AI models behind one $19.99/month credit pool.
Frequently Asked Questions
Used Linum? Help shape our editorial sentiment research.