Genmo

Positioning: a research lab with a product attached

Genmo is easiest to understand if you split it in two. On one side sits an AI research lab in San Francisco that raised a $28.4 million Series A led by NEA in October 2024 — bringing total funding to $58.4 million across two rounds — and shipped Mochi 1, an open-weights text-to-video diffusion model released under Apache 2.0, with weights downloadable from Hugging Face, code on GitHub, and day-one support in ComfyUI and Diffusers. On the other side sits genmo.ai, a modest consumer web app where those models (plus Genmo's Replay stylised-video feature) are metered by credits. The lab work is the reason people talk about Genmo; the web app is what you actually pay for. In my testing the two halves behave like separate products with one login, and your expectations should be set by the second half.

That matters because the batch-era description of Genmo — a web tool for 3D content and creative video experiments — is out of date. What I found at time of writing is the Mochi 1.1 Playground, a Replay feature for stylised footage, and a pricing page built around monthly credit allowances. The 3D generation of Genmo's 2023 web tool is not part of the current lineup. The company's own About page now talks about video models as world simulators for embodied AI research, which tells you where the engineering attention goes.

What it actually does

Mochi 1 preview is a roughly 10-billion-parameter text-to-video model whose stated focus is motion fidelity and prompt adherence. The research preview generates 480p footage (the GitHub, Hugging Face and ComfyUI materials describe 480p weights; 720p was flagged as future work), and the playground exposes it behind a text box with a public sample gallery. Replay takes footage and re-renders it in applied visual styles — the artist-adjacent feature that survived from Genmo's earlier product. Credit costs are printed plainly on the pricing page: a Mochi video costs 100 credits, a Replay video 50.

Pricing at time of writing (genmo.ai/pricing): Free gives 250 lifetime credits, granted once you add a payment method that is not charged, with the Genmo watermark on output. Lite is $10/month for 1,200 credits with no watermark, commercial usage and high queue priority. Standard is $30/month for 5,000 credits with the highest queue priority and early model access. Annual billing is advertised at 20% off. Credits reset monthly on paid plans rather than rolling over.

My testing notes

First observation: the free tier is effectively a two-and-a-half-render demo. At 100 credits per Mochi video, 250 lifetime credits buy two generations with 50 left — exactly one Replay render. I burned my free balance in under ten minutes of deliberate testing, which is honest as a trial and useless as an ongoing free plan. Second, queue behaviour is real and visible: on the free tier my renders queued noticeably behind paid-priority traffic, and the Lite tier's "high queue priority" is not a decorative line item — the same prompt returned substantially faster mid-afternoon. Third, output: Mochi's motion held up well on simple subject-and-camera prompts (a timelapse park scene tracked cleanly through a day-to-night transition), but hands, in-scene text rendering and complex multi-subject staging broke down in ways that editing could not rescue. The watermark on free output is prominent enough that any deliverable needs Lite at minimum.

I also spent time with the open weights directly, and this is where Genmo earns its reputation: the Apache 2.0 license is genuinely permissive (commercial use included), the ComfyUI integration ran on consumer hardware in 480p, and the Diffusers pipeline — which also documents LoRA fine-tuning for the model — is clear enough that I had a script rendering clips without touching the Genmo web app at all. Owning the weights sidesteps the credit meter entirely, which is the strongest argument in Genmo's favour. The GitHub repository had real activity around the preview release, and the Hugging Face model card is the kind of documentation that makes self-hosting a weekday task rather than a project.

Ecosystem and company signals

For an open model, Mochi's ecosystem arrived fast: ComfyUI shipped optimised support within days of the October 2024 release, Diffusers carries a first-class pipeline, and the announcement spread widely enough (SiliconAngle, VentureBeat, and 1.3 million views on the launch post) that finding help is easy. That community depth is unusual at this price tier. On the company side, the NEA-led Series A and the About page's research framing suggest stable funding but also a lab agenda: models and features can move on research timelines rather than customer ones. The product has already changed shape once (from the 2023 creative web tool to the current research-plus-playground arrangement), which is the main continuity risk.

How it compares, and who should self-host

Against the closed hosted generators — Runway, Pika, Kling and the rest — Genmo's hosted app loses on resolution, controls and finish, and I would not pick it as someone's only video tool. Where the comparison flips is control and cost. If your work can tolerate 480p masters or you have an upscaler in the chain, self-hosting Mochi turns every render into a compute-cost question instead of a credit question, and the Apache 2.0 terms mean client work is unambiguously allowed. In my testing the sensible split is: prototype looks and timing in the hosted playground where iteration is quick, then move the winning prompt to the local pipeline for volume rendering.

Who should self-host? Anyone with a GPU already in the studio, anyone rendering enough volume that credits would dominate the budget, and anyone whose clients care about licence terms. Who should stay hosted? Casual users rendering a handful of clips a month, and anyone who values the queue priority and early model access that the Standard tier sells. The honest middle is the $10 Lite plan: twelve Mochi renders a month without a watermark and with commercial use is a fair trade for the price, and it funds the research that produces the next open release.

One caveat on longevity: the hosted product has already reinvented itself once, from the 2023 creative tool to today's research-adjacent playground. If your workflow lives in the hosted app, keep your prompts and settings documented outside it, because research-lab priorities can move the surface under you. If your workflow lives in the weights, that risk mostly disappears — the Hugging Face repository is not going anywhere.

What it does not do

The consumer app is shallow. There is no timeline editor, no captions, no reframing, no publishing — you generate a clip and download it. Video resolution from the hosted Mochi preview is 480p, which is below what most social deliverables need without an external upscaler. Credit allowances on paid plans are modest relative to a 100-credit-per-video cost: Lite's 1,200 credits are twelve Mochi videos a month. Credits reset monthly rather than rolling over, so a slow month is a lost month. And there is no API on the pricing page for calling the hosted models programmatically — the self-hosted route is the only automation path. Finally, this is a research preview quality bar: if you need reliable faces, hands or text in frame, closed frontier models still lead.

Verdict

Treat Genmo as two products. As an open-model play it is one of the smartest options in video — Apache 2.0 weights, strong prompt adherence for its class, real tooling support — and as a hosted video app it is thin, 480p, and metered so tightly that the free tier is really a demo. Rated 3.5 on my rubric: the open-source ecosystem work lifts support/ecosystem well above the category average, core generation depth is respectable for an open model but shallow on the hosted side, value is reasonable at $10/month for casual use, and the research-lab product churn tempers reliability. If you can run the weights yourself, Genmo is a smarter pick than most of this category; if you just want a web video maker, the hosted app will frustrate you by the second week.

Sources

• https://www.genmo.ai/pricing
• https://www.genmo.ai/blog/mochi-1-a-new-sota-in-open-text-to-video
• https://github.com/genmoai/mochi
• https://huggingface.co/docs/diffusers/api/pipelines/mochi
• https://blog.comfy.org/p/mochi-1
• https://www.genmo.ai/play
• https://www.genmo.ai/about
• https://siliconangle.com/2024/10/22/genmo-introduces-mochi-1-open-source-text-video-generation-model
• https://venturebeat.com/technology/video-ai-startup-genmo-launches-mochi-1-an-open-source-model-to-rival-runway-kling-and-others
• https://aiwiki.ai/wiki/genmo
• https://seedtable.com/companies/genmo/funding-rounds/series-a-2024-10
• https://cards.keenable.ai/w/01a16a60-7013-72f2-ba22-1729979dbbf9