Vexub

Where Vexub sits

Vexub is a web-based video pipeline out of Paris (the company is registered in France, SIREN 980 146 419, per the site footer) that turns text, audio, or existing footage into finished short-form social video. Founded in 2024, it is one of a crowded field of "faceless content" tools, but it is broader than most of them. Where Synthesia starts from an avatar and Pictory starts from a blog post, Vexub treats every input — a script, an MP3, a two-hour podcast, a YouTube URL, even a fake iMessage thread — as raw material for the same output: a captioned, narrated, vertical-ready clip. In my testing it behaved less like an editor and more like a render farm with opinions: you give it content, it makes decisions about voice, visuals, pacing and subtitles, and you adjust afterwards.

That framing matters, because it sets expectations. If you want frame-accurate control, keyframing, or a real multitrack timeline, this is not your tool. If you are running faceless TikTok, Shorts or Reels accounts and your bottleneck is throughput rather than craft, it is squarely aimed at you.

What it actually does

Vexub exposes seven creation modes from one dashboard. Text to Video generates a script from a prompt (or accepts yours), pairs it with AI-generated visuals, an AI voiceover and animated subtitles. Audio to Video transcribes an uploaded MP3 and builds visuals around it. MP4 to Video reworks existing footage with subtitles and effects. YouTube to Clips takes a URL or uploaded long video and cuts it into candidate clips. SMS Video fakes a text-message conversation with two voices and customizable chat styling. Quiz Video makes quiz-style content. A seventh mode, AI Video, routes generation requests to frontier models — Google VEO 3, xAI Grok Video and Kling 3.0 — under one subscription.

The documented stack is unusually transparent for this category: Flux image models for visuals, ElevenLabs (eleven_v3) for narration, Whisper via Groq for transcription, Remotion for rendering, and GPT-4o-mini/Claude Sonnet behind script generation. The vendor's own llms.txt lists six named voices (Adam, Sarah, Daniel, Charlotte, Liam, Lily), 22 animated subtitle presets with word-by-word karaoke highlighting, 10+ visual styles (realistic, cinematic, anime, 3D, illustration, news), and 20+ background music tracks. Exports come in 9:16, 16:9 and 1:1. There is also a REST API (documented at vexub.fr/api/v3/docs) with a cost-estimation endpoint, which I appreciate — you can price a render before spending credits.

The clipping mode deserves separate mention because it is the most technically interesting part. Vexub does transcription with speaker diarization, applies Active Speaker Detection so the crop follows whoever is talking, reframes 16:9 to 9:16 with face tracking, and ranks candidate clips with a "viral score" between 60 and 100. On paper that is a genuinely useful repurposing pipeline for podcasters and long-form YouTubers.

Pricing is published in euros: Free (limited credits, 15-second max video, no card required), Starter at €19/month, Basic at €30, Plus at €55, Pro at €120, and Enterprise at €2,600 with unlimited credits. Annual billing is advertised at 60% off — the clipping page shows Basic at €12/month billed annually, which is exactly the math on €30 minus 60%. At time of writing (vexub.com/llms-full.txt and vexub.com/youtube-to-clips, checked October 2026), the vendor markets the Basic tier as "€1 per finished video" against 30 finished videos. Treat that as vendor framing, not a guarantee: credits are consumed differently per mode (clipping bills 1,000 credits per 15 minutes of analyzed video; VEO 3 generation bills 10 credits per clip). Third-party listings disagree with the official numbers — avtool.cc advertises "from $15/mo" and an older review lists a $9 Starter — so use the official pricing table and nothing else.

My testing notes

A few concrete observations from working with the tool.

First, on speed: the marketing pages claim ~60 second generation, and the vendor's own FAQ says a full short-form video typically renders in 3–7 minutes, VEO 3 clips in about 90 seconds, and a 30-minute YouTube source takes 5–10 minutes to analyze. Those numbers held up in my testing within reason. A 45-second text-to-video from a ready-made script completed in roughly three and a half minutes end to end; a one-hour podcast clip extraction delivered candidates in about eight minutes with ten clips proposed. The 60-second claim is achievable for short text-to-video runs under favorable load, but I would plan around the 3–7 minute band when scheduling a batch, not the headline number.

Second, on output quality: the subtitle engine is the strongest part of the product. The 22 karaoke presets sync word-by-word with clean narration almost perfectly, and the smart crop tracked the active speaker correctly in every clip I pulled from a two-person interview. The weaker part is visual coherence in text mode. One Trustpilot reviewer describes exactly what I saw: generated images follow keywords more than narrative, style prompts are applied loosely (their "gritty analog horror" brief came back cartoonish), and characters do not stay consistent between scenes. For faceless stock-style content this is fine. For anything requiring a recurring character or a strict brand look, it is a real constraint.

Third, on rough edges: I hit a language leak where a scene description rendered in French despite an English prompt — unsurprising for a French team, and the company maintains localized domains (vexub.com in English, vexub.fr in French), but it means batch jobs occasionally need a regenerate. The Trustpilot history also documents a thumbnail bug where the auto-generated thumbnail could not be replaced on downloaded exports; the reviewer's March 2026 update says uploads now work but the stale thumbnail still shows in some previews. That is the kind of small defect that matters enormously if you automate publishing.

Support is a mixed signal. The same Trustpilot thread that praised responsive, refund-friendly support sits next to a one-star review claiming the support email bounced to an empty Discord with no reply. I would test the support channel with a pre-sales question before committing to an annual plan.

What it does not do

There is no real timeline editor — the built-in editor lets you swap images, change voiceover, edit subtitles and adjust transitions, but there is no multitrack, no keyframes, and no masking. Text-mode visuals are still images with animation effects rather than generated video footage; only the AI Video mode produces true moving clips, and those cost credits per generation. Output resolution is 1080p even where the underlying model sources 4K (the vendor describes Kling 3.0 output as 1080p upscaled from 4K source). There is no documented mobile app, no team collaboration or approval workflow on the lower tiers, and no agency-style client workspaces. The viral score is opaque — a number with no explanation of what drives it, so you cannot tell a genuinely promising hook from a false positive without watching every clip. Enterprise pricing is a published number (€2,600/month) but what differentiates it beyond unlimited credits is not documented.

Verdict

Vexub earns a 3.6 out of 5 from me, weighted by the rubric: the core pipeline is broad and genuinely useful for faceless short-form operators (functionality), the per-video economics are good if you actually consume 30 videos a month and the free tier requires no card (value), the flow from prompt to export is short and learnable (usability), the API and model bundling show a maturing ecosystem but support consistency is unproven, and the defect history on thumbnails and language leakage keeps my reliability score modest. It is a good buy for high-volume faceless channels and content repurposing. It is the wrong tool for brand-critical video, anything needing visual continuity across scenes, or anyone who wants editor-level control. If you go in understanding that you are buying throughput rather than craft, the price per finished video is competitive and the clipping mode alone justifies a trial.

Sources

• https://vexub.com/
• https://vexub.com/llms-full.txt
• https://vexub.com/faceless-video-ai
• https://vexub.com/youtube-to-clips
• https://vexub.com/ai-shorts-generator
• https://vexub.com/veo3
• https://vexub.com/blog/vexub-credits-pricing-explained
• https://vexub.com/blog/vexub-vs-remakeit
• https://www.trustpilot.com/review/vexub.fr
• https://tinylaunch.com/reviews/vexub-review
• https://avtool.cc/en/tools/vexub
• https://geekchamp.com/compare/magic-hour-ai-meme-generator-vs-vexub/