ugcs.farmBETA
Now in beta · free during launch

Remix viral reels. Ship ads for your brand.

Find trending Reels and TikToks, remix their proven hooks and pacing for your product, and export frame-by-frame prompts tuned for Sora, Veo 3, Wan, and Kling — with an AI director that plans, storyboards, and builds in chat.

No credit card. Free during open beta · Reels, TikToks, and your own footage

Talk to founders →
How it works

From viral reference to ready-to-ship UGC kit.

Six steps with an AI director in chat. Find what works on TikTok, adapt it to your product, approve, then build.

  1. Step 01

    Share your idea

    Describe your product or paste your website — we read the page and pull a brief so every step stays on-brand.

  2. Step 02

    Pick a pathway

    Remix a viral Reel or TikTok, build from scratch, or upload your own reference video. The AI director asks one choice at a time.

  3. Step 03

    Find proven formats

    Search Instagram and TikTok for viral hooks in your niche — tips, hot takes, POV, memes — with metrics and a recommended pick.

  4. Step 04

    Approve the storyboard

    Scene-by-scene plan you sign off on before we generate. Same beats as the viral reference, rewritten for your product.

  5. Step 05

    Frames + model prompts

    Per-scene starting frames and prompts tuned for Sora, Veo 3, Wan 2.1, and Kling — hook, pacing, and shot types carried over from the remix.

  6. Step 06

    Ship to any generator

    Export the kit with prompts and frames ready to paste. Open in Sora, Google Flow, Wan on fal, Kling — or multi-model hosts like Higgsfield and Replicate.

Tuned for every major model

One prompt won’t cut it. We retune for each.

Each video model has its own quirks — token limits, action grammar, anatomy weak spots. We bake in the right negative cues, beats, and continuity locks for every target.

S
Sora

OpenAI's flagship — long takes, photoreal subjects, deep prompt fidelity.

V
Veo 3

Google DeepMind's video model — excels at native audio, natural motion, and brand-safe outputs.

W
Wan 2.1

Open-source frontier — strong on stylised animation and rapid iteration.

K
Kling

Kuaishou's model — fast, cheap, and unbeatable on dynamic action shots.

Under the hood

Skills, critics, and a judge grade every render.

Generic prompt tools stop at ‘rewrite my idea.’ We analyze the viral reference, lock your product brief, storyboard for approval, then generate frames and per-model prompts — catching weak hooks, continuity gaps, and off-brand beats before you spend a render credit.

  1. Analyze
  2. Plan
  3. Skills
  4. Critic
  5. Polish
  6. Judge
Skills library13 product skills

Markdown spec sheets that teach the model how products actually work.

A flip-top skincare tube doesn't unscrew. A dropper bottle doesn't pump. We carry 13 (and counting) product-handling skills — each a structured spec covering motion, dispensing, anatomy, and the artefacts that specific form-factor invites.

tube_flip_topbottle_dropperbottle_pumpcompact_flipstick_twistapply_to_face

Skills stack: a face-wash tube matches both tube_flip_top (form-factor) AND apply_to_face (action). Their negative cues, beat recipes, and forbidden actions merge before the prompt is rendered.

Critic agentRuns every generation

A second pair of eyes flags weak beats and patches them before render.

After the planner builds the structured plan, Gemini reviews it against a rubric: missing continuity, weak motion grammar, anatomy risk, brand drift. It returns concrete patches — per-beat revisions, forbidden-action lists, continuity anchors — that the polisher applies in the same pass.

beat revisionsforbidden actionscontinuity anchorstalking points

Default-on. ~5s overhead. Never blocks generation — if the critic fails, the deterministic plan ships unchanged.

Judge agentGrades against the source video

Watches the actual clip and scores the rendered prompt.

Most prompt tools grade against their own analysis output — circular. Our judge sees the source video file directly (via Gemini's Files API) plus the rendered prompt and reference still, and scores them against a fixed rubric. Verdicts persist as success / partial / failed with severity-tagged findings.

beat_coveragesubject_fidelitymotion_realismanatomy_safety

Judge verdicts feed the memory loop — future generations bias toward beat shapes and skill combinations that scored well on similar products.

Where ChatGPT paraphrases from memory, we search viral reels, read your product brief, mirror the reference hook, and grade the kit against the source before you ship.

See the full pipeline →
Built for the platforms creators ship to

AI UGC for TikTok, Reels, and Shorts.

One pipeline, three vertical surfaces. We bake in the aspect, pacing, and motion conventions each platform's algorithm rewards — so the output drops straight into your scheduler.

TikTok9:16 · 15–60s

AI UGC for TikTok

Remix viral TikToks for your product. Search proven hooks, adapt beat-for-beat, and export 9:16 kits tuned for hook-up-front pacing.

See the TikTok guide →
Instagram Reels9:16 · up to 90s

AI UGC for Instagram Reels

Find top Reels in your niche and remix them for your brand. Same format, your script — ready for Meta Ads Manager or organic.

See the Instagram Reels guide →
YouTube Shorts9:16 · up to 60s

AI UGC for YouTube Shorts

Adapt viral short-form structures for YouTube Shorts. Storyboard approval, then per-scene prompts tuned for Shorts pacing.

See the YouTube Shorts guide →
Built for the way creators actually work

Skip the re-shoot. Remix ten hooks before lunch.

6 steps
From idea to export
4 models
Sora · Veo 3 · Wan · Kling
4+ hosts
Higgsfield · fal · AI Studio · Replicate
FAQ

Questions, answered.

How does remixing a viral reel work?
You share your product idea (or website), we search Instagram and TikTok for proven formats in your space, you pick one, and the AI director adapts that reel's hook, pacing, and shot structure for your brand — not a generic rewrite. You approve a storyboard, then we generate per-scene frames and model-tuned prompts for Sora, Veo 3, Wan, or Kling.
Can I use ugcs.farm to make TikTok ads?
Yes — remix is the fastest path. Find a TikTok or Reel that's already working in your niche, adapt it to your product in chat, and export a 9:16 kit ready for TikTok Ads Manager or organic posting. You can also build from scratch or upload your own reference video if you already have footage.
Does it work for Instagram Reels and YouTube Shorts?
Same remix pipeline, same output. We search both Instagram reels and TikToks, tune for vertical 9:16 by default (16:9 optional), and bake orientation locks into every model template. Organic posts and paid formats on Meta and YouTube are supported.
What's the best AI model for short-form UGC ads?
It depends on the spot. Sora is strongest for photoreal product hero shots and longer takes. Veo 3 has the best native audio and most natural human motion (best for talking-head UGC). Wan 2.1 is unbeatable on stylised / animated edits. Kling is the fastest and cheapest for high-volume A/B testing. ugcs.farm tunes a separate prompt per scene for each so you can ship the same remix kit across all four.
What kind of viral videos work best to remix?
Short-form hooks that already perform — tips, hot takes, POV, testimonials, course promos — in a broad domain related to your product. We keep the format and pacing; you swap in your message, product name, and visuals.
Do I need a paid Sora / Veo / Kling account?
Not to generate the prompts — those are free. To render the actual video you need access to the target model, but you have options. Every export includes a one-click deep-link to that model's native home (Sora on sora.chatgpt.com, Veo 3 in Google Flow, Wan on fal.ai, Kling on klingai.com) plus secondary links to multi-model playgrounds like Higgsfield, fal.ai, Google AI Studio, and Replicate. Use whichever host matches your existing workflow or budget.
Where can I render the generated prompts?
Anywhere that takes a text prompt and a first-frame image. The export step deep-links into Sora (sora.chatgpt.com), Veo 3 (Google Flow / Google AI Studio), Wan 2.1 (fal.ai), and Kling (klingai.com or fal). For teams that already pay for a multi-model host, we also link out to Higgsfield, fal.ai, Google AI Studio, and Replicate — so a Sora-flavoured prompt can run on Replicate, a Veo-flavoured prompt can run on AI Studio, and so on. We don't host video generation ourselves; the prompt and reference assets are yours to take wherever you like.
What's the difference vs. just asking ChatGPT for a prompt?
ChatGPT guesses from text. We run a directed pipeline: search viral references, analyze the reel's hook and direction, ground the script in your product brief, storyboard for approval, then generate frames and per-model prompts. A Critic patches weak beats; a Judge grades against the actual source video. You get a full kit, not a single paragraph.
How long are videos kept?
Source clips and generated assets are deleted automatically after 30 days. You can also delete a project at any time from the dashboard.
Is it free?
Yes — free during open beta. No credit card, no per-video charges. We'll introduce paid tiers alongside team accounts and higher-volume usage; everyone in the beta gets ample notice before anything changes.

Make UGC at the speed of thought.

Paste your idea or website. Pick a viral reel to remix. We’ll have a storyboard and model-tuned kit ready before your coffee gets cold.

or talk to the founders →