Alternative
Sandcastles is research-first: it finds outlier videos across other creators’ channels and helps you remix what worked. Melda is source-first: it starts from your own recordings and writes the idea you already have. Both end in a script. They disagree about where the script should come from.
Other creators’ top-performing videos, analyzed for the pattern that made them work.
The podcast, video, or call you already recorded, mined for the ideas only you can back up.
One tells you what shape is working. The other writes the thing only you could have said.
Fair description
Taken from their own public positioning, checked against their live site.
Sandcastles describes itself as the easiest way to research top channels, find outliers, and remix them into your own videos. The product wraps a research engine: you build a feed of channels across Instagram, TikTok, and YouTube Shorts, browse a database of outlier videos, pull the transcript from any of them, read an analysis of why a particular video performed — hook, structure, format, style — and then generate hooks and scripts on top of that research using frameworks drawn from high-view content. Projects hold the ideas you want to keep.
That research layer is real work and it is not something Melda attempts. Analyzing outliers across three platforms every day, keeping a competitor feed current, and surfacing the top performers in a niche is a substantial system, and if that is the part of your process that is broken, Sandcastles is aimed directly at it. Their plans are credit-based, with each analysis or script costing a credit, starting at $39 per month billed annually according to their own pricing FAQ, checked August 2026.
So this page is not an argument that research-first tooling is a mistake. It is an argument about what happens after the research, and about which creators have a different bottleneck entirely.
Fit
The clearest way to choose is to ask what you already have, not which feature list is longer.
The split is cleanest for people who already publish long-form. A coach with two hundred hours of recorded calls, a podcaster forty episodes deep, a founder with a YouTube back catalog — these creators are not short of things to say. They are short of a reliable way to get the best thirty seconds of any of it onto camera without it flattening into something anybody could have posted.
For someone starting cold, the calculation inverts. With no archive there is nothing to extract, and knowing which formats currently work is worth more than a voice-matching engine with no source to match against.
Side by side
Both pipelines produce a script. Everything upstream of that is different.
| Stage | Research-first (Sandcastles) | Source-first (Melda) |
|---|---|---|
| Where the idea comes from | A database of outlier videos from other creators’ channels. | Your own recording — the moment you already explained something well. |
| What gets analyzed | Why a high-performing video worked: hook, structure, format, style. | Your source material: which claims are specific, and what proof sits behind each. |
| What the script is built on | A storytelling framework taken from videos that performed. | One extracted idea, your take on it, and a hook matched to what the idea can deliver. |
| Whose voice it lands in | The pattern was learned elsewhere, so voice is yours to enforce afterward. | A voice profile built from how you actually talk, applied before you read the draft. |
| Needs source material? | No. That is the point — you can start with nothing of your own. | Yes. Without a recording or transcript there is nothing to extract. |
| Answers the question | What kind of video should I make? | What is the best thing I already said, and how do I say it in thirty seconds? |
Limits
Stated plainly, because a comparison page that hides its own gaps is not worth reading.
Melda takes the long-form you already own, transcribes it, and pulls out the specific ideas worth posting — each one saved with the quote or example that makes it credible. You add your take. A hook gets matched against patterns saved in your swipe library. Then it writes a short-form script in your voice, structured for talking-head delivery.
When the source is thin, it says so. Fewer ideas, sometimes none, instead of confident filler. That is a deliberate choice: a script you cannot defend on camera is worse than no script.
Together
Yes, and it is the setup we would actually recommend to a creator with budget for two tools.
Research tells you where attention already is. Source material tells you what you can say about it that nobody else can. Used together, the research narrows the topic and the format, and your own archive supplies the specific claim, the client story, the number you can actually stand behind. That combination is much harder to imitate than either half.
The failure mode to avoid is letting the pattern do all the work. If the framework came from someone else’s video and the content came from a general model, the result is a competent short that sounds like the fifty others built from the same template. The thing that makes it yours is the proof, and proof only comes from material you actually produced.
Next step
Research-first is one shape. Clipping tools and general assistants are the other two most creators weigh.
Drop it in, see which ideas come back with real proof attached, and decide from the draft rather than from a feature list. The first script is free and no card is required.
FAQ
Straight answers to the questions creators ask before they start.
For one specific job, yes. Both end with a short-form script. They disagree about where the script should come from. Sandcastles starts from other creators’ top-performing videos and helps you remix the pattern. Melda starts from your own long-form material and writes the idea you already have on tape. If outlier research across other people’s channels is what you want, Sandcastles does that and Melda does not.
Sandcastles describes itself as a way to research top channels, find outliers, and remix them into your own videos. Its site lists channel discovery across Instagram, TikTok, and YouTube Shorts, a database of outlier videos, transcript downloads, hook templates, analysis of why a video performed, and script writing on top of that research. Plans are credit-based and start at $39 per month billed annually, according to their own pricing FAQ, checked August 2026.
Melda has no outlier database, no competitor channel tracking, no cross-platform analytics, and no clip cutting. It will not tell you what is working on someone else’s account this week. It works from material you already own, which is a real limitation if you have none.
A research tool, honestly. Melda needs a source: a podcast episode, a video, a call recording, or a transcript. With nothing to draw from there is nothing for it to extract, and a research-first product is the more useful place to start.
Melda is built for that specifically. It works from your own words in your own recordings and keeps a voice profile so drafts read as things you would plausibly say out loud. Any pattern-first workflow carries a drift risk in the other direction, because the pattern was learned from someone else’s video.
Yes, and the two starting points complement each other. Research where the attention already is, then write the version only you can back up with your own material. One finds the shape; the other supplies the proof.
You can start free with no card required. Paid plans are $19, $49, and $99 per month, with 20% off when billed annually. Every feature is available on every plan; credits meter how much heavy generation you run.
Creators, coaches, and experts who already produce long-form material — podcasts, videos, client calls, workshops — and want short-form that carries their actual proof instead of a generic take on a popular format. Melda is built by Preston Chin, a creator with more than 10 million views in three months, and Jimmy Zhang.
