All articles
Create·8 min read·

The tools we actually use, and what each one is bad at

A working creator stack, written by the people running it. What each tool is for, where it sits in the week, and the thing about it that annoys us.

Most tool roundups are written by someone who opened a free trial, took a screenshot and moved on. This one is the stack we actually run Scrptly on, which means it is shorter than most lists and considerably less enthusiastic.

Every entry below has two parts: what we use it for, and the thing about it that is genuinely annoying. If we could not write the second part honestly, the tool did not make the page. That rule cut the list roughly in half.

Nobody paid for a slot here and there are no affiliate links in this article.

Working out what to make

Scrptly

Our own tool, so weigh it accordingly. It takes a Reel that already worked and pulls out the structure underneath: the opening line, the pacing, where the beats land. We built it because the useful question is never "what is trending", it is "why did this specific video hold people".

The annoying part: it describes what a video did, not whether the same thing will work for you. We analysed 4,385 Reels for our hook study and the strongest finding in the set, that short hooks beat long ones, reverses in two categories. A tool that tells you what worked somewhere else is a prompt to test, not an instruction.

Claude

Does more of the week here than anything else on this list. It writes and ships pages, runs the SEO system, keeps the research notes, and built the talking Halloween mirror described further down. Most of what we publish passes through it.

The annoying part, and this one is worth taking seriously: it will hand you a confident, complete-looking, wrong answer if you let it work from its own old notes instead of checking the live source. Ours quoted a dataset size of 3,056 videos for weeks, including in a public repository, because it kept reading a stale note. The real number was 4,385. Nothing in the output looked uncertain. The fix is boring and it is the whole trick: make it verify against the thing itself, not against what it remembers.

Making the thing

Higgsfield

Image and video generation. We have put 599 assets through it since February, 385 images and 212 videos. It is what we reach for when a shot needs to exist and filming it is not realistic, including the face for the magic mirror build.

The annoying part is the billing model, not the output. Credits expire, and they expire quietly. We lost 4,979 unused credits on 31 October without noticing until we went looking for something else. Video is also far more expensive than it looks next to images: roughly 14 credits for a five second 720p clip against 2 to 6 for a still, so a session of "just trying a few versions" costs more than you would guess. Check your expiry date before you top up, not after.

Canva

Thumbnails, carousels, anything static that needs to look finished without opening a real design tool. Its genuine strength is that a non-designer can produce something acceptable quickly, which is the correct bar for most content work.

The annoying part: the video side is much weaker than the static side. The timeline is fine for stitching clips and captions together and frustrating for anything that needs frame-level control. Use it for the thing it is good at and move the edit elsewhere.

Blender

Free, open source, and the tool to learn if you want 3D scenes, product shots or motion that does not look like a template. The fact that it costs nothing and is genuinely professional makes it unusual on any list like this.

The annoying part: the learning curve is the real price. The software is free and the weeks you spend before you can make anything usable are not. It is worth it if 3D is going to be part of what you make regularly, and it is a bad use of a fortnight if you need one shot by Friday.

Editing

CapCut

The mobile edit. Cutting, captions, pacing and sound, on the device the footage is already on, which removes a transfer step that otherwise quietly eats an afternoon.

The annoying part: the auto captions need checking every single time. They mistime on fast speech and they confidently mis-transcribe names and anything niche, and captions are the part of a Reel most people actually read. Budget time to fix them rather than trusting the first pass.

Voice and audio

ElevenLabs

Synthetic voice, and the realistic end of it. We used it to build a talking Snow White mirror for Halloween: a tablet behind two way glass that listened all evening and answered in character, with its mouth moving. That project is open source if you want it.

The annoying part is a specific one worth writing down, because it cost us most of a build day. Several defaults fail silently rather than erroring. A seven second silence timer makes a conversational agent start talking to an empty room. A ten minute session cap ends things mid-evening with no message. Background speech filtering is off by default, which is wrong for any room with more than one person in it. None of these announce themselves; the thing just behaves oddly and you go looking in the wrong place.

Getting your own work back

This is the part of a content workflow nobody plans for. You posted something two or three years ago, you have changed phones twice since, and the only copy that still exists anywhere is the one on the platform. The original is gone from your camera roll, your old laptop, and whichever cloud drive you were using at the time.

It comes up more than you would expect: recutting an old post that did well, putting a showreel together, or moving your own back catalogue onto a platform you have just started using. A downloader pulls the published version back at full quality, which is usually better than whatever compressed copy you still have.

Worth being plain about what these tools are, because the honest version of this recommendation has a second half. A downloader does not check whose video it is pointed at. It will pull down anyone's. We are talking about using it on your own back catalogue, and that distinction matters to anyone who has had their work reposted without credit. If you are going to use someone else's footage, that is a permissions conversation with them, not a software question.

What is deliberately not on this list

Scheduling tools, analytics dashboards and the various all-in-one creator suites. Not because they are bad, but because we either do not run them or have not run them long enough to tell you what breaks. A list is only useful if the absences mean something.

We also left out everything we have used for less than a month. The interesting part of any tool is the thing you only discover in week six, and we would just be repeating the marketing copy.

The short version

The stack, and the catch in each one
ToolWhat we use it forThe catch
ScrptlyTaking apart Reels that already workedTells you what worked, not what will work for you
ClaudeWriting, shipping, research, most of the weekConfidently wrong when it works from stale notes
HiggsfieldGenerated images and videoCredits expire quietly, video costs far more than stills
CanvaThumbnails, carousels, static designVideo timeline is much weaker than the static side
Blender3D scenes and motion, freeThe learning curve is the real price
CapCutThe mobile edit, captions, pacingAuto captions need checking every time
ElevenLabsSynthetic voice that holds upSeveral defaults fail silently rather than erroring

If one thing on this list is new to you and saves you an afternoon, it did its job. If you already run all seven, the expiry warning on Higgsfield is probably the useful line.

Scrptly takes any Reel apart frame by frame: the hook, the pacing, the structure underneath. Point it at something that worked and see what it was actually doing.

Analyse a Reel free