I have a lot of AI tools installed. I use maybe four of them regularly. The gap between what I thought I would reach for and what I actually reach for has turned out to be one of the more useful things I have learned this year.
This is an honest account of what is in my stack, what each tool is actually like to work with, and why some things that looked essential turned out to be occasional experiments.
The ones I use every week
Kling AI (with Omni)
Kling is my primary video generation tool and has been for a while. The Omni model specifically changed how I work. It handles character consistency better than anything else I have tested, responds well to emotional and psychological prompting, and produces results that feel considered rather than generated.
Its personality: methodical. It takes direction well. If you put the work into your prompt, Kling rewards that. If you do not, it gives you something technically fine and completely forgettable.
The limitation I keep running into is precise character interaction. Two characters in the same frame doing something specific is still unreliable. But for single character work, atmosphere and cinematic pacing, it is the most consistent tool I have.
Midjourney
Midjourney is where I go first for character development and key scene composition. It has an aesthetic sensibility that the other image tools do not quite match. Results feel art directed rather than generated.
If I were to pretend that it was a person, I was say that it's very opinonated. It has 'taste' and it will express that taste whether you want it to or not. This is a strength when you are exploring more broadly and a frustration when you have something specific in mind that it keeps reinterpreting.
I use it to establish what a character looks like before I take them anywhere near a video tool. The character base portraits come out of Midjourney first.
Grok (Aurora)
Aurora is Grok's image generation model and it has become genuinely useful for reference sheet work specifically. Where Midjourney has taste, Aurora has compliance. It follows instructions more literally, which is exactly what you want when building a structured character reference sheet that needs to be technically consistent across multiple angles.
Its personality is more helpful and direct. Less creative flair than Midjourney but more controllable. I typically bring my protraits or base scenery from Midjourney into Grok to expand upon and create new or additional references of the same base image input.
ElevenLabs
The tool that surprised me most. I came in expecting a fairly mechanical voice synthesis experience and found something with genuine nuance. The ability to direct performance through prompt language rather than just selecting a voice type changed how I approach sound design. I say 'sound design' very loosely here as it's definitely my weak area.
Its personality is like a very good actor who reads the brief carefully. Give it emotional context and it responds to that context. Brief it like a spec sheet and it sounds like one.
I use it for character voices, sound design and music. I find music generation briefing quite tricky, evem with the help of Claude or Chat GPT to help structure the input prompt, so I typically go with a feeling rather than being too instructive on what exactly I want. Alternatively I'll describe what is happening and leave Elevenlabs to do the interpretation.
Claude
Writing, prompt development and workflow thinking. I use Claude as a thinking partner more than a writing tool. Working through how to structure a complex prompt, figuring out why something is not working, planning a shoot before I run it.
It also built this website, which feels worth mentioning.
Figma Weave
Node-based workflow design. I completed the Lighthouse AI Academy course on Figma Weave earlier this year and it changed how I think about AI production pipelines. The ability to chain tools together visually and see the whole workflow as a system rather than a sequence of separate steps is genuinely useful.
I do not use it every week but when I am planning a more complex production it is the first place I go to map out the pipeline before I start.
DaVinci Resolve
Editing. Not an AI tool but it is where everything comes together. The AI-generated clips are only as good as the edit that holds them together.
The ones I keep for experimentation
Higgsfield
I have a year's subscription to Higgsfield and I want to be honest about how I use it. It is not a regular part of my workflow. I pick it up when I want to test a new model or try something I cannot do elsewhere.
The most interesting thing it has produced recently was not a video. Higgsfield released HiggsCraft, a Minecraft mod that brings AI image, video and structure generation into Minecraft Java Edition. I spent a session getting it installed and running, working through NeoForge compatibility errors and version mismatches, just to generate AI structures inside a Minecraft world.
It worked. It was completely impractical. It was also exactly the kind of thing I think is worth doing. You find out what a tool actually is by taking it somewhere unexpected.
The ones I tried and moved on from
The ones I use occasionally
Runway, Sora, Krea
I tested all three. None of them became part of my regular workflow. That is not a verdict on the tools themselves but on fit. Runway has impressive capabilities but Kling gives me more consistent results for the specific work I am doing. Sora and Krea similarly did not replace anything I was already doing better elsewhere.
A note on Google Veo and Omni
I have tested both for emotional and psychological character work. In my experience they handle this less well than Kling or Seedance. The outputs tend toward literal interpretation of language rather than responding to the emotional context underneath it.
This might change. These tools are evolving fast enough that anything I write here could be outdated by the time you read it. But as of right now, for the kind of character-driven work I care about, they are not where I start.
What I actually learned from all of this
The tools that stayed in my regular stack are the ones with a discernible personality. Kling is methodical. Midjourney is opinionated. ElevenLabs is a good actor. Aurora is compliant. Claude is a thinking partner.
The ones I moved away from were not necessarily worse. They just did not have a clear personality in the context of my specific work. And it turns out that is how I choose tools. Not by feature list but by whether working with them feels like a collaboration or a transaction.