Choosing the tool for the job

Hi, I'm Mel. There are more AI creative tools than there are hours to test them, and they all claim to do everything. So I'm going to make one thing: a field guide page for a frog that does not exist. The frog is only the thread. Watch where each tool enters the job, where it should leave, and what still needs a human decision.

One split sorts most of the field. Creating brings something into existence that was not there: a new photograph, an unrecorded voice, a scene that was never filmed. Finishing arranges what now exists: type, margins, timing, colour and the crop that fits. A generator can invent the frog, but it does not know your layout. A design tool can finish the page, but it cannot photograph an imaginary animal.

General assistants are the broad starting point. ChatGPT combines writing, analysis, images and data in one place. Claude suits long work, complex thinking, files and tools. Gemini connects closely to Google's image and video tools. These are starting points, not winners. Pick the assistant that fits most of your ordinary work, then move to a specialist when the output asks for one.

For pictures, you can stay inside a general assistant — ChatGPT or Gemini — or use a dedicated generator. Midjourney centres the look of the image. Ideogram is built around typography. Adobe Firefly is positioned for commercially safer production. FLUX and Recraft are designer-facing alternatives. I am not ranking them. Use one long enough to learn how it responds, and change tools when the job changes, not because another launch appeared in your feed.

This prompt describes the photograph as well as the frog: side view, eye level, a macro lens, shallow depth of field, cool shadows and a sharp eye. Then it adds one memorable invention — a translucent throat glowing faintly amber. Lens, light and framing give the generator a picture to make. The invented detail gives the picture a reason to be remembered.

The first frog was close, but the legs were too plain. An edit added orange markings, and it also changed the front legs more than I wanted. So I corrected that in the same conversation: remove the orange from the front leg, make the bands more natural, and keep the toes orange. That back-and-forth is normal. Stay with the same image and correct whichever detail moved. Starting again would solve one detail by replacing the whole frog.

The page also needs a population chart. At this point I stop asking for appearance and establish the source. Six years, six values, typed into a table: twelve thousand in 1975, falling to sixteen hundred in 2025. These figures are fictional because the species is fictional, but they are now fixed. Whatever comes next has to use these numbers.

Then I asked the image generator for an infographic about that decline. Look at the result. It is beautiful, legible and confident. It also invents a different scientific name, a percentage scale, habitat history, pesticides, disease, local extinctions and a possible extinction. None of that came from my table or prompt. Image generators used to warn us with broken text. Better typography removed the warning while the risk stayed. A polished figure can still be fiction wearing the clothes of evidence.

So I rebuilt the chart in Canva from the table. It is less dramatic, and that is fine. The years match. The values match. If I change a number, the line changes with it. Use AI to help decide what a chart should communicate, but draw the chart somewhere that understands data. The source should control the picture.

These three images look like stages of one task, but they do different jobs. The table is the source. The infographic is an illustration invented from a sentence. The chart is a visual expression of the source. Only the one on the right can carry the numbers into a meeting, a report or a decision. Good design cannot rescue a broken chain of evidence.

Now the finishing. The photograph, labels, range note, chart, headline and spacing have to sit on one page in one visual system. This is Canva, though Figma, Affinity or PowerPoint could do the same job. The dissolve shows the difference: the ingredients existed on the first frame; the finished communication exists on the second. That arrangement was never going to fall out of the image generator.

This is the page I would show another person. The invented species is clearly presented as a designed example, the anatomy labels point to what they describe, and the chart comes from the table. Finishing turns separate outputs into one understandable thing, and makes inconsistencies visible enough to fix.

The same still can become a short video. I loaded the finished frog into Google Flow and asked for restrained motion: slow breathing, a gentle pulse in the throat, one firefly and a camera that stays still. That instruction protects what already works. Starting video from a strong frame gives the motion tool composition, lighting and character consistency before it has to invent movement.

The result is only eight seconds, but it is enough. The frog breathes, the firefly moves and the camera holds. Flow, Runway, Kling and Luma all work in this image-to-video space. Choose on whether the tool preserves the still, follows the motion brief and gives you a clip you can cut into the finished piece.

One more choice happens before production, not after it. Can you use the output commercially? Do you own it? If you pay later, does that fix work you already made? Different vendors answer those questions differently. Some free-plan work remains non-commercial even after you subscribe, and ownership can depend on whether the output includes licensed library material. Free plans are good laboratories, but if you intend to publish or sell the result, read the current terms before you generate it. This is a workflow check, not legal advice.

So choose by asking four questions. What am I creating? What tool finishes it? What evidence has to survive the process? And what am I allowed to do with the result? Pick something small this week and take it all the way through. You will learn more by finishing one imperfect piece than by collecting another list of tools.