Image generation is the most commoditised corner of AI, which makes tool-chasing pointless and job-matching valuable. Four jobs cover business use, each with a different right answer, and the legal questions matter more than the picks.
The four jobs at a glance:
| Job | Right tool | Why |
|---|---|---|
| Quick utility images | ChatGPT / Gemini built-in | Already in the tool you pay for |
| Crafted visuals | Midjourney | The wide-use quality benchmark |
| Licence-comfortable work | Adobe Firefly | Trained on licensed stock, enterprise indemnity |
| Volume / in-house | Flux, Stable Diffusion lineage | No per-image fees, full control |
Quick utility images
Diagrams, illustrations for a post, a mock hero image to brief a designer: this job belongs to whatever you already pay for. ChatGPT and Gemini both generate capable images inline, iterate conversationally, and handle text-in-image far better than the early years. The convenience beats specialist quality for anything internal or ephemeral, and the marginal cost is zero.
Crafted visuals
Where the image is the product, campaign art, distinctive site imagery, anything a designer would once have comped, Midjourney remains the wide-use benchmark for sheer visual quality and controllable style, with a real learning curve in exchange. The craft did not vanish; it moved into prompting, curation and iteration, and a saved house-style recipe is what keeps outputs looking like one brand rather than a mood board.
Licence-comfortable commercial work
For organisations whose legal teams ask where training data came from, Adobe Firefly is the category answer: trained on Adobe’s licensed stock and built into the Creative Cloud tools designers already use, with indemnification for enterprise customers. The images are competent rather than category-leading; the paperwork is the product, and for some buyers that is exactly right.
Volume and in-house generation
At scale, or where prompts and outputs must stay on your infrastructure, the open-weight image families, Flux and the Stable Diffusion lineage foremost, run on your own hardware with full control and no per-image fees. It is the image sibling of the open-weight LLM decision: real control, real upkeep, justified by volume or data posture rather than fashion.
The cautions that actually bite
Three, in descending frequency:
- Resemblance risk. A tool’s commercial licence does not immunise an output that resembles a copyrighted character, artwork or a real person; avoid prompting toward any of them.
- Ownership softness. UK law on copyright in purely AI-generated images is unsettled, so build nothing that depends on exclusive rights to a raw generation; human editing and composition strengthen your position.
- Documentary dishonesty. Generated images presented as real photos of your team, premises, work or results sit on the wrong side of the same line as fabricated reviews, and platforms increasingly label synthetic media anyway.
Illustrate freely; never fake the record.
Where I fit in
Image tooling shows up across my work: site imagery inside Managed Websites builds, generated assets in content pipelines with the honesty rules enforced in the system rather than remembered, and self-hosted generation where volume justifies it. If your firm is spending real money on stock imagery or design hours for routine visuals, that line item is usually reducible in an afternoon; it is a small, satisfying corner of an Automation Audit.