Module 3, Lesson 4: Specialist Tools — Image Generation, Code Assistants, Transcription, Search

Estimated reading time: 9 minutes

Beyond the Big Four

The general-purpose AI assistants we've covered — ChatGPT, Claude, Gemini, Copilot — can do a lot. But the AI tool landscape is much broader than those four. There's a growing ecosystem of specialist tools that do specific things very well.

This lesson covers the major categories of specialist AI tools: what they do, when to use them, and which ones are worth knowing about.

Image Generation

AI image generators create pictures from text descriptions. You type what you want to see, and the tool produces an image. The quality has improved dramatically since the early days — modern tools can create photorealistic images, illustrations, logos, and artwork.

Major tools

Midjourney — Widely considered the leader in image quality, particularly for artistic and photorealistic images. It runs through Discord (which takes some getting used to) and through a web interface. Paid only — plans start around $10 USD/month.

DALL-E 3 — OpenAI's image generator, built directly into ChatGPT. The advantage is convenience: you can generate images in the same conversation where you're working on text. Quality is good but generally a step behind Midjourney for artistic work.

Stable Diffusion — An open-source image generation model that you can run on your own computer (if you have the right hardware) or access through various online tools. Free to use locally, but requires technical knowledge to set up. Many online tools are built on top of Stable Diffusion.

Adobe Firefly — Adobe's AI image generator, integrated into Photoshop and other Adobe Creative Cloud tools. Its key advantage: it's trained only on licensed images, which makes it safer for commercial use. It's included with Creative Cloud subscriptions.

Leonardo.ai — A popular tool for generating game assets, illustrations, and concept art. It has a free tier and paid plans, and offers good control over the style and quality of outputs.

When to use image generation

Creating illustrations for presentations or documents
Social media graphics and content
Concept visualisation and brainstorming
Placeholder images and mockups
Creative exploration and art

Honest limitations

Copyright is still unclear. The legal status of AI-generated images varies by country and is evolving. Be cautious about using AI-generated images commercially without understanding the risks.
Details can be wrong. AI image tools still struggle with hands, text in images, specific counts of objects, and complex spatial relationships.
Bias is real. Image generators can reflect and amplify biases present in their training data, particularly around race, gender, and cultural representation.
They don't replace designers. These tools are excellent for drafts and exploration, but professional design work still benefits from human expertise.

Code Assistants

AI code assistants help people write, debug, and understand computer code. Even if you're not a developer, it's worth knowing these exist — they're changing how software gets built.

Major tools

GitHub Copilot — The dominant AI coding assistant. It works inside your code editor (like VS Code) and suggests code as you type, similar to autocomplete but much more powerful. It can write entire functions, explain existing code, and help debug errors. Costs $10 USD/month for individuals; free for students and open-source contributors.

Cursor — A code editor with AI built in from the ground up. Rather than being a plugin for an existing editor, Cursor was designed around AI-assisted coding. It's popular with developers who want AI deeply integrated into their workflow.

Claude and ChatGPT — Both general-purpose assistants are surprisingly capable at coding tasks. You can paste code into a conversation and ask for explanations, debugging help, or modifications. This is often the simplest approach for people who aren't professional developers.

When code assistants are relevant (even for non-developers)

Automating repetitive tasks with simple scripts
Understanding code that someone else wrote
Building simple tools or workflows
Learning to code with AI as your tutor

Transcription and Audio

AI transcription tools convert speech to text — meetings, interviews, lectures, podcasts. They've become remarkably accurate.

Major tools

Otter.ai — A popular transcription tool that integrates with Zoom, Google Meet, and Microsoft Teams. It can join your meetings automatically, transcribe in real time, and generate summaries. Free tier includes limited transcription minutes; paid plans start around $17 USD/month.

Whisper (by OpenAI) — An open-source speech recognition model that's exceptionally accurate across many languages and accents. You can use it through various apps and services built on top of it, or run it locally if you're technically inclined.

Microsoft Copilot in Teams — If your organisation uses Microsoft Teams with Copilot, meeting transcription and summarisation is built right in.

Google Meet transcription — Google's built-in meeting transcription, available on certain Google Workspace plans. Works well for English; other language support is growing.

Descript — A tool that combines transcription with audio and video editing. You can edit audio by editing the transcript text — delete a sentence from the transcript and it's removed from the audio. Particularly useful for podcast and video creators.

When to use transcription tools

Recording and summarising meetings
Transcribing interviews for research or journalism
Making audio and video content searchable
Accessibility — creating captions and subtitles
Note-taking during lectures or presentations

Honest limitations

Accents and technical terms. Accuracy drops with strong accents, heavy jargon, or poor audio quality. Always review transcripts for important content.
Speaker identification isn't perfect. Most tools can distinguish between speakers, but they sometimes get confused, especially with similar-sounding voices.
Privacy matters. Sending audio to a cloud service means your conversations are being processed externally. Check your tool's data policies, especially for sensitive discussions.

AI-Powered Search

Traditional search engines give you links. AI-powered search tools give you answers, synthesised from multiple sources.

Major tools

Perplexity AI — An AI search engine that answers questions with cited sources. It's like having a research assistant that reads multiple articles and gives you a summary with references. Free tier is generous; Pro costs $20 USD/month for more powerful model access.

Google AI Overviews — Google now shows AI-generated summaries at the top of many search results. You don't need to do anything special — they appear automatically. Quality varies; always check the linked sources.

ChatGPT with web browsing — When you ask ChatGPT a question that requires current information, it can search the web and synthesise results. Useful for quick research.

Microsoft Copilot search — Copilot's standalone chat includes web search, providing sourced answers to questions.

When to use AI search

Getting quick overviews of unfamiliar topics
Comparing information from multiple sources
Finding recent information that may not be in an AI's training data
Research where you want synthesised answers, not just links

Honest limitations

Citations need checking. AI search tools cite sources, but they can misrepresent what those sources actually say. Always click through to verify important claims.
They can miss nuance. A synthesised answer may oversimplify complex or contested topics.
Not a replacement for deep research. For academic, medical, legal, or financial research, AI search is a starting point, not an endpoint.

Key Takeaways

Specialist AI tools exist for image generation, coding, transcription, search, and more — each doing specific tasks better than general-purpose assistants.
You don't need all of them. Identify which categories are relevant to your work and explore those.
Free tiers and open-source options exist in almost every category.
Copyright, accuracy, and privacy concerns vary by tool category — understand the risks for your specific use case.
General-purpose tools can handle many specialist tasks adequately — you only need a specialist tool when you need higher quality or specific features.

Practical Exercise

Explore one specialist tool.

Choose one category that's most relevant to your work or interests:
- Image generation: Try Midjourney, DALL-E (via ChatGPT), or Leonardo.ai
- Transcription: Try Otter.ai or upload an audio file to a tool that uses Whisper
- AI search: Try Perplexity AI (perplexity.ai)
Complete a real task using the tool. For example:
- Generate an image for a presentation slide
- Transcribe a short meeting or voice memo
- Research a topic using Perplexity and compare with a normal Google search
Note down:
- How accurate/useful was the output?
- What surprised you (positively or negatively)?
- Would you use this tool again? Why or why not?

Knowledge Check

1. Which AI image generation tool is integrated directly into Adobe's Creative Cloud?

a) Midjourney
b) DALL-E
c) Adobe Firefly
d) Stable Diffusion

Answer: c) Adobe Firefly — it's built into Photoshop and other Adobe tools, and trained on licensed images.

2. What does Perplexity AI do differently from a traditional search engine?

a) It only searches academic papers
b) It synthesises information from multiple sources and provides answers with citations
c) It generates images based on search queries
d) It only works with paid subscriptions

Answer: b) It synthesises information from multiple sources and provides answers with citations, rather than just giving you a list of links.

3. What is a key limitation to be aware of when using AI transcription tools?

a) They can only transcribe English
b) They require specialised hardware
c) Accuracy drops with strong accents, technical jargon, or poor audio quality
d) They can only transcribe pre-recorded audio, not live speech

Answer: c) Accuracy drops with strong accents, technical jargon, or poor audio quality — always review transcripts for important content.