Code, Images and Music: How I Integrate AI into My Projects

I have used Claude, ChatGPT, Gemini and Suno for very different tasks: reviewing code, solving errors, writing documentation, building visual prompts, editing images and exploring musical demos. The key has not been finding one tool that does everything, but learning how to direct each process and review the output critically.

01 Why I use more than one tool

No model responds to the same problem in exactly the same way. Sometimes I need to analyse a long HTML file; at other times I need to debug an issue from a screenshot; in another stage I may want to visualise a poster or quickly hear a musical direction. Using several tools allows me to compare proposals and select the one that best fits each stage.

02 Real uses inside my projects

On my websites, for example, AI has helped accelerate changes ranging from a language selector to a chatbot integration. However, a modification that looked correct has sometimes broken paths, animations or external resources. That is where the less visible part of the work begins: roll back, compare versions, isolate the issue and test again.

03 A prompt works like a brief

A useful prompt is not simply “make something professional.” For code, I need to explain the structure, expected behaviour and constraints. For images, I define composition, palette, lighting and what must not change. For music, I describe structure, dynamics, instrumentation and the purpose of the demo.

Code example

Review this website without changing its visual identity.

Goal:
Add an English version with an ES/EN selector.

Constraints:
- Preserve animations, Spotify and the chatbot.
- Do not expose private keys.
- Use paths that work on hosting and in local tests.
- State exactly which files were modified.

Before editing, identify dependencies and possible side effects.

Image example

Create a vertical 4:5 poster for a progressive rock and post-hardcore band.

Aesthetic:
Atmospheric, geometric, dark and contemporary.

Elements:
Volcanoes, fragmentation and red gradients over black.

Avoid:
Power-metal aesthetics, excessive effects, generic typography and a template-like appearance.

Final use:
Instagram publication, with clear space for date and venue.

Musical demo example

Create a progressive rock and post-hardcore demo.

Structure:
Atmospheric intro, restrained verse, expansive chorus,
irregular instrumental bridge and an intense ending.

Instrumentation:
Ambient guitars, defined bass, dynamic drums and subtle synthesisers.

Mood:
Melancholic, tense and contemplative.

Avoid:
Theatrical vocals, unnecessary solos and overly bright production.

Goal:
Use the result as a reference for developing an original composition later.

04 Image generation is an iterative process

With images, I rarely keep the first version. I have had to ask for more realistic results, remove objects, preserve specific elements from a photograph, correct proportions, adjust lighting and change styles that drifted away from the original goal.

• First, I generate a proposal to test composition and mood.
• Then, I identify what should remain and what should change.
• I write a narrower instruction to avoid unnecessary modifications.
• I review text, hands, faces, logos, instruments and small details.
• Finally, I adapt the piece for a website, cover, poster or social platform.

This is closer to directing a sequence of versions than pressing a button. AI can generate an attractive image, but it does not always understand its communication purpose or preserve consistency from one edit to the next.

05 Suno as a pre-production tool

In music, Suno allows me to turn a written description into an audible reference. Sometimes an idea exists as a guitar progression, a voice note or an abstract concept; a quick demo helps evaluate how a full structure could behave.

I use it to explore possibilities, not to replace composition or performance. A demo may suggest a dynamic shift, texture or transition, but it can also fall into generic formulas or introduce elements that do not represent the project's identity.

06 My current workflow

1. I define the problem, the expected result and what must not be changed.
2. I choose the tool according to the task: long code, iterative conversation, visual exploration or a musical demo.
3. I provide enough context, while excluding credentials, private data and unnecessary files.
4. I save a working copy before applying any major change.
5. I work in small steps and ask for the modified files to be identified.
6. I test the result in real conditions: desktop, mobile, browser and server.
7. I compare alternatives, correct errors and rewrite anything that does not represent my intent.
8. I document the solution so it can be maintained without relying on the original conversation.

07 Where AI fails

Speed can create a false sense of certainty. I have seen convincing code that ignored the existing structure, technically correct but unnatural translations, images that changed details I never asked to modify and demos that solved everything with overly predictable formulas.

• It can invent functions, paths, configurations or results.
• It can fix one area and break another that was not visible in the prompt.
• It can sound overly confident when context is missing.
• It can repeat generic visual aesthetics and musical structures.
• It can lose decisions made several versions earlier.
• It does not independently understand the emotional, commercial or ethical intention of a project.

Human review is therefore not a decorative final step: it is the part that turns a plausible output into usable work.

08 Privacy, authorship and responsibility

Before sharing files, I check whether they contain personal information, credentials, client data or material that should remain on my own computer. In development, keys must stay on the server and outside public code. With photographs, documents and audio, I also consider whether I have permission to process them.

Platform terms can change as well. Before publishing or commercialising a result, it is sensible to review the current terms, keep the original files and be able to explain which parts were created, edited or assisted with AI.

09 Conclusion

Artificial intelligence has reduced the time required to move from an idea to a first version. It lets me test paths that previously required more resources, compare alternatives and document my process more clearly. It has also taught me that moving faster requires better backups, clearer instructions and a stronger ability to detect errors.

I do not see Claude, ChatGPT, Gemini or Suno as replacements for learning programming, design, production or music. I see them as tools that amplify what I can already direct. The result improves when there is judgment to ask, select, discard and rebuild.