|
|
||||||||||
Trying Free AI Tools for content creation: Writing, & Images
My personal opinions & experiences
Keywords:
chatgpt, website, sources, copilot, privacy
Key Takeaway:
There are several free AI platforms online.
Category Insights:
AI SERIES — AI is rapidly changing how we work, learn, and interact with
technology.
SECTION 1: Using Free AI Tools for Image Generation
There are several free AI platforms online that work well for content creation.
I’m exploring them using the free options as I have not purchase any
subscriptions. The goal is to see what you can reasonably do for free, and when it makes
sense to upgrade. I’m testing writing, & image generation, and news.
Below are my personal results and tips along with my opinions as to which AI that I
prefer for each task. My goal is to see what you can reasonably do for free, and
when it makes sense to upgrade. Results are not scientific.
Google Gemini: Fast, Near Photo-Realism (With a Catch)
By far, my best results with free AI image tools came from Google Gemini. It has
quickly become my go-to recommendation for fast, high-quality visuals, though the experience
taught me a lot about how AI "thinks."
The Good: Near Photo-Realism in Seconds
I started by asking Gemini to generate an image of a woman with an umbrella
walking in the rain in a big city. The first image materialized in a staggering 10 to 20
seconds. (Note that generation times do vary depending on the complexity of your request
and the current server workload). Gemini took my basic description and masterfully
filled in the details to create a gorgeous, near photo-realistic picture.
From there, I started refining. I updated my prompt to be more specific, changing it to
a red umbrella and a leather coat. The second image was generated just as quickly, and
the quality was spectacular. I have include 4 of those images below.
The Quirks: The AI Guessing Game
However, this is where the quirks of AI generation show up. In that second
prompt, Gemini independently decided to change the city background entirely and place
the woman right in the middle of the street! Because I didn't explicitly specify a
location, the AI just made an assumption. Maybe she’s in a crosswalk or a bike lane, but
I probably should have told the AI to keep her on the sidewalk!
I experimented with several more variations on this theme by changing the
prompt slightly to alter the clothing. Every single image was visually spectacular and
almost looked real. For fun, fast results, Gemini is quite good.
The Frustration: The "Loop" Effect
Gemini certainly isn't perfect. There are times when you are trying to
micro-tweak an image and the AI just refuses to understand your request. During my
testing, it would occasionally place objects in the wrong spot or add elements I never
asked for. Most frustratingly, it would sometimes just regenerate the exact same image
without making any of the changes I requested.
Despite the occasional refinement headaches, Gemini tends to beat the
competition away in terms of speed and realism. It's the best choice for my
purposes. It is a fantastic tool for creating images—just remember that if you don’t
explicitly specify a detail, prepare for the AI to make its own creative
assumptions!
The last image shows how Gemini can produce unexpected and sometimes humorous results. The prompt that I used was to show a spinner style mechanical harddrive with bad sectors sitting at a 45 degree angle on the workbench with the top removed. As you can see, it completely went off the rails! The more you look at it, the more problems you'll see. There are at least 6 major issues with this image. It just could not comprehend what I was asking. I think I went through 20 iterations to final get what I needed.
The Hits, Misses, and Realities of Using Microsoft Copilot for AI Imagery
Like it or love it, Microsoft Copilot and its AI-generated imagery are
here to stay. Over the last several months, I’ve put its free image generation tool
through its paces with mixed results. While it has noticeably improved since its
early days—allowing me to generate dozens of usable images for my website—the
experience has been a mix of impressive wins and frustrating
bottlenecks.
The Good: Quick, Creative, and Fun
Where Copilot truly shines is with creative, straightforward prompts. I
experimented with uploading family photos and asking it to turn them into simple
cartoons. The results were excellent, usually wrapping up in just a minute or two
depending on the complexity of the photo. In fact, a cartoon portrait I made for my
niece was a massive hit when I posted it on Facebook. For stylized, personal
projects, it’s a fantastic (and fast) tool.
The Bad: Where Specificity Goes to Die
However, things get messy when you need precise, technical
accuracy.
I recently tried to generate a network diagram from a text prompt. Structurally, the
layout looked great, but Copilot kept making the exact same error over and over.
Even after rephrasing the prompt multiple times, the AI simply couldn't grasp the
concept of what I wanted. Ultimately, I had to open up an image editor and manually
correct the mistake myself before I could use it.
My Advice: Patience and Precision
If you’re going to use Copilot for web design or content creation,
keep these two things in mind:
Expect to iterate: It often takes several tries and multiple prompt tweaks to
get an image exactly the way you want it to appear.
Be brutally specific: If you leave any room for interpretation, Copilot will
fill in the blanks—often adding unwanted elements or completely misinterpreting
your intent.
It is also worth mentioning that if you subscribe to Microsoft 365
Personal or Family, you actually get greater access to additional, premium
Copilot features. If you're already paying for the subscription, you might as
well take advantage of it!
The last image shows how CoPilot also had trouble producing a drive with the cover off. It produced a very strange looking drive with a lot of odd results. The prompt that I used was to show a spinner style mechanical harddrive with bad sectors with the top removed. As you can see, it didn't do any better than Gemini; even with a simplified prompt. AI seems to have issues with mechanical hard drives!
Expanding the Test: Does ChatGPT Free Tier Cut It for Imagery?
Naturally, I couldn't test Microsoft Copilot without also giving ChatGPT
a fair shake. I've been using ChatGPT's free tier, and while it has always been quite
good for text generation, my journey with its image generation capabilities has been a rocky
one.
The Early Days: Painfully Slow
When I first tried generating images on ChatGPT, the performance was abysmal. I submitted a text prompt for a simple cartoon (no photo uploaded), and the tool stalled out. It ran for hours, and it wasn't until the next morning that I saw it had finally completed—and even then, I was completely underwhelmed by the result.
My next few image requests never finished at all. I suspect the system
was severely overloaded at the time. I have no idea. I suspect performance would improve on
a paid plan and / or during off-peak hours. That said, as OpenAI has continuously improved
and has rolled out newer model versions over time, I did notice the speed and quality
gradually ticking upward. Maybe the system was overloaded at the time. I have no
idea.
Today’s Reality: Faster, Better, but Crippled by Limits
Fast-forward to today, and ChatGPT's image generation is undeniably faster, and the output quality is vastly improved. However, it comes with a massive catch: the free tier limits you to just five images per day. (Note: OpenAI dynamically adjusts these caps based on demand, but it remains heavily restricted).
Because AI image generation inherently requires multiple iterations
to get right, hitting a hard wall after just a few prompts completely destroys your
workflow. Being told you have to wait 24 hours to tweak a detail is an immediate
dealbreaker. I'll stick to Microsoft Copilot or Google Gemini for image generation.
Others:
Yes, there are many other AI models that can generate images. However, some do
not offer a free tier. I will update this article if and when I get the opportunity
to try them. Grok is high on my list of AI tools to explore.
The Verdict on Free Tiers for Image Generation
Because of these strict limitations, I've decided that ChatGPT's free tier simply isn't worth my time for image creation. If you're looking for a free tool to generate graphics for a website or blog, you are much better off sticking to the free tiers of Microsoft Copilot or Google Gemini, both of which offer far more daily flexibility.
Paid image/video models exist and are usually better and faster—but some can be expensive. For one-off projects, try the free tier first to see if it meets your needs.
SECTION 2: Writing with AI (This section will be rewritten with more details soon!)
ChatGPT by OpenAI
This is the tool I use the most for writing, & blogging, coding & webpage
design. It’s great for:
- Rewriting my own articles for better flow and clarity.
- Turning an outline or even just a title into a full draft.
- Fixing grammar and adjusting tone (casual, friendly, or professional).
- Generating HTML and CSS that I can adapt into this website.
- Coding with literally any programming language.
Unexpected Benefits
AI will give you ideas you may not have expected. Some of the page
formatting used in various places on this website was created by ChatGPT. I was able
to integrate many features into my template for use anywhere I choose on this
website. In fact, I have learned lot of new things about HTML and CSS by utilizing
code generated by ChatGPT generated. This was a game changer. AI can take 'good'
input and make it better and more efficient while eliminating mistakes. AI can make
you more productive if used properly.
Pros
- Strong at structure and readability.
- Helpful suggestions and quick edits.
- Good at matching a style if you paste a sample.
Cons
- Free plans can be slower at busy times.
- Daily or session limits may stop long tasks.
- Facts may be outdated; you still need to verify.
Tell Chat GPT that it should play the role of copy editor. Paste a document you wrote and say, “Match this style.” Add: word count, audience level, and key points.
CoPilot by Microsoft (Within Microsoft Word)
I also tried CoPilot. It was easy to use but I don't always
like the results. It generated 3 different choices for each item I asked it to
rewrite. Each was technically correct and worded differently. However, the overall
tone and wording just wasn't right. In my use case, I want to use the content on my
website it would often generate text that sounded to technical or too cerebral for
my target audience. I wanted to keep things on my website easy enough for anyone to
understand. Perhaps I wasn't prompting it correctly to get results I wanted.
General Tips for Better Results for Writing Tasks
- Be specific: Say the audience (e.g., “high school level”), length, and tone.
- Provide examples: Paste a paragraph of your style to match.
- Chunk big tasks: Free plans have limits—work in sections.
- Verify facts: Ask for sources and check them yourself.
- Save prompts: Keep your best prompts in a doc for reuse.
- Respect privacy: Avoid sharing sensitive data with online models.
- Expect filters: Image tools block some content; rephrase if needed.
When to Consider Paying
- You need faster results and fewer rate limits.
- You want higher-quality image or video generation.
- You rely on current, cited information for work or school.
- You value priority support and advanced features.
Always try the free tier first. If it saves you time or prevents mistakes, a paid plan can
be worth it.
MOVE SELF HOSTED PORTION INTO POST 88
Self-Hosted LLMs:
I tried running local models for experimentation and website summaries.
I can’t recommend them for most people yet. When I asked for a summary of a
local author’s website, the model attributed the author’s original work to
someone else. This was a deal killer for me.
- Llama 2: Feels outdated now. It simply doesn't perform as well as newer models. I probably won't use it anymore.
- Llama 3.1: Much better than 2, but still missed important details in my tests.
- Gemma2:9b If you have an 8 Gig video card, this model may be a good fit for you. It fits well within the memory constraints of a typical consumer GPU. This is a good option for running AI locally.
Why Run Locally?
- Privacy: Your content stays on your own machine and won't be used for training purposes. This is a strong selling point if your dealing with confidential or your own copyrighted materials.
- Offline work: It can be used even if you have no internet or internet is limited.
- Customization: You can add your own documents.
Running Locally Can be Challenging
- Hardware need: Using just your CPU-only can be slow; NVIDIA GPUs with at least 8GB+ VRAM help considerably.
- Model size: Larger LLM will consume use more RAM/VRAM and disk space. We recommend at least 32 Gig of RAM, but even more is recommended.
- Quality: Self-Hosting may still lag behind top hosted models that run on MUCH larger systems with hundreds of servers and GPU's. A home user or small business simply can't afford that much hardware.
If you care most about privacy and can handle the setup, local models can be
worth it. For everyday accuracy and speed, hosted AI is usually easier.
(For more background, see my related post #34.)
My Recommendations (NEEDS UPDATING)
| Tool | Best For | Speed (Free) | Accuracy | Notes |
|---|---|---|---|---|
| Open AI ChatGPT | Writing, editing, structure, coding | Less than 30 seconds for most results | Always proof-read and verify everything | Great for drafts & rewrites, check sources |
| Microsoft Copilot | Simple images & cartoons | 1-2 minutes per image | Good for casual art | Complex diagrams may need edits |
| Google Gemini | Image Generation | Seconds | Good quality art | Must be PG |
| Grok | Breaking news & trends | Fast | Varies by source | Ask for sources; expect bias |