Chat GPT - AI idiots guide?

[Lucan Unlordly]

I'm struggling to get my head around why Chat GPT messes up a good result when asked for a simple edit on a basic graphic?

I've uploaded a circular wreath graphic and asked for a 2d effect and a colour change which it's done well. I then ask for it to be kept exactly the same but for the silver colouring to be brightened, which again it does well......................but squashes the design turning the circle into an oval! I ask it to keep the circular shape and nothing changes.

This is not a one off. An image that I asked the background to be removed from changed the ethnicity of the subject!

HELP!

 

[Lisa Thomas]

I've only just started using CGBT but have found it easier to ask it to make the designs editable in Canva and make the corrections myself.

 

[erenackerman]

Ugh, I feel your pain! That ethnicity change is wild.

The issue is ChatGPT doesn't actually "edit" images—it regenerates the whole thing from scratch every time you ask for a tweak. It can't lock the original pixels, which is why it ruins your perfect circle.

Honestly, for minor tweaks or background removals, just take the first good AI image and drop it into Canva or Photoshop. It’ll save you a massive headache!

 

[Data Swami]

I'm struggling to get my head around why Chat GPT messes up a good result when asked for a simple edit on a basic graphic?

I've uploaded a circular wreath graphic and asked for a 2d effect and a colour change which it's done well. I then ask for it to be kept exactly the same but for the silver colouring to be brightened, which again it does well......................but squashes the design turning the circle into an oval! I ask it to keep the circular shape and nothing changes.

This is not a one off. An image that I asked the background to be removed from changed the ethnicity of the subject!

HELP!

What version of the model are you using? And what version Paid or Free?

Also what is the prompt you have given it?

However there are alot more tools that can easily remove the background Affinity i think its called is free and they have background removal and colour tweaking

If you really want to try Chatgpt, Here are a couple of prompts to try though:

Edit the uploaded image only. Do not redraw, regenerate, reinterpret, resize, crop, stretch, squash, rotate, skew, or recompose the design.

Keep the wreath exactly circular, with the same canvas size, same centre point, same outer diameter, same inner diameter, same proportions, same spacing, same 2D style, same transparent/background treatment, and same position as the uploaded image.

Make one change only: brighten the existing silver colouring so it appears lighter, cleaner, and more luminous.

Preserve everything else exactly as it is, including:

• the circular shape
• the wreath thickness
• all leaf/ornament shapes
• all outlines and edges
• all shadows/highlights except where needed to brighten the silver
• the original 2D graphic style
• the aspect ratio
• the background/transparent areas

Do not turn the circle into an oval.
Do not scale the artwork horizontally or vertically.
Do not alter the composition.
Do not add new elements.
Do not remove any elements.
Do not change any colours except the silver areas.

Output the edited image with the same dimensions and proportions as the uploaded original.

OR seperately:

Use the uploaded image as a locked reference. Treat the geometry and composition as fixed.

Only brighten the silver colour values. This is a colour adjustment task, not a redesign task.

The wreath must remain a perfect circle, not an oval. The artwork must keep the same width-to-height ratio, same centre, same diameter, same thickness, same placement, and same canvas dimensions as the uploaded file.

Do not redraw the wreath. Do not reinterpret it. Do not improve the design. Do not change the shape, proportions, layout, linework, background, transparency, style, or decorative details.

Only make the silver brighter.

And a generic Remove background prompt

Remove the background only. Do not change the person or subject in any way.

Keep the subject’s face, skin tone, ethnicity, hair, clothing, body shape, expression, lighting, colours, texture, and proportions exactly the same as in the uploaded image.

Do not retouch, beautify, recolour, stylise, redraw, replace, or reinterpret the subject.

Only remove the background and make it transparent. The final image should be a transparent PNG-style cutout with the original subject preserved exactly.

 

[HFE Signs]

Hopefully in a few years time it might start generating images in true vector format, then people like us will love it, for now it isn't quite there and certainly not capable of print resolution.

 

[Data Swami]

Hopefully in a few years time it might start generating images in true vector format, then people like us will love it, for now it isn't quite there and certainly not capable of print resolution.

You can defo create in true vector format can even do 3D model format too but not via the regular image way. There are a couple of tools that specialise in vector images and creating 3D model files from AI prompts too. But you can still try prompting ChatGPT etc to convert the image you give it into a true SVG vector file etc.

 

[HFE Signs]

You can defo create in true vector format can even do 3D model format too but not via the regular image way. There are a couple of tools that specialise in vector images and creating 3D model files from AI prompts too. But you can still try prompting ChatGPT etc to convert the image you give it into a true SVG vector file etc.

I'll check this out, last time I tried (quite a while ago) it gave me a vector pdf but it was just a load of random lines.

 

[Data Swami]

I'll check this out, last time I tried (quite a while ago) it gave me a vector pdf but it was just a load of random lines.

Ye make sure to use the Pro and Extended models if you are using ChatGPT to get the best chance of it working properly. Otherwise one of the many specialied AI tools. Or if you want to be fancy you could technically use something called ComfyUI which can use loads of different models and use your computers own hardware. Loads of free workflows out there for it for different things like character creation from prompts or images all the way to generating video

 

[Lucan Unlordly]

The biggest problem for me is using prompts that works well on one piece of rectangular artwork - 'please retain the exact size and specification, enhance the overall image and turn it into a watercolour painting' etc., which when used on a virtually identical image is presented as a square! I'm not trying to re-invent the wheel.

 

[Data Swami]

The biggest problem for me is using prompts that works well on one piece of rectangular artwork - 'please retain the exact size and specification, enhance the overall image and turn it into a watercolour painting' etc., which when used on a virtually identical image is presented as a square! I'm not trying to re-invent the wheel.

affinity might suit your needs better then. its free but some generative ai features are paid for but its tailoured for being the alternative to adobe so fits much of what you seem to be asking

 

[fisicx]

Or….

Support your local graphic developer and give then the work.

 

[Lucan Unlordly]

Or….

Support your local graphic developer and give then the work.

He's had to get another job to support his design work as AI has taken the bread and butter

£50 an hour is also too steep for 99% of what I need doing.

 

[bizvids]

Try using Nano Banana; it's much more forgiving IMHO.

 

[Lucan Unlordly]

Try using Nano Banana; it's much more forgiving IMHO.

There are a lot of sites that appear to go beyond what I need. Lots of colour examples featuring the extremes of AI. Chat GPT in theory does the job. Upload Image, request changes.