Product placement and text

[stevew]

one of the biggest frustrations in AI is it garbles any text.

Eg I had an image of a clients van all branded with logo and text but the van was dirty.

I asked ChatGPT to clean the image up.

The result was a lovely clean white van but parts of the text and images garbled.

After a couple of prompts I finally had to go into illustrator and remove certain areas and re-do the text.

Does this workflow address this issue ?

I noticed in the promo video the shampoo bottle had some garbled text on the bottle.

[marmottevendee]

This is a very known “disease” of AI image generation.
Latest models like FLUX or now even QWEN image, are likely to do better with texts. Have you tried those?

[stevew]

Hi guys,

I have edited the sample application to test the AI generated results (which are great ).

The issue is you cannot do any product placement ( which is the key to producing most commercial videos).

So if the guy had a company logo on the t-shirt or a product in his hand any image / video would not make sense.

Consistency of character is great but text is just as important,

Any solutions?

[Sophus]

Like mentioned above, this is a general problem with AI image generation. It’s the same with getting normal looking hands.

Also newer models like Flux, who can create text with a better quality, still have massive problems.

Therefore, you need to do manual editing in a photo or video editing application. For example, render the original iClone animation with text and logos and then superimpose them on the AI image.

[stevew]

Any progress with this latest version.

[animagic]

Since you can now use reference images one way is to add an image that you want on the shirt or other surface and refer to that in the prompt. If there is text, I usually quote that in the prompt as well.

Below is an example: