Forum Discussion
AI image generator
Just playing around with the image generator to see what it comes up with. I am giving clear instructions and it is producing images that I would want to use. However there is something wrong with every image; someone has three feet, a hand that is back to front, 6 fingers, zombie eyes, one bare foot and one with a shoe on and in some cases missing half a body! How do I get round this? Does anyone have any tips?
23 Replies
It's really frustrating, I didn't manage to get one image in the 30mins of trying that I could actually use.
Our team has a collection of Articulate AI Image generation oddities. Here are a few:
I'm curious, what were the prompts you were using that generated such strange images? I try to stay away from prompts that focus on someone holding something or hands near their face because AI doesn't do hands well at all. I did create one image that I liked, but the skin was wrong on the knuckles. I opened it in Photoshop and removed the extra wrinkles.
I used these prompts for the three odd images:
- For both image 1 and 3: "a construction worker standing next to a pickup truck".
- For image 2: "an architect looking at a building plan"
Yeah, I got weird images for the 1st prompt too. Try this prompt:
"Pickup truck on construction site with a construction worker standing near it"
I got good results using my original prompts using other AI generative tools.
Using Adobe stock photos is a much better use of my time.
Yeah, I think Articulate's model will require continued training. You need to be quite specific in your prompting. I also use Copilot Pro and the Techsmith AI tools.
I have to agree, people are not well done. Even images have spelling mistakes - I asked for an image of a book and the title was misspelled. I'm very disappointed. I've also noticed that the language text to speech has problems with some words and there's no way to correct the phoneme. :(
I have found that specifying the exact text desired will at least 50% of the time, create an image with the correct spelling. I also noticed that the case of the desired text in the prompt is matched in the generated images.
Maybe its best to not include text, adding it later to the image.
I got one image the first timee I used it, and it was great. Since then I only get a bug alert:
...or an unhelpful AI assistant message:
The other AI functions seem to work, but Generating an image? ...not so much. Any suggestions?
Hi JonBrown-9de5e9,
Thanks so much for reaching out! Sorry to hear you've run into this snag with AI Assistant.
Is this happening with a specific set of keywords you're using when generating the image? I'm curious to know if this is happening on more than one keyword search!
I'd recommend if you're still experiencing this issue, please then do a repair of Storyline 360.
Looking forward to hearing from you!
Hey Luciana...
Thanks for getting back to me. After re-installing, rebooting, cache clearing - the usual basics - all is well and my images are generating. Now, if I can only improve on my descriptive writing, maybe my characters won't show up with hands growing out of their heads 🙄.
Jon
Hi Community!
I've been trying to generate flat UI design images and icons using Articulate's AI generator in SL360, but some details of my prompts are simply ignored. We're using vector graphics (exported as PNGs) in our courses with transparent/no background, and I'd like to do the same in SL, however, I couldn't figure out a way to generate images with no/transparent background. It simply doesn't care if I add this part to the prompt or the style description.
Any tips or tricks regarding this issue?
Thanks,
Attila
Attila, can you share your prompt
Shared it below.
Hi,
This is the prompt: "Hacker in grey hoodie behind a white laptop, hoodie covers his eyes, no logo on the laptop, transparent background"
Style:
The same as "Vector Illustration", I've just changed the background description.
"Flat design, vibrant limited palette, simplified geometric shapes, minimal detail, exaggerated proportions, 2D perspective, high res vector, symmetrical composition, iconographic elements, transparent background, modern cartoon-like aesthetic, clean lines, bold colors, playful scene, stylized characters, contemporary illustration style"
I tried but was not able to generate transparent background. So work around is to use a PNG background remover (Freely available online) and then import that image back into rise.
I am trying to generate an image of a wall calendar for 2025 with the month titles in English. I have listed the months. I get the months out of order, the months in other languages, the months duplicated. The numbers days in the months start with 2, 4 or 7. I just need a visual of a wall calendar. I have changed by prompts from general to very specific and I don't get anything that resembles a wall calendar.
Hi MaryGregg-f1028! Thank you for sharing this feedback with us! I'm also seeing jumbled text when trying to generate a 2025 wall calendar. I've shared this experience with our product team so we can continue improving AI image generation.
