Generated images are not responsive to the specifications in the prompts

[S31995984e2fl]

Version: Probably the latest.

OS: Windows

Prompt used: "A pencil-sketch using "only these colors", #0E512E, #5EBB47, #595959, #999999, #FFFFFF, #141414, fill randomly with geometric symbols relating to fitness, ageing, exercising, health, and lifestyle. Do not include pencils in the image".

Expected results: Image without pencils in it. Anything related to the symbols (geometric) requested - not an image full of cripple beings/aliens.

There must be a huge bug with Firefly prompts. Unless it becomes useful to ordinary humans, using ordinary language, this tech is not useful at all. Suggestion: Link it to a well-trained language model, and maybe it can begin to consider every detail in the prompt, not just the ones you think it should respond to.

[droopydog500]

Hello @S31995984e2fl,

Thanks for your message. When I read your message, my first reaction was that based on what you are seeking, using Firefly in Illustrator might be a better path. I find it makes better icons/symbols than the web-based Firefly and you can have tighter colour control, especially with the added controls in the beta version of Illustrator.

A few observations on your prompt. Generally, it is best for the main object to be as far left as possible and clearly defined. I think the main object here is the "symbol", so that should be as far left as possible preceded by the least number of impactful qualifiers possible.  It should be followed by some more general instructions.

The mixture of these words: "fitness, ageing, exercising, health, and lifestyle" I think is so wide, I would break them down into separate generations and pick from those results and then assemble them into a larger image (using Adobe Express or Photoshop).  

Firefly cannot limit to colours using the hex values. You can use words to describe the colours, but they won't be as precise as you want. So, you could then take the results into Photoshop to change the colour, or use Illustrator which does allow more control of the colours, especially in the beta version.

So, after all this, my first pass on fitness generated this:

I am not sure whether that is on track or not for what you want.  Here is a link to it if it is helpful: fitness symbol, geometric icon, pencil sketch, duotone, red black 

Please let me know if this is helpful or if we can help you further.

My best,

    droopy

[Dave Gustafson]

Hi,

The prompt is the problem, you are working against yourself. I tried it and I'm not getting pencils or mutants. Don't say anything with a pencil, this poisons the water! Use the left panel under content type and select Art. Then scroll down to Effects and choose Concept Art, then scroll down to Sketch and choose Sketch. Then Geometric and finally Flat Design. Flat Design can be toggled off if you want a more 3d appearance. The Hex color code might not work in Firefly never tried it.

Try this:

A sketch using "only these colors", #0E512E, #5EBB47, #595959, #999999, #FFFFFF, #141414, fill randomly with geometric symbols relating to fitness, aging, exercising, health, and lifestyle.

Or

sketch using "only these colors", #0E512E, #5EBB47, #595959, #999999, #FFFFFF, #141414, fill randomly with geometric symbols. ; relating to fitness, aging, exercising, health, and lifestyle concept

Or

A geometric sketch using "only these colors", #0E512E, #5EBB47, #595959, #999999, #FFFFFF, #141414, fill randomly with geometric symbols. ; relating to fitness, aging, exercising, health, and lifestyle concept

Or

A geometric sketch using "only these colors", #0E512E, #5EBB47, #595959, #999999, #FFFFFF, #141414, fill randomly with geometric images relating to fitness, aging, exercising, health, and lifestyle

 

 

 

 

 

 

 

 

 

I'm just taking shots in the dark. Personally the word "symbols" should be replaced by "images" but that's me. If you like the look of an image import that as a "Style Reference" in the left panel. Use the styles, effects, techniques, etc.. to further refine the image to your vision in the left panel.

I tried this in another AI generator, DALL-E and I get similar results, so it's not a bug.

 

[droopydog500]

@Dave Gustafson, when the original poster said "symbol", I thought they meant like an icon or logo.  They might have meant more broadly which your images reflect. 

[Dave Gustafson]

Yeah, the word symbol is very confusing, it confuses me, so Firefly is tripping on that also. Aging was spelled wrong so Firefly might have skipped that also, I corrected that in my prompts.