Looks like music generation is on their roadmap. Fun!
DeepFloyd IF: open-source text-to-image model
201–210 of 237 posts
Re: DeepFloyd IF: open-source text-to-image model
#202This could be super cool for logos. I've tried using Stable Diffusion to generate logos and it does pretty good at helping brainstorm, but the text is always gibberish so you can use its idea, but you have to add your own text which basically means creating a logo from scratch using its designs as inspiration.
Try controlNet with stable diffusion, I’ve been having good results for text.
Re: DeepFloyd IF: open-source text-to-image model
#203Meh, results feel hodge podge like a bunch of models were stitched together
Re: DeepFloyd IF: open-source text-to-image model
#204The current license makes this largely unusable for nearly any purpose. Really disappointing release from SAI.
I think this is just for pre-release, and they will release fully licensed for Commercial. It doesn't make sense to have a model like this that can do game changing text and logos... but then not license it for commercial. If they don't, that would be ridiculous.
Re: DeepFloyd IF: open-source text-to-image model
#205Earlier quoted context omitted.
HF also wrote a blog post on how you can mess around with the model in a python notebook using their excellent Diffusers library: https://huggingface.co/blog/if
I knew the model would have difficulty fitting into a 16GB VRAM GPU, but "you need to load and unload parts of the model pipeline to/from the GPU" is not a workaround I expected. At that point it's probably better to write a guide on how to set up a VM with a A100 easily instead of trying to fit it into a Colab GPU.
What about people having RTX 4090 with 24GB or even dual? Does it run on it?
Re: DeepFloyd IF: open-source text-to-image model
#206This could be super cool for logos. I've tried using Stable Diffusion to generate logos and it does pretty good at helping brainstorm, but the text is always gibberish so you can use its idea, but you have to add your own text which basically means creating a logo from scratch using its designs as inspiration.
Well, surely you don't expect to just take the generated stuff and use it right away? For logos you usually need something that some entity can hold the claim on, so you probably need a human touch in any case.
MidJourney's v5 model did a great job producing our logo https://summer.ai/static/nearby/android-chrome-192x192.png
Admittedly it was one of several hundred I had it spit out for me. But the design was completely original and caught me by surprise.
I tried making modifications but everyone kept telling me they prefer the original exactly as MidJourney made it!
Re: DeepFloyd IF: open-source text-to-image model
#207This could be super cool for logos. I've tried using Stable Diffusion to generate logos and it does pretty good at helping brainstorm, but the text is always gibberish so you can use its idea, but you have to add your own text which basically means creating a logo from scratch using its designs as inspiration.
Well, surely you don't expect to just take the generated stuff and use it right away? For logos you usually need something that some entity can hold the claim on, so you probably need a human touch in any case.
Authorship is not a requirement for trademark, though having copyright as well is nice.
Re: DeepFloyd IF: open-source text-to-image model
#208Tried using right now, and it's way better than Stable Diffusion (be it 1.5, 2.1 or SDXL).
But is harder to get a good picture. This fine tuned with a good RLHF will be amazing.
Re: DeepFloyd IF: open-source text-to-image model
#209> Gorbachev holding meatball pasta in both hands. 1980s synth futuristic max headroom aesthetic. Neon lights.
> Aristotle in ancient greek clothes. Toga. New york, rain, film noir, fog, art deco, neon lights, blade runner sci fi
Seems to be holding up recently well with the first promt. Second was only OK.
Re: DeepFloyd IF: open-source text-to-image model
#210Related: https://stability.ai/blog/deepfloyd-if-text-to-image-model
(via https://news.ycombinator.com/item?id=35743727, but we've merged that thread into this earlier one)