There's a new AI player on the grid: Promptus. Promptus brings together lots of existing AI facilities and systems in a single easy-to-use system that can produce some utterly stunning results.

Promptus is several things combined:
A web-based image and video generator.
A downloadable app for generating images and sharing your facilities with others.
A whole series of Microsites that specialse in specific types of images.

If you want to test the waters it's quite possible to generate some images for free, signing up a free account gets you some free credits to play with the facilities and download full size images, and you can easily buy more. And a fully paid for account gets you unlimited access to a whiole suite of very advanced tools and facilities including video generation.

Below are just a few WAM images I generated using the system tonight, but the capabilities and facilities are truly impressive.

Want to create some game-show style green gunge images? Go to the Splosh-Gunge microsite and type in a description of the person you want to see being gunged, and get back four assorted images of them meeting their messy fate! And it'll even make most or all of the background audience match the gender of the person being gunged, entirely automatically.
https://splosh-gunge.promptus.ai/

Want more control? Head on over to the main Promptus website and either download and install the app, or just use the web portal. Craft prompts, select models, or delve into the depths of ComfyUI and Cosyflows for some really advanced abilities. There's a blog and video tutorials explaining how to use everything.
https://www.promptus.ai/

That sounds suspiciously like a sales pitch or referral farming.
What's the differentiator Vs using individual sites or existing aggregator sites sites like fal, CivitAI, etc?

What do the microsites do that are different? Presumably they use the same underlying models such as Veo, WAN for video. Imagen, Flux, OAI for images? Or do they use custom models? if custom models, what are they on or just pulled from online repositories?

Call me cynical but I don't see anything new here other than a credit system where you pick and choose a model to run against? The gunge images look like Imagen or OAI/Sora images for example?

Yeah sounds more like a GTM strategy than any specific mod of the model.

Nope, no referrals involved, I just thought it might be of interest to people. I know some of the people behind it so have a personal interest in promoting it but am not getting anything from doing so - my own business is firmly real-people WAM, while I enjoy playing with AI wam I don't see it replacing modelled scenes any time soon. It's good, but not yet that good. Time will probably come, but not next week.


Ease of use, basically. Yes, it uses (lots of) existing models, it's not a new AI. But for less tech-savvy users who may not be ready to dive fully into running their own models and GPUs it makes things easy. You can also opt to share your GPU compute resources on the network to earn credits I believe (I don't have a GPU in the Mission Control PC so haven't experimented with that side of it).


It's actually quite simple, you set up a microsite with specific choices of models and helper text and then when a user comes along and enters a prompt the helper text gets added and the whole thing submitted to the model(s - up to 4) that the site owner has selected. So if you find a parictlar models is very good at a specific type of image, you can set up a microsite to use that as a nice easy to use way to generate more images.

I love pre-Raphaelite art, and women in boilersuits. Sadly the actual pre-Raphaelites didn't paint women in overalls, but several of the models now available are quite good at that style, so I can have my own alternative version of Burne-Jones' "The Brass Tower" - see attached image.


Sora isn't available in the UK, I think I used Nano Banana for the ones above.


It's not any kind of model mod, no, just an easy way to generate - you can play with it for free at thumbnail size, create an account and sign in and you can download the full size versions, and you get free credits to experiment with. Probably way too simplistic for advanced users but hopefully some folk might find it useful or fun. No ulterior motive.

Makes sense. I guess I struggle to understand the market for it with Google having free tiers for Gemini, Imagen and VEO and OpenAI with similar offerings with GPT and Sora. Sora1 is available in the UK with video and images, Sora2 not yet. Ideogram likewise have free and paid tiers etc.

The red flag for me was the open models such as flux QWEN etc. If it's the base models, great Being able to run them without needing a GPU and renting the GPU as needed is great for less tech users but for wam material you need trained Loras which not many are doing/doing well. I released some loras on CivitAI which were subsequently stolen by SeaArt and other services for use in their generation site. In other cases, users took them and loaded as their own. My concern was that was also this case here for the microsites. This doesn't appear to be the case though.

I think the main advantage is combining things together, making it simple to use, plus the resource sharing facilities. It's still very much under development but the team said i could let people know about it - been playing with it privately for a while.


Aha - no, nothing like that and definitely no stolen loras. The team is more about prompt engineering and ensuring users get the best out of whatever model they chose to use - so far I've seen everything from hyper photorealistic, through stuff like my replica Victorian oil paintings, all the way to 8-bit pixel art. The potential range is huge and it's fun to experument with the same prompt across different models and see how varied the results that come back can be. The three images below were done using Seedream 4.0, with my "pre-Raphaelite" overlay on the prompts. Planning to use images like these - very clearly painting-style, not pretending to be real photos - to illustrate some of the house and historic characters backstories when I finally get round to rebuilding the Saturation Hall and Langstonedale websites, which is long overdue. The thirtd image below for example - the woman in the white satin blouse is clearly the wife of the Seventh Earl of Langstonedale, painted in the early 1880s, with her trusted companion in the green top, wearing some of their best clothes into the village duckpond together.

It's not my thing but I guess you need to let the market dictate demand. If it's there it'll grow.

The resourcing aspect is a bit paradoxical when it comes to image generation. Those that have a capable GPU such as as a 3090, 4090, 5090 or RTX pro 6000 are probably already inclined to run it themselves locally. It would be significantly cheaper than the electricity costs incurred. they all pull over 400W under load and the RTX pro 6000 will top 600W unless underclocked.

I hear you on the prompt engineering, I've tried since dalle3 to try and explain each model needs it own prompt structure. GPT and Gemini can help no end to write better prompts yet next to no one seems to use them to help. Personally, I think prompting is a skill everyone should brush up on

I've attached a few pics in various styles. Pretty happy how well my model is developing.