Self Promotion ComfyUIPromptBuilder and Beginners guide to generating AI images, Face Swapping and LoRA Training from your home PC.

[Ezdub69]

So I've always struggled a bit writing prompts, remembering them, getting image information metadata and image description and now more so that I use Krea 2, plus I don't like using online websites, like Grok, ChatGPT and so on and wanted to keep things on my local pc.

So I decided to build an app to help with that, which uses local AI models running through LM Studio and no need for it to be connected to the internet at all. (PLEASE NOTE: I'm no UI design so the look of the app is pretty plain but it functions well)

I've called it ComfyUIPromptBuilder but the end prompts can be used for other image generation websites and not just for Krea 2, It is after all just a prompt maker.

It does the following, don't worry that two of the headings are Krea 2 I just used that as it is the main model I use for my image generation in ComfyUI.
Krea 2 Prompt Builder - build a prompt from structured fields (medium, character,
wardrobe, pose, scene, shot size, camera angle, lens, camera, aperture, lighting, genre,
mood) instead of freeform chat. Each free-text field has its own "Vary" button for asking
the model for a fresh alternative to just that one detail.

Krea 2 Assistant - a focused chat interface with two jobs: describe an attached
reference image in plain, faithful prose (Vision mode), or rewrite/convert an existing
prompt - from Civitai, another tool, wherever - into Krea 2's format, including targeted
edits like "remove the necklace" or "change the pose" (Rewrite mode).

Image Analyser - browse a folder of generated images and read back their embedded
generation metadata (prompt, negative prompt, model, LoRAs, seed, steps, CFG, sampler,
scheduler, denoise), for both ComfyUI's own metadata format and the standard A1111-style
"parameters" text some save nodes use.

Saved Prompts - a browsable history of everything you've saved from the Builder and
Assistant tabs, ready to reload or export to a text file.

If you want to have a play with it I've stuck it on GitHub with a full install guide which to get it up and running, link below.

Any question please ask, would also like any feedback if you have any.

[Login or Register to view]

[Login or Register to view]

[Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view]

 

[Ezdub69]

It was discussed last week in the Fake Emma Watson thread about sharing how I create my fakes and it would be nice if I would share a how-to guide.

I was going to put this in the AI Fake discussion thread but decided to make this thread for beginners who may want to start dipping their toe into the rabbit whole. (If the admin/mods want to move this post there please go ahead) and they may have questions

So I’m not going to post the whole guide on how I set things up as it will a very long post so I’ve put a document together with everything you should need to get up and running.

It is all based on my experience and I have no doubt there are probably better ways of doing it but this is what worked for me at the start of my journey.

First thing you may ask yourself is can my system run this type of thing, well it is going to depend and everything I do runs from a laptop with a Nvidia 8GB GPU 16GB system memory laptop, so you don’t need some super high spec PC to get things going.

For the software to use I’ve chosen Forge Neo as a starting off point, I love ComfyUI and use it mostly fulltime now but I found when starting out Neo was much more user friendly and no complicated workflows and so on to deal with.

In the how-to there is two options to go with a DIY toolkit option, if you have some technical skills already and the other option is using an AI tool to get 99% of the job done for you.

You can find all the details and some resources in the link below, I’ve put the starting guide into 3 formats pick use whichever floats your boat TXT file, PDF and Word Doc which is called ‘Install Neo Guide’ which I, would give a good read before moving on, hopefully I’ve made it as easy to follow as possible but happy to answer any question if anyone has any.

Please feel free to chime in if you have better suggestions for those that know more than me, as I’m sure there will be many.

I will put together guides on Face Swapping in Neo and how to train a character LoRA later, hopefully in the next week or so.

[Login or Register to view]

 

[hoeappreciationmonth]

youre a hero dude thanks will give it a shot soon

 

[Ezdub69]

Sorry for not getting around to doing my two other guides, just not had much time as I've been working away but should have them done this coming week.

 

[Ezdub69]

So as a follow on from my LM Studio ComfyUI Prompt Builder a few people have asked if there would be a way of building something that does not need LM Studio and AI models to be used and downloaded.

So I created a new Pre-set Krea 2 ComfyUI Prompt Builder that does not need any of the above from my original Builder.

It contains around 9000 pre-sets for various things like subject, clothing, pose, setting, camera, lighting, style broken down into sections which you can select manually or use the wildcard options, as well as locking certain options. You can save prompts, add your own custom pre--sets lists and quite a bit more.

All you need to have installed beforehand is python and the app on the first run will install PySide6 and it all runs from the local folder, no internet connection needed after the very first install.

And to give you some context of how many random prompts it could generate using the wildcard button I asked ChatGPT to give me a rough estimate.

Distinct prompt combinations, roughly 3.76 decillion, honestly the numbers are huge there are rules of course but from my testing of it, it kind of works better than the LM Studio version in some ways and of course you can tweak the created prompt to your liking and if you're using trained LoRA, that may need trigger words.

You can find the new app on my GitHub page [Login or Register to view] it is a very small zip file on the release page and the full code is on the GitHub page.

I've kept the xxx pre-set list quite small to start with but if there is interest I can build a much larger list which can be added very easily after the install.

Some screenshots of the app.

[Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view][Login or Register to view]