In Review

In Progress

Completed

Main Roadmap

Hi Guys, I have read the <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="https://graydient.ai/loramaker-instruction-guide/">guide</a> at least 4 times now, I’ve watch the videos… and I must say that when you read it on it’s own, it all makes sense (for the most part) but when it comes to following it on <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="https://my.graydient.ai">https://my.graydient.ai</a> … I can’t figure it out. I’m not sure if the guide made sense at one point and then things changed and it just doesn’t anymore. but as a newbie coming in, it’s a mess. My Lora doesn’t work, and I’m finding it impossible to troubleshoot why. I want to first start by saying that I’m following the guide and I’m using the browser (my.graydient.ai webpage) to interact with and make images. I’m not using the telegram bot. - the Guide doesn’t always say if commands are for the web or the bot. Here’s where it first gets confusing for me… in the guide it says: <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="https://graydient.ai/loramaker-instruction-guide/#:~:text=So%20if%20I%20want%20to%20use%20the%20model%20above%2C%20I%20would%20write%20%3Cjon%2Dcarnage%3E%20in%20my%20prompt%20and%20also%20add%20%3Crealvis51">So if I want to use the model above, I would write &lt;jon-carnage&gt; in my prompt and also add &lt;realvis51&gt; (not pictured)</a> The word Realvis51 comes <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="https://stable2go.ai/concepts">from our models page</a>. 2 things things:<ol><li>Do I write the trigger word with these &lt; &gt; ?, i.e <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="https://graydient.ai/loramaker-instruction-guide/#:~:text=So%20if%20I%20want%20to%20use%20the%20model%20above%2C%20I%20would%20write%20%3Cjon%2Dcarnage%3E%20in%20my%20prompt%20and%20also%20add%20%3Crealvis51">&lt;jon-carnage&gt;</a> ? because a little lower it just show “photo of zrtq person”, it doesn’t say “&lt;photo of zrtq person&gt;”. (so now I’m thinking, are the &lt;&gt; just for the bot?)</li><li><a target="_blank" rel="noopener noreferrer nofollow" class="link" href="https://graydient.ai/loramaker-instruction-guide/#:~:text=So%20if%20I%20want%20to%20use%20the%20model%20above%2C%20I%20would%20write%20%3Cjon%2Dcarnage%3E%20in%20my%20prompt%20and%20also%20add%20%3Crealvis51">&lt;realvis51&gt; </a> - this is confusing.. I get the concept, and you even added that it’s from your model page…. but on your model page, there’s nothing call realvis51, when I do a search, I get 7 results, but they are all Loras, not base model, and they only show up because realvis51 is in the description of said loras. </li></ol>I created my lora, I have 10 images, I upscaled them, brought them into lightroom, made sure the white balance is the same on all photos, I cropped them all 1:1, I removed anything in the image that could confuse the AI. I edited all the photos so that they are crispt, no noise. skin tone perfect. I instructed ChatGPT on how to describe an image for Lora training (from the guide) and I feed it each photo 1 by one to get a good description, I modified it to make it more simple and proceded to upload all 10 photos with their own unique description. now, let jump to Tip #9 – Know the base model biasesI 100% understand what you’re talking about, but I don’t know how to test it. First, Lets start from this page: <img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0194b787-04dc-7263-b6a9-cd9d5a8c7fdd.png" data-width="100%" data-align="left"> Then I’m going to Filter: FULL MODELS, REALISTIC, and NSFW…. I see like 800+ Models <img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0194b788-ce9e-7e64-85df-9d23528939df.png" data-width="100%" data-align="left"> The idea here is that I need to test out these base models before I train my Lora to know which base model resembles my Lora best? ok.. so.. let’s go to the Lora Training to see what’s available… <img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0194b78d-6618-755b-995f-00faf2ce8823.png" data-width="100%" data-align="left"> Let’s pick the first one, anymagine31-xl If I search for this: I get 0 results. <img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0194b78e-d1c6-71de-a063-36c0e26edfb8.png" data-width="100%" data-align="left"> let’s pick the last one in the list… <img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0194b78f-59dc-7e11-84b7-b2489804f881.png" data-width="100%" data-align="left"> This nothing… ok, lets try StableDifusion 1.5 instead. <img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0194b78f-e3bc-72ca-b1eb-95bb0d6fd519.png" data-width="100%" data-align="left"> Hum… well it seems that it actually did find something now. <img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0194b791-36c7-7cbb-9dcd-d97841cb433c.png" data-width="100%" data-align="left"> ok, let’s say it found multiple base models, my next steps are to test the base models, with one of my descriptions of my Lora training photos right? and I’m trying to find the which photo comes out most resembling my lora without training it? So far, I haven’t had much luck producing realistic looking photos with just a base model, it all looks like a 3D animation. The only time I can get a decent looking photos are with “workflows” - and there are no mentions of workflows in the guide. Also, from what I can tell, all the workflow I’m interested in are FLUX based, so does that mean that my model wont work with any Flux base models? I don’t see an option to train my Lora with FLUX. Now lets move on to the next section: <a target="" rel="noopener noreferrer nofollow" class="link" href="https://graydient.ai/loramaker-instruction-guide/#:~:text=How%20to%20use%20your%20Lora">How to use your Lora</a> The example shows: Correct: a photo of zhzhzy person &lt;happyguy134&gt; &lt;realvis51&gt; What the F is &lt;happyguy134&gt; ? is this prompt for telegram? in the troubleshooting section: WEIGHTSYou can control how much or how little to enforce the training with weights. The default weight of a LoRa is 0.7. So &lt;mustacheguy:0.8&gt; is 1x more influential, and &lt;mustacheguy:0.6&gt; lowers it by 1x. If you get blue squares, tuning this helps. again with the “&lt;mustacheguy:0.6&gt;” what is this? the lora prompt for telegram? - where/how can I adjust the weights on the browser?

https://graydient.ai/loramaker-instruction-guide <<-- Needs clarification

Anon577051

Sometimes I create a model in the wrong community and the privacy feature doesn't let me see it from my preferred Telegram bot group. It would be cool if there was a web interface to manage where it shows up and who could access it, to make that more clearSo if a person wants to share it, there would be a pulldown for "public"

The ability to "move" and delete models, and training data

Captain

paste an Instagram profile handle and we crawl it for the catalog

Allow catalog to import photos from URL

It seems like a lot of models are personal or fantasy models, so a checkbox for this or making it the default would help