Mudge

This is Part 4, and yes the last part, of this introductory article guide to LORA creation. Here I will present links, many to LORAs and images I created while writing this, to provide a comparative sampling of what can be made without needing to know a lot of technical details. Seriously, I basically just uploaded the images and let the tensor.art training platform do its thing. So now you need to decide whether to...Step 3: Dive in head first or check out one of the tensor.art specific step-by-step guides for LORA creation.ADDENDUM: Since writing this there have been (or will be by the time you read this) changes here at tensor.art. As I write this what those changes may entail, how they may affect TOS policy or LORA creation is unclear. However, in a effort to get ahead of things, I wrote a follow-up article about FACSIMILE images, at least how I define them, and what that means as many of my test LORAs used such AI generated images. The article can be found in my articles. (Can't seem to copy links today.)If you have a dataset ready and feel like you're good to go then get started, you can peruse this article at your leisure while waiting for your LORA to train. Or, if you're still uncertain, there are many great informational articles published here on Tensor.art. To begin I’d suggest skimming at least one of the step-by-step guides before diving in blindly. Here are links to a few:https://tensor.art/articles/875734357225025459https://tensor.art/articles/806929088523547379https://tensor.art/articles/849095063096129962https://tensor.art/articles/732090335657730218https://tensor.art/articles/862381990022788927https://tensor.art/articles/806909589372248366All steps in the process ultimately lead to...Choosing your LORABottom line the best LORAs cost the most to train. There are a few that you can bank credits for in a short period of time, depending on how many images you have in your dataset. The two best for this are probably the Lightning FLUX and SD 3.5L. But they are very different.How different?As a test I created a dataset comprised of 6 Dystopian, 5 Fantasy, and 9 Wasteland warrior AI generated images using the comedian Katherine Timpf as my character template. All that means is, instead of describing a character in detail, my prompt ask was “Katherine Timpf as a” or “depicting Katherine Timpf as” kind of syntax. To round the dataset out I included 6 images of a character holding a sign reading “Kat”, 4 ‘photoshop’ style AI retooled (background cutout/replaced, indrawn touch ups, etc) genre themes images, 2 upscaled headshots, and 10 random other images. That should make for a solid, if not better, dataset than using just 10-12 images, right?Let's find out. I have generated comparison samples for both versions of the WASTELAND WARRIOR KAT LoRAs: https://tensor.art/images/876085333362493423?post_id=876085333362493424&source_id=njeyo1nrnEW3oPYsaX309xkkhttps://tensor.art/images/876061863513690934?post_id=876061863513690935&source_id=njeyo1nrnEW3oPYsaX309xkkhttps://tensor.art/images/876062475546563295?post_id=876062475546563296&source_id=njeyo1nrnEW3oPYsaX309xkkThe first image sticks close to what the image set was intended for. The Witch art, which actually looks pretty amazing, was the best test I could think up that had completely no aesthetic similarities with the source images. I think they prove the Flux LORA is fairly versatile with prompt interpretation on its own whereas the SD 3.5L version seems less so. Flux costs more but the images should speak for themselves.For example this is what we get when combining WASTELAND WARRIOR KAT with other LORAs:Wasteland Warrior Kat FLUXhttps://tensor.art/images/876088023085772229?post_id=876088023085772230&source_id=njeyo1nrnEW3oPYsaX309xkkhttps://tensor.art/images/876090879239001728?post_id=876090879239001729&source_id=njeyo1nrnEW3oPYsaX309xkkhttps://tensor.art/images/876091719978895752?post_id=876091719978895753&source_id=njeyo1nrnEW3oPYsaX309xkkNot terrible. Now let’s try…Wasteland Warrior Kat 3.5Lhttps://tensor.art/images/877386430354090215?post_id=877386430354090217&source_id=njeyo1nrnEW3oPYsaX309xkkhttps://tensor.art/images/877386722411870693?post_id=877386722411870694&source_id=njeyo1nrnEW3oPYsaX309xkkhttps://tensor.art/images/877388513413183158?post_id=877388513413183159&source_id=njeyo1nrnEW3oPYsaX309xkkObviously the first couple with the polar bear were a quick test. Simple prompt. No reference to the character or character description. (She even disappeared entirely from a couple image generations, like the 3rd image above, for some odd reason.) Now here’s four with more or less the same prompt, but generating very different artistic styles:https://tensor.art/images/877534034421380333?post_id=877534034421380334&source_id=njeyo1nrnEW3oPYsaX309xkkWith SD 3.5L the prompt has to be worded very specifically whereas with FLUX you can write a few words and probably get a decent looking pic. Even the polar bear prompt generates a slightly different looking image without adding extra descriptors when using the FLUX version: https://tensor.art/images/877593127802635927?post_id=877593127802635928&source_id=njeyo1nrnEW3oPYsaX309xkkOf course you don’t have to use either. But, by now, you should have a dataset ready (if not already uploaded) and all that’s left to do is train your chosen LORA then…Step 4: Name and Publish your LORAI know what you’re thinking... What did Mudge do that necessitated spelling out the obvious final step as if it’s difficult?I chose a LORA name “Bare Ten” (because it only used 10 images), which sounded better than “Just Ten” in my head. Also I thought it fit well with my low-key “Barefoot” naming meme. The problem? I forgot “bare” is one of those words that is a red flag for auto-mod filters everywhere. It’s okay to laugh because that wouldn’t be a problem, except for the fact I often use my LORA names as trigger tags.You see where this is going. Too bad I didn’t. But I quickly discovered the problem as I had to log in to see images created using those tags. So now I feel kind of silly because it just never occurred to me. The fix was simple, rename the tags. Which I did. I even came up with a more descriptive name for the LORAs. The links to them (and a few more I created) are below.Ten Scenes Emilia: https://tensor.art/models/871749988005217621?source_id=njeyo1nrnEW3oPYsaX309xkkhttps://tensor.art/models/872118629343169865?source_id=njeyo1nrnEW3oPYsaX309xkkTen Scenes Pirate Girl:https://tensor.art/models/871704761999581375?source_id=njeyo1nrnEW3oPYsaX309xkkVintage Photo Nichelle:https://tensor.art/models/872788354331067482?source_id=njeyo1nrnEW3oPYsaX309xkkhttps://tensor.art/models/872497183465720805?source_id=njeyo1nrnEW3oPYsaX309xkkWhat the above all have in common is they’re either trained on 12 images: 10 main scenes and 2 poster art mockups with title text or fewer than 20. I don’t know if it helps to train a LORA including images with text or not but I figure it can’t hurt. Besides I am testing my work prompt anyway to be sure it creates a decent looking title so might as well use the sample images, right?Well, that’s it. Thanks for reading. I hope this helped. All that’s left is a refresher on the…Simple Annoying Stuff:IMAGES: Whether curated from an public domain image archive or sourced from free range pictures you corral and lasso in an image search, there’s always two things of primary importance. First, clarity. Second, quality. The better your source image the better your training will be, which means a better model result.UPSCALING: Not necessary for images of decent quality. However, for the best results, use a tool that outputs images of high-quality but not necessarily large file sizes.IMAGE OPTIMIZATION: This is sort of related to the above. If you have a tool that outputs too large of a file size this may become necessary. However, and here’s a trick I discovered by accident. Some tools will only ever produce images in the 1+ MB size range. Problem is optimization often loses some of the quality. What to do? Find a tool that outputs smaller already optimized files. Often you can upscale images through another tool and not only get an optimized file but a fairly decent image with more pixel resolution. Doesn’t always work but when it does, golden.RETRAIN: Before your LORA expires you can RETRAIN it, meaning use all your already uploaded pics and generated tags, to create a NEW LORA. And, no, it doesn’t have to be the same model you already trained on.ADDENDUMTHIS IS KIND OF IMPORTANT. I believe, to properly implement LORA TRIGGER TAGS, you need to BATCH ADD your trigger word(s) to the BEGINNING of your tags. I've been testing some of my early LORAs (before I knew about the importance of manual tag editing) and it seems like there's no real difference in output. If someone could verify whether this is a necessary manual step in the training process, or if the auto training actually implements your trigger word, that would be immensely helpful. Thanks!Now Let's Go Make a LORA!Tap the menu tab at upper right of screen. On the drop down menu you should see "Training" (just above "Creator Dashboard"). Click on "Training". This will take you into the first of the LORA training pages. From here you begin the process of uploading your images, or a prepared dataset in a ZIP archive. For noobs like us we will be batch uploading images and let Tensor auto create tags. Now you can add specialized tags and prune the auto-generated tags, but that’s a whole different article. Once the images are uploaded and the tags generated pick a LORA. Do you have the credits? If so, great! If not you have a couple choices: 1) Bank more credits. 2) Pick another LORA. 3) Use fewer EPOCHs. Once your settings have been adjusted to your liking and you've filled in all necessary information begin a training session. If it fails don't panic. Try again in a few hours when there's less traffic on the site.Congratulations! You've made a LORA!

So you want to make a LORA, but aren't sure where to start? Hopefully, after reading these articles, you'll decide this is something you'd like to try. I've tried to include examples and helpful resource links.ADDENDUM: Since writing this there have been (or will be by the time you read this) changes here at tensor.art. As I write this what those changes may entail, how they may affect TOS policy or LORA creation is unclear. However, in a effort to get ahead of things, I wrote a follow-up article about FACSIMILE images, at least how I define them, and what that means as many of my test LORAs used such AI generated images. The article can be found here: https://tensor.art/articles/884619347480304594 If you’re like me you have probably found yourself sitting there after your image generated, staring at the screen, thinking: ‘I don't like how this looks.’ Wishing you had a better LORA to use for your artwork. Or maybe you were muttering something with more colorful metaphors because, well, things and stuff happens.Well, curse at your screens no longer! Making a LORA, at least here on Tensor.art, is surprisingly simple.No. Really. If I can do it so can you.There are many how to guides for LORA creation that will take you step-by-step through the process. This article isn’t one of those. This offers some practical advice and, hopefully, useful information based on my own observations (good, bad, and d’oh!) from creating a couple of passable LORAs.To begin have a basic idea of the what and why of your goal for creating a LORA. Even if it’s just to see how it’s done, you need a good basic thematic end result in mind. Meaning what type of character or art style you'd like to create with your LORA. Anything is possible from Dark Jedi, Barbarian Warrior, Frank Thorne art style, Pies, race cars, Whimsical Kittens, Post-apocalyptic Landscape, and etcetera et al.Main question: What kind of art are we wanting to create?Not sure. Then ask yourself: What character or style do I need (or want to create) that there isn’t already a LORA for. Or that no LORA exists for your preferred generative Model. Or that you just would like a better version of. (Not all LORAs are created equal. You can try three similar character LORAs and often get very different results.)For example there are many Character LORAs for actress Kaley Cuoco. But there were none of her as a fantasy warrior. Of course with the right LORA combos you can create awesome Kaley Cuoco Barbarian Warrior artwork. But what about a blood-splattered Warrior? Sure, we can create that, but using how many LORAs to get the image just right?I took this idea and created BARBARIAN WARRIOR KALEY using 25 images in the training dataset. The images I curated include 7 head and shoulder shots, 3 mid close-up body shots (head to knees), 10 full body shots in dynamic poses, 1 artistic image, and 4 sample images using a test prompt similar to the one used to generate the LORA sample images generated during training then in-painting to touch up the face and eyes.Here’s the published LORA link: https://tensor.art/models/876777897912706295?source_id=njeyo1nrnEW3oPYsaX309xkk so you can view the gallery of generated images. This LORA appears to do dungeon backgrounds and blood-splatter very well. In fact it may seem too bloody, depending on your prompt, and not enough like Kaley even at the default weight of 1. Lowering the weight to .8 lessens the persistence of blood-splatter, but this may affect the likeness since the source images were AI generated facsimiles. But it's good enough we can use it with a Character LORA, if we want a more realistic likeness.Now, having seen some of what is possible, all you need is to…Step 1. Decide what type of LORA you want to create, and what you want it to do for you.(PART 2 HERE: https://tensor.art/articles/876855087064988656)

Many may have noticed there's an image to text feature in many AI generators. If you're like me you've probably just shrugged it off, after all why waste credits on that?Well, while searching for something else, I stumbled on a free img2txt tool. Probably not as good as the one here but, intrigued, I uploaded a movie still. What I got was not that spectacular a description, yet useful. The kind of basic description that's easy to work with.Here's the prompt I created from one of the outputs. Text in green is the AI generated text. Text in yellow was my edit. Text in blue were my additions to the prompt.In a dimly lit, mysterious temple chamber, three women stand in a circle, with one in a striking red robe dramatically raising her arms as if invoking a powerful spell. The second woman exudes an air of apprehension, while the third observes with intrigue. In the center, a figure lies on a table, draped in white. Explore themes of mysticism, women's empowerment, and the tension between fear and curiosity. Hyper-realistic, cinematic, Technicolor, creative, extremely intricate detail, majestic, mythical, museum-quality, anatomically correct, realistic, realism, lifelike textures, Occult Symbolism, chiaroscuro, volumetric lighting, hyperdetailed, art trending on Artstation, DeviantArt, classic insufficiently attired occult art in style of Frank Cho, Stephen Youll, Don Maitz, Frank Frazetta, Luis Royo, Julie Bell, Frank Kelly, Ninjartist.Not a terrible prompt. I'd probably change a lot more, especially that last sentence (in the original green text prompt), to give it a more sinister horror movie vibe. However it works okay as is. Granted the images created with basic models sometimes don't look that great. Save, of course, for Flux, which was used to generate the cover image. What's interesting is how simple the descriptions created were.If you just need help describing a scene I'd say give the text to image tool a try. I certainly will in future. Especially for movie stills.Have fun.

Need a last minute greeting card with a personal touch?Good news, your AI image generator can do that. And it's very simple.Step 1. Choose your AI model. My preference is Flux Schnell.Step 2. Write your prompt.Here's the one used to create the digital greeting card image attached to this article:Thanksgiving card Art, idyllic first Thanksgiving scene, artistic, creative, 75mm, top holiday text reads: Happy Thanksgiving!Step 3. Save to local device/HD. Attach to message. Send.That's it.Addendum: You can get as detailed as you want with your prompt. I kept it simple. But consider these prompts:Thanksgiving card Art, idyllic first Thanksgiving scene with pilgrims and Indians, artistic, creative, 75mm, top holiday text reads: Happy Thanksgiving!Thanksgiving card Art, idyllic family Thanksgiving scene of multi-generational women and children in kitchen preparing Thanksgiving feast, artistic, creative, 75mm, top holiday text reads: Happy Thanksgiving!I created images for each. You can find them in my gallery. Notice the differences, and that's without adding any additional descriptors for style. Be as inventive as you want to be!Note: You may need to generate more than one image to avoid unusual results, like heads in cooking pots, but that's half the fun. Happy Thanksgiving!

This is just an Addendum to my introductory article on making Christmas Magic. Creating a fun seasonal postcard type image is fairly simple. If you like whimsical art then your descriptions should reflect this. Simple descriptions work best. Remember the more detailed and focused your prompt the closer to what you imagined that picture may be the AI output will get.Good descriptors are key to creating seasonal themed art. In fact, in many instances, you can just add a few keywords to a Favorite prompt to tweak it's aesthetic.Keywords and descriptive ideas you may want to use in your prompt: Christmas, Yule, Yuletide, seasonal, holidays, mistletoe, Christmas Fantasy, Christmas Magic, Christmas decor, holiday decorations, Yulelog burning in fireplace, with a mistletoe border, with a Christmas tree in background, Christmas Elves...You get the idea. 😉Here's a fun prompt to play with:(Christmas Fantasy:1.1), elfin Elves, Christmas Magic, triadic colors, windswept hair, extremely high-resolution details, photographic, realism pushed to extreme, fine texture, incredibly lifelike, dynamic lighting, upon a smokey tavern stage in the City of Christmas three beautiful barefoot Christmas Elves wearing Santa hats and red and green Christmas Elf costumes battle each other with candy cane wands filling the air with the colorful sparkling word "Noel" during a Yuletide magic festival, above them a festive Christmas banner reads "Yuletide Magic", epic fantastical Christmas art, epic fantasy, vintage Christmas postcard art, Hyper-realistic, windswept hair, intricate detail, majestic, high-quality resolution, UHD, highly detailed face,distinct facial features, photorealistic,75mm, 20 million pixels,4k,8k,16k,UHD,realism,realistic,vivid,key visual, film still, cinematic color grading, depth of field, sharp focus, professional photography, masterpiece, ultra detailed photorealistic, raw, anatomically correct, Hyperrealistic, splash art, concept art, mid shot, intricately detailed, color depth, dramatic, 2/3 face angle, long shot, side light, colorful background, whimsical.Addendum: A quick explanation. Everything after "hyper-realistic" are descriptors trying to do one thing, get a higher quality better clarity more realistic image. Is it all necessary? Probably not. You can delete it entirely and replace it with something like... "artistic, creative, extremely intricate detail, Christmas art, vintage holiday postcard art, art in style of Stephen Youll, Don Maitz, Howard David, Julie Bell, Frank Frazetta, Luis Royo,, trending on Artstation, DeviantArt, classic Christmas scene." And still get a decent, if very different looking, image. Or not replace it at all. Go ahead, give it a try.(When done here check out my articles on crafting holiday cards. What you learned here should prove useful.)Remember the image attached as a cover for this article used the above prompt with Lora's to tweak the result, but you can use whatever model you want. That image and more can be found in my gallery. Plenty of prompt ideas for you to remix or copy-and-paste.Here's that same prompt without Lora add-ons:https://tensor.art/images/799639532594004905?post_id=799639532589810603https://image.tensorartassets.com/cdn-cgi/image/anim=true,plain=false,w=1024,f=jpeg,q=85/posts/images/794519944456031492/200f186e-c522-4cf8-8b4d-6c31cfccc832.png[Is there a way to post images inside articles?]I hope this has been useful.

Writing a prompt is simple. However Avoiding Unfortunate Image Results, that can sometimes pose a challenge. To begin let’s start with some basics and a few work prompt challenges…Step 1. Chose Your ThemeWhat do you want to create? An epic fantasy battle? A snapshot of models on a runway? Pirate ships in a storm? A lush garden full of gnomes? Giant Mecha battling on an alien moon? Clowns tossing pies at each other? With AI there is virtually no limit. Sci-fi space opera, Frankenstein’s monster at the prom, anime unicorns surfing on rainbows, funny chibi characters, if you can imagine it, and describe it in a prompt, the odds are you will be able to create it. You are the artist. AI is your brush.Let’s say you want “vintage” Hollywood style pics. Very simple to do with the right model, right? Sure. But you can often achieve similar results with a good prompt.You knew that? Of course you did. But what adjectives you might like to use for a really retro Hollywood style will depend upon the genre. For instance to describe Film Noir words like: pulp, sultry, brooding, sensual, dark, shadowy, steamy, &tc may leap to mind. Those same adjectives might not work as well to describe a western gunslinger or comedic clown scene. Would they?Step 2. Choose Your WordsThere’s no wrong or right approach. If you want feel free to copy and paste vintage scene descriptions or sections of old movie reviews into your prompt. Yes, that can generate images, just be aware the copied text will need major adjustment.But, wait, you may be saying to yourself, aren’t prompts supposed to be well structured? Yes. But more important is consistency of directions. You can generate an image with:Cinematic, Hyper-realistic, dancing toads, chibi.Hyper-realistic, Nicholas Cage, superman punching Lex Luthor, Metropolis.Comic Book Art, Maisie Williams, Raven casting spells, Teen Titans.Technically those prompts have a subject and style/aesthetic description, which a good prompt needs to generate an image. Go ahead. Try them for yourself. That first prompt is simple. Your results shouldn’t vary too much. The second and third prompts will likely generate wildly different results. This is because what they lack is a coherent directive.Step 3. Write a Strong DirectiveWhat is a strong/coherent directive?Something like:A hyper-realistic cinematic action scene of Nicholas Cage as Superman punching the villain Lex Luthor into a building in the City of Metropolis.Go ahead. Try the prompt. You will get interesting images. However they may not necessarily be close to the described action.Why?When images are generated that do not precisely depict what was in the prompt it’s usually because of vague directions (meaning poorly written) or filtering. Yes, filtering. In comic books punches fly like raindrops in a hurricane. But certain AI generators may balk at depicting violence, unless you use a model specifically designed for comic book action scenes.Give the prompt another try rewriting it to depict whatever actor you’d like to see as (or with) your favorite superhero or superheroine, and no they don’t have to punch a villain. It can often be just as challenging to depict them rescuing a damsel/dude in distress or just licking an ice-cream cone. You just need a clear description and strong main subject.Step 4. Choose a Strong Main SubjectThis can seem highly subjective. If you are a fan of Game of Thrones, you will assume everyone knows who Maisie Williams character was. Likewise, to fans of comics, characters like Superman and Spiderman are considered well known (as is which comic imprint they belong to). It may have thus come as a surprise, if you count yourself amongst the ranks of the Fandoms mentioned, to see the results of the third prompt:Comic Book Art, Maisie Williams, Raven casting spells, Teen TitansBecause of her pop culture fame this will generate images of Ms. Williams. But will it definitely generate images of Ms. Williams as the Teen Titan Raven? If you tested the prompt you know the answer. You’re likely to get a better result with a simple declarative sentence like:Maisie Williams as the Teen Titan superheroine Raven.Another example. Consider the image this prompt produces:Cyberpunk, heroine, neon cityscape.Now compare with the image created with this prompt:A flamehaired Cyberpunk heroine eating noodles under neon signs against a futuristic cityscape.Once you have a strong main subject you can easily, and simply, refine your image with detailed aesthetic and/or style cues.Step 5. Consider Aesthetic and StyleAs a general rule the more detailed, and tightly written, a prompt the better the output. This means judicious yet creative use of descriptors paired with aesthetic and style directives.Aesthetic means anything that describes the desired visual style. A few examples: 3D, art-deco, artistic, bas-relief, cinematic, epic fantasy, fresco, game art, impressionism, oil painting, painting, photo, photorealistic, realistic, surrealism, whimsical, and etcetera.Style, while it can overlap with aesthetic, is really just how you stress what should have more weight in the image. IE: epic fantasy painting, photorealistic game art, sci-fi space-opera, Halloween horror, Christmas Fantasy, whimsical impressionism, and etcetera et al.Remember the cyberpunk heroine from our previous example? Think of ways to tweak the prompt to your liking by adding style and aesthetic descriptors and see what a difference those tweaks make. Don’t have time right now, no problem, try this:A flamehaired Cyberpunk heroine eating noodles under neon signs against a futuristic cityscape, cinematic, photorealistic, sci-fi, intricately detailed futuristic architecture, realism pushed to extreme.Wow, quite the difference, right?To get even more depth out of your generated images you can also expand the details. Right now that’s a static photo. But what else might be happening? Are there drones hovering above? Does she have a pet, if so what kind? What if this was a “bustling street scene” or she was “sitting under the shadows of flying cars”?You can create myriad creative flourishes in your art just by adding a few additional details. Try to avoid overwriting the prompt by stacking your descriptors. However, just as you want to avoid overwhelming the AI with too much detail so, too, will you need to be careful how you combine descriptors.Step 6. Temper Your DescriptionsThe two main difficulties, I have found, with attempting artistic precision is overwriting your descriptions and combining too many descriptors.How is that a problem? To keep a prompt below a set word threshold often means taking shortcuts like reducing a sentence to keywords : “night, cemetery, Halloween theme” or eliminating entire sentences by combining descriptors : “Happy Holidays, winter scene, day, golden hour, cinematic, cookies” in a prompt. It’s not necessarily bad methodology, but sometimes subtle contextual cues may be lost when a prompt gets simplified in this way.Did you notice anything different? Two additional variables are specified: night and day.When I chose the word “temper” for this section title I did so not just to mean “keep it simple” but to suggest you hone your prompts, if not to perfection, at least to include as much basic information as possible. You are not just any artist, you are a blacksmith hammering the imperfections out of metal to forge a strong sharp blade, your prompt. That means considering whether to include directions for lighting, background details, setting description, specifying an architectural style for buildings, amplifying style by including references to genre and artistic style (or specific artist's style, "in the artistic style of [insert artist name]") or any other details you think will help make your art look better.Just remember: A prompt doesn’t necessarily need to be overly bloated with directions for style, lighting, color, and everything else you might read about in guides. Just choose what seems right to you. You’ll quickly discover what works and what doesn’t, and that can be half the fun of creating AI art!Speaking of fun Here’s one last prompt to try. It’s a subtle difference from the images generated by the prompts above, but a good one. :A flamehaired Cyberpunk heroine eating noodles sits in a café on a busy street under a neon sign that reads: “Noodles”, a drone flies overhead, fog shrouded futuristic cityscape, cinematic, photorealistic, sci-fi, intricately detailed futuristic architecture, realism pushed to extreme.I hope the challenges posed in this article have been fun and helpful.