by @Fureko

Hey people! In this guide, I will help you generate your own pictures with Tensor.art!


I took this guide from Reddit and added a few explanations. Also, if you have any other questions or if I've made a mistake, let me know and I'll update this Rentry :3.


What is Tensor.art?

Tensor.art is an online website where you can generate your own pictures with many parameters (it's similar to Stable Diffusion but online, which is useful for people who can’t generate images locally).

Tensor.art is free for everyone, but you have a daily credit limit. Free users get 50 credits per day, or you can pay for a membership to have more credits and access advanced features such as speed, upscaling, etc.


How to begin with Tensor.art?

  • First, go to Tensor.art. On the main page, click on the Sign in button here:

signin

  • Then, create an account with your email or use a Google account, as you wish:

signin

  • Once you are signed in, click on the Create button here:

create

  • Now you have the creation page to start generating pictures!

create

Now, how do you set up the different settings?

First, you need to select a model from Tensor’s large library and choose a few LoRAs.

note: What are a model and a LoRA?

  • An AI model is a program trained on various pictures in different styles, such as anime, realistic photos, landscapes, etc. You use a [Prompt] to tell the AI what you want, and it converts your words into a picture.
    • We use a file from this model called [Checkpoint], which represents a specific step in the training process that yields the best results.
  • A LoRA (Low-Rank Adaptation) is a method to add additional training parameters to the main models you use. It can incorporate a specific character, style, extra details, etc. This approach is beneficial because it avoids re-training an entire model, making it faster, cheaper, and saving memory since models are usually around 4GB-6GB while LoRAs are only about 100MB-600MB.
  • Now, on the creation page, click on the little > and in the search bar, type NTR MIX (or you can search for other models as you wish—there are many, but I will use the same one from the Reddit guide). Warning In this guide, I recommend selecting the 3.5b version of this model as used in the Reddit guide.

signin

signin

  • When you change the model, some creators add suggestions in the settings to help you use it. You can either include these recommendations or not, but in this guide I will include them.

signin

  • Now you will see some changes on the creation tab when you add these settings, but I won't use this LoRA and will show you how to add another one.
  • First, start by deleting the LoRA already loaded using the little Trashcan button and the Add LoRA here:

signin

  • I will add two LoRAs: 748cm and HauntedPastel:

signin

signin

  • If, when adding the LoRA, the settings recommendation pops up, just cancel it and don't add it.
  • Now we have our two LoRAs, and it's time to set up our settings to create an image.

    • First, make sure that both LoRAs are set to a strength of 0.8, then click the Negative button to display it and remove the positive prompt.
  • The Positive prompt will ask to the AI [What you want in your picture]
  • The Negative pormpt will ask to the AI [What you don't want include in your picture]

signin

  • Secondly, go back to your settings, and:
    1. Select your Aspect Ratio (Custom or the one provided by Tensor).
    2. Click on the Advanced button to display more settings.
    3. In Sampler, select Euler and in Scheduler, select Normal (you can also choose different parameters to find what you like).
    4. For Sampling Steps, I recommend between 20-30 steps—25 is good, but if you choose more, your generation will cost more credits. Always set Clip Skip to 7 and ENSD to 0.

signin

  • Finally, it's time to write your prompt. I suggest you start with this prompt:
    ((score_9_up, score_8_up, score_7_up)). (((A MATURE ADULT man, male focus, solo))), {YOUR DETAILS WILL GO HERE. REMEMBER TO USE ()}. Detailed hair, detailed eyes, absurdres, masterpiece, best quality, high resolution, 4k, perfect illustration, detailed art style
  • And for the negative prompt use: lowres, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, censored, bar_censor, (pregnant), chibi, loli, simple background, conjoined, child, childish, kid, small, short, boy, girl, young, undersaturated, low_resolution, poorly_drawn, bad_anatomy, bad_proportions, disfigured, gross_proportions, malformed_limbs, missing_arms, missing_legs, extra_arms, extra_legs, fused_fingers, polydactyly, long_neck, camera, text, duplicate, cloned_face,, Also Don't forget to check A1111, then click on Generate! and you'll see the cost of your generation.

note: The parentheses () are used to "force" the AI to focus on a specific detail; the more you use them, the stronger the effect will be.

signin

  • After a short wait, the AI will generate your picture!

signin

  • But wait, is there a little problem? Yes! Sometimes the eyes and face are misgenerated, but don't worry; that happens because the AI struggles to recreate faces and hands. There's a handy tool called Adetailer and I will also show you how to ask the AI to generate the same picture again.
    • Click on your generated image and look for Seed

signin

  • Now, copy the Seed from your settings, turn on Adetailer in your settings, and select face_yolov8n.pt

info: I must warn you that turning on the Adetailer will double the generation cost, so I suggest you turn it off, generate your image, and once you have a good picture, reuse the seed with the Adetailer turned on.

signin

  • There's also an easier way: if you don't remember the seed or otherwise, you can use the Remix function by clicking on the icon in the top right corner, and Tensor will use all the settings from that picture.
    IMPORTANT: TO RECREATE THE SAME PICTURE, YOU MUST USE THE EXACT SAME SETTINGS, INCLUDING THE PROMPT. CHANGING THE ASPECT RATIO WILL RESULT IN A DIFFERENT PICTURE.

signin

  • Now we have our perfect picture with a well-generated face!

signin

  • Now you finally know how to generate good pictures with Tensor.art! But remember, this is just the basics—you will need to experiment and learn how to craft a good prompt and play with the settings.

note: As a final note, the function called Upscale will make your generated picture larger and of better quality. Personally, I don't use it because JanitorAI converts and reduces your image into .webp and adjusts the resolution to use less memory.


Thank you for reading this guide and I hope it helps you!!

Edit
Pub: 27 Mar 2025 19:10 UTC
Edit: 30 Mar 2025 16:23 UTC
Views: 260