Knowing the right model to pick for your video is like seasoning a dish - the model must work with you and add a kick your vision 👁️
What do Frame-by-Frame models do?
Frame-by-Frame models are our traditional animation style. They have more flicker which offers a trippier effect than Text-to-Video. Frame-by-Frame has less fluidity than Text-to-Video with motion and movement, and are most comparable to stop motion animation.
How many credits do Frame-by-Frame models use?
Frame-by-Frame models utilize 1 credit per second. Let's say you have a 4 minute music video, this would utilize 240 credits by default. However, you may want to re-render 50% of the video, this would utilize 120 credits, for a total of 360 credits.
What are Allrounder models?
Allrounder models are designed for versatility - they typically perform well across a wide range of prompts and artistic styles. Here is a quick view of the Allrounder models 👇
OpenJourney: A model striving to achieve the beautiful aesthetics of Midjourney. It's versatile, designed to excel at producing artistic, concept-driven visuals with a focus on creativity and fine detail, blending realistic textures with stylized aesthetics for a balanced and imaginative output
DreamShaper: Our most popular Allrounder model - it's highly stylized, imaginative, evoking dreamlike imagery while blending realism and artistic abstraction for a unique visual aesthetic
Stable Diffusion XL: A powerful, general-purpose model designed for generating high-resolution, photorealistic, and versatile imagery, with enhanced detail and composition, with a sweet spot for face generation. Works well in adapting various styles and subjects. The video resolution of this model is doubled than with all other models, which means it can produce beautiful, high-resolution frames but is also significantly slower than all other models
Juggernaut XL: A leader in realistic models. Based on Stable Diffusion XL, it's best to use shorter prompts for this one. The model is robust and general-purpose, perfect for creating high-resolution, photorealistic, and highly detailed imagery, excelling particularly in face generation and fine-tuned compositions. It adapts seamlessly to various styles and subjects. The video resolution of this model is doubled than with all other models, which means it can produce beautiful, high-resolution images but is also significantly slower than all other models
What are Specialist models?
Stylized models are crafted for focused expertise – they excel in specific styles, themes, or tasks, delivering exceptional results within their niche. However, they may struggle or produce less effective results outside their specialization. Here’s a quick overview of stylized models 👇
Realistic Vision: Optimized for creating ultra-detailed, photorealistic imagery, excelling in producing lifelike textures, environments, and subjects with exceptional accuracy and clarity. Looks great with flashy colors
Analog Diffusion: Designed to emulate the nostalgic aesthetic of vintage film photography, producing images with grainy textures, muted tones, and authentic retro vibes reminiscent of analog cameras. Great for demonstrating emotions, traditions and narratives
Anything 3: High quality and highly detailed anime, manga, and cartoon-style images that excel in detailed character designs. It offers vibrant colors and dynamic compositions while being flexible across various anime aesthetics
Cyberrealistic: Maximum realistic images that are modern and clean. Minimal prompting is needed. Has a knack for futuristic, cyberpunk-inspired aesthetics and lifelike visuals with a modern yet high-tech edge
RealCartoon3D: Perfect for stunning 3D cartoon visuals. Offers vibrant colors and detailed textures, with smooth and stylized designs for a polished animation aesthetic
Custom Models
Creating a custom model can be very powerful and unlocks the ability to enhance and stylize your video. They're also super fast to create - a custom model typically takes 30 minutes or less to train. You can set up two different types of models at the moment. Here is a breakdown of each 👇
Person 🤵
Check out this tutorial on how to create a custom Person model:
Step into your scenes by adding yourself as a custom model. You'll want to upload a minimum of 15 images so that the neural network can get to know your Person better. On the website, we specify different types of images and the quantity. As an example, 10 to 20 images of yourself, 10 close-ups of your face, 5 chest up images, 3 full-body images. Be sure to change things up by adding different backgrounds, lighting, and angles. Also, no sunglasses or masks - the neural network wants to see the real you 🧘
When creating the model, you'd want to select Person from the drop-down list and select your base model. OpenJourney is the default when creating a custom model, but you can select a model that you feel would work best for what you're trying to create🖼️
Object 🎢 and Style 🌇
Set the mood to your scene with a custom model set as Object or Style. For objects and we recommend uploading more than 15 images. When creating a style model, the more, the merrier! You can select up to 199 images to upload.
When creating an Object model, be sure that you're only adding images that show the same object. As an example, if you want to create a model based on apples 🍎 , be sure to only have images with apples in them with different backgrounds, lighting and angles.
If you're creating a Style model, be sure that you're adding variations of the style you're wanting to train. Different lighting, backgrounds, angles and moods helps the neural network to best understand the style that you're going for 🚀
When creating an Object or Style, you'd want to select the appropriate option from the drop-down list and select your base model. OpenJourney is the default when creating a custom model, but you can select a model that you feel would work best for what you're trying to create 🖼️
I've selected my Person, Object, or Style. What do I do now?
Get ready to upload! Once you've completed uploading your images, you should see them all visible right from the site. You'll want to complete the steps by checking off the boxes at the bottom, ensure that your Category and Base Model are selected, and then create a name for your model. Once you're ready to go, select Agree To The Terms and Conditions, and you're all set! We'll start training your model immediately.
It's important to note that training timeframes are based on several factors, including how many images you uploaded, the base model you've selected, and more. Once your model is fully trained, you should be able to Start A New Project and see your model at the bottom in the Custom Models bar.
Once you're working with your custom model and creating a prompt, you'll want to add sks object, sks person, sks male, sks female, or sks style to your prompt - this is so the neural network knows exactly what you're wanting to emphasize from your model.
Can I revise my custom model by adding or removing images after it's been trained?
At this time, this feature isn't supported. However, each month as a subscriber you receive an allotment of custom models to train! This means you can train a new model with the images you want to add or remove to fine-tune your ideal custom model.
NOTE: The models displayed in the videos are all using the same prompting and settings. It's important to mention that prompting and the model you select go hand-in-hand, working together to achieve the best results.
NOTE: Frame-by-Frame custom models are different than Autopilot Styles. For more info about Autopilot Styles, please check out this article.