Responsible AI: Read through our AI Manifesto to learn more about our Mission with AI tools
Private Beta: This feature is currently available only to selected beta testers. If you'd like to get access, let us know.
Overview
Magma AI is a tool that expands your artistic toolset and helps your to overcome art blocks, visualise ideas faster and speed up the overall design process. Magma AI is intended to be an extension of your art skills, and best results are achieved when you use it in collaboration with other drawings tools provided by our platform.
Accessing AI Assistant
Please note that AI feature is only enabled for artists taking part in our Beta program. If you'd like to join and get access, let us know.
Request access to beta
Once you receive confirmation that you're part of the Beta program, once you open the canvas you will see small "AI cloud" icon in the toolbar:
When you use Magma AI for the first time, you'll see a modal with the terms of use and you'll have accept them to get started.
Tool settings
The parameters available in AI tools can be divided into two groups:
AI settings - allow you to fine-tune subject and style of the output
Output settings - define the quantity and quality of the received results
AI settings
Below you will find our simplified guides for Magma's AI settings. For a more complete guide on stable diffusion settings & breakdowns, see this article
The Most Complete Guide to Stable Diffusion Parameters
Please note the some values might only be visible in specific AI modes and/or with "Advanced" settings tab enabled, as specified in the table below
Setting | Description | Supported modes |
AI mode selection | Allows you to switching between operation modes: Render / Variants / Inpaint / Outpaint
Each mode is described in detail below | All |
Layout options | Options to customize the AI tools look and behaviour, described in details below | All |
Render from | Helps AI to understand what content will be rendered (drawing or photo)
If "Empty layer" is selected, AI will ignore canvas content and generate image based solely on prompt | Render |
Strength | Higher value means that the end result will vary from the original artwork more | Variants, Inpaint |
Guidance | Higher value means that AI will stick to the text prompt more strictly | All |
Advanced settings | Button to toggle advanced settings on and off | All |
Sampler | Computation formula applied to generation algorithm resulting in different visual styles of the end result | All |
Seed | Random number that automatically changes with each new generation and guarantees that the next result will be different from the previous one. | All |
Seed lock | Prevents the seed from being automatically updated. | All |
Mask blur | Defines how much should the mask be blurred in order to blend the result into surroundings more smoothly | Inpaint, Outpaint |
Negative prompt | Textual description of things that should not appear in output.
Sample negative prompt: lowres, bad anatomy, bad hands, cropped, low quality, blurry, out of frame | All |
Prompt history | Button to open prompt history window | All |
Prompt | Textual description of what artwork should be added to the canvas.
Sample prompt: cute dog, portrait, blurry background, vivid light, illustration, high quality | All |
Output settings
Number of results | How many outputs should be generated simultaneously to choose from |
Generation resolution | Resolution of the generated image. Higher resolution requires more time. |
Number of steps | Number of refinement step done by the AI, higher number usually produces higher quality result, but requires more time. |
Generate button | Starts the generation process |
Time estimate | Estimate on how long it will take to generate all the results.
Tip: All the parameters above affect the generation time. Lower chosen params to receive results faster. |
Layout options
Pixel Perfect | Helps to improve quality of AI analysis of the existing canvas content in the Render Mode |
Show backdrop | Turns the dark backdrop around AI bounding box area on/off |
Use different settings for each mode | If enabled, all the AI settings will not be copied when switching between AI modes |
Switch mode when applying settings from history | Makes it easier to apply settings shown in Prompt History, as it allows not only to copy parameters, but also switch the mode in a single action |
Reset all settings | Resets all the setting to default value |
General flow
Add necessary canvas content and frame it in bounding box
Adjust prompt and parameters
Wait for results
Choose best result and click "Accept"
Optionally, while the results list is open, it's also possible to right-click any of the results and select additional options from the dropdown menu:
Accept: The same as green "Accept" button, closes result list and places selected result on current layer
Clone to layer below/above: creates a new layer and places on layer above (useful when we want to keep more than 1 result)
Remove: deleted results from the list
AI modes
Render mode
Overview
Render mode allows you quickly polish your drafts or render photos in entirely different style, while preserving your original shapes, lines, and contours. It's useful for:
quicker ideation and brainstorming
texture and colors inspiration
speeding up creation of game / animation assets
You can optionally also use a prompt without any visual input, which can additionally help with generating reference images or textures.
Cheatsheet
Variants mode
Overview
Variants mode generates multiple slightly different versions of your artwork while retaining certain amount of reference to the original idea. It's useful for:
adjusting style of your artwork
generating more design inspirations
creating multiple versions of game / animation assets
Pro tip: Try adjusting the "strength" parameter to adjust how far the variants differ from the original
Cheatsheet
Inpaint mode
Overview
Inpaint mode helps you to quickly add characters or objects to your existing artwork, providing you with a useful color and lighting information. It's useful for:
adding elements to your existing images
generating objects as a base for further overpaint
Cheatsheet
Watch a video guide
Please note that this video was made in previous version of Magma AI tool, so your user interface might look slightly different. Please refer for the tool settings description above for details.
Important: the methods of drawing the mask has changed - it's now possible to draw mask directly in the AI tool rather than having to create mask as on a separate layer.
Outpaint mode
Overview
Outpaint mode allows you to fill in blank space on the canvas with the described objects. It's useful for:
changing composition of your artworks
building out your ideas from multiple smaller pieces
Cheatsheet
Watch a video guide
Please note that this video was made in previous version of Magma AI tool, so your user interface might look slightly different. Please refer for the tool settings description above for details.
Prompt history
Users can open prompt history by clicking "History" button in the AI tool settings:
Prompt history allows everyone to see what prompts were used for a given artwork previously. The list is updated in real time and contains prompts of all users.
By default only prompts from current sessions are shown, but it's possible to load previous sessions as well by clicking "Load previous sessions" link:
AI attribution system
As pointed out in our AI Manifesto, we believe that ability to distinguish between human-made and AI-made artworks is essential and it is a very important component of building ethical AI systems.
In Magma, we're tracking how exactly was the AI used to achieve final results by marking each layer with a relevant icon.
Right now, the attribution system is limited to a couple of options:
AI-generated - "Render" mode was used on this layer at some point in the past. It means that is no other additions were made, the layer might be in large part a result of AI generation.
AI-assisted - "Variants", "Inpaint" or "Outpaint" mode was used in the past on this layer. It means that the artists workflow included use the AI tools, but most likely majority of work was performed directly by the artist.
Imported - User has pasted an external image onto the layer. We're not able to track what happens outside of Magma, so this might mean any origin of the pasted image (artists own work, downloaded image, result from other AI systems, etc).
If various modes and methods were used on the same layer, it will have all of them listed in the tooltip:
If layer or drawing is duplicated, the information above is copied along as well.
In the future we'll be looking into expanding this system to provide even more granularity around how AI Assistant was used in each canvas.
It's also possible to understand the extent to which AI was used on a given canvas by opening the Prompt History window, which lists all previous AI generations along with their respective prompts.
Troubleshooting
I'm getting error telling me some words are restricted
There are some keywords which are not allowed in the prompt field - these might include both living artists & studios names, as well as some other NSFW or sensitive content.
You can read more about Magma prompt limitations here: https://help.magma.com/en/articles/7260990-magma-ai-prompt-limitations
I'm getting "Failed to generate drawing error"
This means that AI tool was not able to generate an image for you, and there might be a couple of reasons:
NSFW (Not Safe For Work) content -- we don't allow imagery generating nudity, gore, or any other controversial subject. Most of the time this will happen if you try to include or suggest such things in the prompt -- but sometimes it might also happen randomly, in which case you can try again to see if the problem disappears.
Non-optimal input -- since AI is usually trained using images with square or rectangular shape, it might sometimes generate unexpected results when tasked to create imagery with an unusual shape or ratio. If you keep getting error while having a selection which is for example very long or wide in shape, try adjusting it and trying again.
Wrong parameters -- another cause for the error might be issue with AI input, for example: tool expected to receive some pixel input, but the canvas is empty. In that case please check cheatsheet section for the relevant mode to make sure you've provided all the required input
Some of my results are not showing up
Sometimes a result might be qualified as NSFW by our filter, in which case preview will be impossible, and we will display a sad Pyro icon instead:
Similarly as in the question above, it might mean that prompt contained some NSFW content, but it also can happen randomly in some cases. In most cases adjusting the prompt and/or retrying will solve the issue.
It takes very long time to generate an image
First of all, please check your parameters -- high resolution and high number of steps might result in generation taking much longer.
Secondly, the waiting time might depend on number of artists currently using the system. Although we're doing our best to keep the times low for everyone, in case of higher than usual traffic you might expect some delays.
That said, if you regularly expect waiting times longer than 1 minute, it might be a technical issue -- feel free to contact our tech support.
I keep getting the same results all the time, nothing changes
It most likely means that you've "locked" the generation seed so that each run of the algorithm returns the same results. Please see the "Tool parameters" section above to see where the Seed lock option is located and disable it.
I'm receiving weird / distorted results
It's usually difficult to tell what causes the distortion, but here's a list of things to try:
If using selection, make sure it's rectangular, or even better, square
Try adjusting your prompt by rephrasing it, removing unnecessary elements, etc
Use lower quality resolution (higher resolution is currently known to sometimes cause issues, we're working on a fix)
If your Steps parameter is very low, try a higher setting
If your Guidance setting is very low, try a higher setting
If using the variants mode:
Make sure that current selection includes as little as possible of transparent area - the canvas or selection should be fully filled in for best results
If the variants differ too much from the original image, try using a lower Strength setting
If using the inpaint mode:
Try to make sure that the mask of inpainted element fits as much as possible into the current perspective of the drawing, and a realistic size (for example: mice is way smaller than a bear)
If using the outpaint mode:
Make sure that on top of transparent area that will be filled in, there's also enough of non-transparent area in the selection, as it will be used by the AI to determine the context and style of generated image
🛟 Need more help?
Get more support from our team and the community on our Magma Discord server