Yara is a companion tool for ComfyUI, based in the terminal. It can:
Download "yara.exe". Open a terminal in the same directory/folder as "yara.exe", and run the program through the terminal by simply typing "yara".
The first time you run, you must select your ComfyUI output folder, and then a config file will automatically be created. You can open the folder containing the config file with the argument "yara config", to edit it manually (most of the options are just for configuring "yara preview").
To save pending generations to a file, run
yara save [name]
You can alternatively use "yara save -wr [name]" instead, if you wish to additionally save the currently active/in-progress generation. Note that Yara cannot save a partial generation; saving and loading an in-progress job will restart the generation from the beginning.
Now, you may clear out the queue or close ComfyUI. When you later want to resume generation, queue them up again by running
yara load [name]
You can print out a list of all saved queues by typing
yara list
and you can delete a saved queue with
yara delete [name]
Warning: After saving/loading generations, resulting images will not have the workflow embedded in them; i.e. you can no longer drag/drop them into ComfyUI to recreate the workflow. The generation details (prompt, model, loras, seed, etc) are still embedded within the image, though, and you can view that by either reading the image file as a text file, or using Yara's image-generation-info function ('yara image').
From ComfyUI Github Issue #69, importing workflows from the api prompt format is planned, so this will hopefully be fixed later.
Sometimes, I load up a long queue of generations, but forget what they were. Or, I may have messed up some of the prompts (e.g. forgot to remove a lora), and want to know which prompt ID's to delete. To examine the active queue, you can run
yara examine
and it will print out the IDs of all prompts, as well as their model(s), lora(s), and positive prompt text.
When I mess up a bunch of prompts, I often want to delete many at once. Doing so in ComfyUI is cumbersome due to some UI issues, so you can instead do it here. To cancel queued generations, run
yara cancel [prompt IDs]
where [prompt IDs] is a space-separated list of prompt IDs (the incrementing numbers labeling queues when you use "See Queues" in ComfyUI).
You can also append "+" to a prompt ID to cancel that prompt as well as the next 100 prompts up, or specify an inclusive range of prompts using "-" as a separator.
yara cancel 60+ // Cancel generations 60, 61, 62, ..., 157, 158, 159
yara cancel 25-30 // Cancel generations 25, 26, 27, 28, 29, 30
ComfyUI doesn't prevent Windows from sleeping, but sleep mode halts ComfyUI generations. You can use yara to conveniently toggle sleep mode with
yara caffeine // disable sleep mode
yara melatonin // enable sleep mode
By default, 'melatonin' will have Windows sleep after 30 minutes of inactivity. You can customize this length in the config file.
Running
yara wait
will halt the terminal until the ComfyUI queue is empty. While waiting, it will print the number of remaining generations every five minutes. It also will estimate (incredibly roughly and naively) the amount of time until all generations are finished.
This is mostly useful for halting the terminal until ComfyUI generations are done. Often, I disable sleep mode, then chain 'yara wait' with 'yara melatonin'. This lets me queue up a bunch of generations, and go leave my computer - when ComfyUI is finished running, sleep mode will be re-enabled, so my computer won't be running needlessly. I also might use this to execute other commands once ComfyUI is finished, such as if I want to both generate images and train a LorA overnight, but don't want both to be running simultaneously.
As a shorthand, you can use
yara cwm
to disable sleep mode, wait until the queue is empty, and then re-enable sleep mode. (cwm standing for Caffeine/Wait/Melatonin).
Run
yara image
to start an interactive session. Enter the filepath of an image, and it obtain the generation data of the image. Model(s), LorA(s), positive prompt text(s), and negative prompt text(s) will be printed to the screen, while the full generation data will be copied to your clipboard.
(note: you can just drag/drop the image into the terminal window, and it will automatically input the image's filepath).
This is particularly useful since when the workflow isn't embedded into the image (as discussed in the "Saving, Loading, Deleting, and Listing Queues" section above).
If you want to preview the generation output without having the ComfyUI window open, you can run
yara preview
to open an always-on-top window that automatically displays the most recently generated image. Settings to configure the window location/size, or to toggle always-on-top/mouse passthrough and more are available in the config file ('yara config').
To open the folder containing the config.json file, run
yara config
Run
yara cai [URLs]
where [URLs] is a space-separated list of the URLs of the CivitAI models/loras/etc you want to download. It will open a browser window to download them, and will copy the title, URL, filename, keywords, and description to your clipboard.
I mostly use it for the latter feature, as I keep a text file with relevant information for LorAs and this makes it easy to copy/paste all the key info. If you only want to copy the information to your clipboard, without downloading anything, add the '-nd' flag:
yara cai -nd [URLs]
To display available commands/arguments, use
yara help
Some of the commands can be shortened, for convenience:
| Command | Alias | |---------------------------|--------------------------------------------------------------------------------------------------------------------| | save | s | | load | l | | delete | d | | examine | e | | wait | w | | caffeine | c | | melatonin | m | | preview | p | | image | i | | help | h |
This is built for the latest ComfyUI release as of July 22, 2023. Future updates may change the API and thus break parts of this program.
If you have an issue, question, or request for some feature/config option, feel free to make an issue or message me.
This is currently Windows-only. A linux build should be able to be compiled from source, though, I think (with a couple minor missing features). I mostly just don't feel like doing it since I don't think too many people are going to use this, but if anyone actually wants to use it and is on Linux, message me or make an issue post.