For a lot of time I used A1111 as my main tool. Well, for a good reasons - it was the first tool ever available to be used, and because of it, a lot of plugins was written for it pretty fast. But now let's just compare two tools?
Automatic1111 can load LoRAs, LoHAs, LoCONs and so on (if proper addons were installed). Comfy can do same with it's LoraLoader node, but it is on official level, built-in. A1111 can load TI and Hypernetworks. Comfy can too. A1111 is pretty fast. Comfy is pretty stable (and also quite fast). Both A1111 and Comfy have Img2Img. But Comfy allows for more automatic mid-generation parameter switching. Both tools support SDXL, SD2, SD1.5. But A1111 requires more steps to use refiner model. Both tools support inpainting. A1111 is sightly better at it, but Comfy already have custom nodes made for inpainting + if you seek true king of inpainting, look at InvokeAI. Both tools allow outpainting. Both tools have Upscale by model. Both tools support infinite prompt length. Both tools support attention/emphasis. ComfyUI support multiple modes of those. Both tools support variations. A1111 just makes it easier to use. A1111 have prompt editing tool. And so does ComfyUI through additional node Both tools have checkpoint merge. ComfyUI does it in realtime. Both support ESRGAN, unCLIP, and other things.
Only A1111 have pretty interesting, useful scripts like XYZ plot. ComfyUI have some problems replicating it with ease.
So, if pretty much everything in ComfyUI already at A1111 level, why even bother to choose A1111 anymore? Comfy gives much more control over images and their variants. It is also much easier for me to actually share workflows, too! All you need is to download a single file, put it in UI, press Generate and you are fluffin' golden! :)
What you listed is 99% about catching up, not surpassing. It'd be much more interesting to see the list of things ComfyUI can do which A1111 WebUI can't.
What you listed is 99% about catching up, not surpassing. It'd be much more interesting to see the l
Well, while technically true - ComfyUI has surpassed A1111 in SOME things. Most of them are usually the convinience of experimentation. For example:
1) You need to create a hugely labor-intensive picture that goes like so: * Create plasma noise from nothing * Make SD create winter on the right side of picture while making summer on the left side. * All that in the specific model (Anything3, for example) * Then take resulting picture * Enlarge it with model 4xUltraSharp * And using it as a base, inpaint Soviet Winter and American Summer on the left and right respectively * Using ANOTHER model (Realistic) * With ANOTHER CLIP-SKIP value * And with LoRA activated on top of that.
Is this possible with A1111? Sure! ComfyUI makes whole process much more AUTOMATIC than A1111? >:D
Comfy truly excels at being what it is - modular. You don't need to write complex script anymore when you want to add simple feature - just make a single block and include it into your workflow. New noise types, new algos, new filters. However, as both communities progress, they won't be really able to overcome each other, since same technology will be released for both generators sooner or later. Comfy just excels at being much more automatic... than Automatic1111 >:D
Well, while technically true - ComfyUI has surpassed A1111 in SOME things. Most of them are usually
If you want more of the same, ComfyUI seems to be the right tool. But can you just imagine the result of a chain of 3 models and 10 prompts in your brain?
The way I see it, you'll manually go step by step, perfecting each prompt, switching models back and forth, manually fix issues with inpainting on every upscale etc. And eventually, you'll arrive at your perfect node graph for one image. Is it repeatable? Maybe. Assuming every single prompt is 100% stable. Because once you drop to 50% success, your chances become nil to zero. Once you change a major part of the initial prompt, for example by introducing LoRAs, your chain disintegrates. And if you want to inpaint issues or edit in Photoshop, you're still stuck with step-by-step processing.
If you want more of the same, ComfyUI seems to be the right tool. But can you just imagine the resu
Usually you won't need more than 2-3 models. Like SDXL - one for basic generation, one for refinement. Perhaps, if you feeling THAT lucky, one in the middle for even more detailing. Also you usually need 2 to 8 prompts... depending on how many regions you have and how many addons you use. And, of course, sometimes you just want to experiment the living hell out of what you got, with all those 'BW filters' or 'sorted pixel images'. That's actually what is to come in the future in MY gallery >:D
So, yeah. I can't just manually go one by one with such complicated workflow just to see that I can't get good result midway. I need to have bunch (10, 30, 50) images to compare, and get one that is looking like the best one! >:)
Usually you won't need more than 2-3 models. Like SDXL - one for basic generation, one for refinemen
One of the main things I like about CumfiUI over A1111 is the fact that ComfyUI does not keep the models loaded in vRAM between batches. Sure this may take longer as it needs to load the Model each time you run a process, but it allows you do to other things without having to unload ComfyUI.
One of the main things I like about CumfiUI over A1111 is the fact that ComfyUI does not keep the mo
Thats an incredibly valid point. When I was using A1111, and wanted to use separate pieces of that tool (I usually splitted my workflow onto 4+ parts, beginning from generation, then upscaling, then hiresfixing, then cleaning, then upscaling again...) it took 3 to 4 images for my VRAM/RAM to overflow and require restart of A1111. With Comfy, I can generate 400 pictures in a row with a pretty complex workflow, and it still wouldn't crash. It may be slower than A1111, but more stable.
Thats an incredibly valid point. When I was using A1111, and wanted to use separate pieces of that t
