I admit I don't use it much, end is the important one for me. The case for start, it's there for CN chaining, starting with a CN then swapping to another one. It's most visible with lineart CN, if you set it to 100% it won't add much details, but if you set it to 90%, at the final steps it'll add random things that benefit the gen. The case for end, freedom to SD, an example, with segmentation that ends early, at 30-50% of the gen, you can nudge SD to create multiple subjects without constraining it to the given areas. I usually set all my CNs at 95% at max unless it's tile, because giving SD freedom for that final step improves gens a lot. They're very powerful feature, specially end. I hesitate to add parameter that can't also be well explained. Start/End for ControlNet: do you have any good resources on this? I've played with it before, but don't find it intuitive at all. There is already a converging and a stochastic option. Normal vs Karras is an option because (bad?) lora exist that work with one but not the other. Limited samplers: could expand, but again I'd like reasons for additional options beyond "it exists".Will certainly accept code that makes it a bit more flexible though. Otherwise there are way too many possible configurations, and it's hard to maintain and test. Yes the installer is mainly meant to support the case where you don't modify Comfy yourself.That makes it optional, editable and reusable. There is no preprocessor at the moment, but I plan to add a button/toggle which will pre-process the current image on demand. ControlNet is a work in progress, definitely will add more.Start/End for ControlNet: do you have any good resources on this? I've played with it before, but don't find it intuitive at all.Customize checkpoints, lora, samplers, etc.SDXL is supported, but not very good at inpainting (yet.).ControlNet: scribble and line art supported at the moment, working on more.Queue and history, preview results and apply them as layers.Work at any resolution, will generate at native SD resolution and upscale/downscale to fit.Automated SD install, or bring-your-own-ComfyUI.Inpaint and Outpaint (prompt optional, it uses ControlNet, Clip-Vision/IP-adapter to get good results).Prompt is optional, you can get great results without one. You can select an area and push a button and it works, similar to Photoshop generative fill/expand. The focus is on iterative workflows with lots of inpainting/outpainting. Much improved! Unfortunately I still don't get arganman's dome.This is my Stable Diffusion plugin for Krita! The idea is to build something streamlined, rather than another copy of Auto1111 UI. I change Input Strength to 0.1 (original image has smaller impact), Steps to 50, and Guidance Scale to 13 (more strict adherence to prompt) and get this result: But wait - I haven't adjusted any of the settings. Result is improved but not very impressive, especially when compared to argaman123's. I've tried using his exact image and prompt with DiffusionBee to compare the results. Reddit user argaman123 has posted a fantastic example of Image to Image. Input Strength - how much the original image influences the prompts, with 0.9 being highly influences and 0.1 being not very influential.With Image to Image generation you can adjust the following options: Image To Image lets you do many cool things like increase that quality of your drawings (turn MS paint drawings into worwhaks of art), change the contents of your images, or combine different styles. Seed - a random number used as a starting point in the generation.A higher number is more strict, a lower number will create more varied results Guidance Scale - Put simply, it lets you specify how strict the AI should follow your prompt.Batch size - How many images in a single generation.Steps - Sampling steps are how many cycles the AI will take to generate the image.The interface has helpful style suggestions and you can change the following settings under the options button:
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |