The workflow is designed for a single image. I didn't test much it for batches more then one. Since v10.3 you can make multiple images in one job.
The features of that workflow (in the newest version):
Basic generation and editing (TXT2IMG, IMG2IMG, inpaint);
LORAs with control of automatic adding keyword into prompt (through LORA Manager);
Advanced generation (multiply TXT2IMG control-net , inpaint, outpaint, onlymasked inpaint);
Checkpoint loader lora with some additional features for Pony (scores presets, rating presets, negative presets and so on);
Saving metadata in Civitai's format;
Advanced prompting (styles selector, control on lora's keyword, negative/positive presets, morphing support, prompt formatting);
5 separated detailer groups with some pre-set settings, additional/combined prompt for each of them. You may add more detailer groups to connect them in the same way or delete it if necessary. I think it is pretty simple;
Almost fully switch-toggle control: if you use the workflow as is, you don't need any manual connection, muting, bypassing, activating.
(The previews of the adetailers are exceptions. They have global toggles, but sometimes it's more useful to activate or bypass some of them manually. I didn't find a better decision for it.Since v11, there are no toggles for it; the previews are inside adetailers groups.)Model upscale
Lama object remover
Not totally, but mostly nodes in the 'backend' (nodes under Lora Manager node) have tips on nodes and their links that give some help if you want to bring some modification;
Detailed (probably too detailed) a readme node with explanation of the logic of work and some advice.
Few bookmarks/shortcuts for fast navigation
Token counter for positive (+Lora's keyword + styles)
Florance-2 tag caption
The main goal of this workflow is to make the basic functionality of the WebUI and some extensions I use as functional as possible, mostly for Pony, as a residual matter — sd and sdxl. Some of these attempts have already failed.
Yolo/segmentation models:
https://huggingface.co/ashllay/YOLO_Models/tree/main
https://huggingface.co/Anzhc/Anzhcs_YOLOs
Additional prompting:
https://github.com/asagi4/comfyui-prompt-control
Credits:
The main concept is from this workflow https://civarchive.com/models/1507055/text-to-img-complete-workflow-multi-loras-adetailer-wildcards-and-more-comfyui by @Mittoshura.
Description
V11.0 Subgraph:
This version is mostly about trying to optimize by turning nodes into subgraphs. In the first order of Adetailer groups, trying to make all ‘interface’ clearer and simpler and save advanced features.
turned many things into subgraphs;
fixed some small issues in Adetailer nodes and one important issue in the 'eyes' group;
remade some structures of Adetailer groups (deleted some nodes, moved others to their respective groups, deleted toggle for previews);
changed back to the previous system of token count (the new one just got stuck after a certain count and showed '77' always);
repositioned groups and nodes (again) (besides Adetailers, the main changes are in Main toggles and generation settings groups);
remade bookmarks (again) (deleted one, and 'E' (by default) now shows ‘Main toggles and loaders’, ‘generation settings’, ‘Florence-2 tag image caption’, ‘Lora’s’, and Main Prompt on one screen. In my opinion, the new position and structure of these elements are much more useful);
changed 'additional negative' from a string to a multiline string;
pretty much remade the README node, I hope it is now more readable;
remade the links in the notes' nodes, remade the reminder note.
V11.5 Subgraph:
This is kinda fix.
Deleted FaceID (IPAdapter mode). After some more tests, I realized that it is pretty useless. This is fine for sd1.5, but on Pony it almost doesn’t work, especially on realistic. On SDXL it's better, but still not stable enough or the quality is pretty bad. In my opinion, controlnet/Img2img/Reactor could give more useful results and are easier to use.
Remade adetailer's subgraphs.
Slightly changed control-net subgraphs and florence-2 subgraph.
Turned back "wireless" links to adetailer's subgraphs and other nodes.
Few other small changes.
V11.6 Subgraph:
Fixed the issue with the Pony's checkpoint's node/toggle not working when scores and ratings were prompted for any values.
FAQ
Comments (4)
This workflow currently performs very well with Pony and Illustrious, but with noobai, images will have errors such as screen tearing or distorted figures. Could you fix this in the next version, or tell me what settings I should use? PS: I'm using vae_sdxl
Thank you for your response. I've never tried the workflow with NoobAI. Can you provide me with the most common checkpoint that you are using? Maybe I could do something with it.
P.S I tried NoobAI-XL (NAI-XL). I think dealing with clip skip, with a value of "1" and slightly lower CFG (around 5–5.5 probably), it works, even though somehow I didn't get multiple characters. Some schedulers didn't work, like 'karras' in my test. Anyway, try putting clip skip as 1 and CFG as 5, and anyway text me back the result.
@DemonAlone https://civitai.com/models/2105328 prompt just 1girl no lora
@kg843 This scheduler 'linear_quadratic' with the model I gave you in the name above doesn't work for me either. Try the same sampler (euler_ancestral) but with schedulers 'normal' or 'simple'.
I see the name of the model in the workflow you uploaded, but it is not clear what mode exactly you are using. If you give me a link, I could be more precise.
The workflow you uploaded is an older version with a ponycheckpoint node issue; at some point I broke the link to the output of this node (this is fine for Pony's checkpoint, but for others it just adds useless tokens and wastes token limits). That should cut off quality Pony's tokens (score9_up and so on). I recommend using a more modern version of the workflow.