invoke-ai
[enhancement]: Add a better interface for editing prompt words
What should this feature add?
There are two very useful extensions in Webui. Based on the existing "Show Dynamic Prompts" interface, I think it is possible to create another "Prompt Editor" interface and incorporate similar functions of these two extensions.
sd-webui-prompt-all-in-one SD WebUI Tag Autocomplete
In the Webui, these two extensions are almost universally used by everyone. Since most anime model descriptions use tags as the basis for prompt words, once you get used to this plugin, it's really hard to abandon this level of convenience!
Now that the webui has no further development path, after seeing the numerous powerful features of invoke, I believe this is the only tool in the AI drawing field that can match the professionalism of Comfyui. I hope to gradually restore the excellent functions of Webui (including the extended functions of the community, as many of these extensions are definitely used by everyone), and it is certain that Webui can be completely replaced in the future!
oh, oh, that one was on my personal list for quite a while! I've already done some research on making the autocomplete function (one where you can import a .csv with prompt keys, trigger words and even effectiveness scale in some cases)
As for sd-webui-prompt-all-in-one thingy I had a bit similar "prompt constructor" concept in mind for COMPEL/CLIP compatible models where you can essentially convert positive/negative prompt fields in Linear UI into "building blocks".
In my thoughts, said blocks could have weight and be reusable, for example ("blue sky", "yellow sand, beach, palm trees", "crayon art, simple colors").and(1,1,1) could be converted into three separate blocks, back and forth, as long as syntax is correct. I thought of it as a simple UI additional, quite transparent and non-intrusive for the the overall InvokeAI workflow.
I could work on this after my current QoL plan is done.
oh, oh, that one was on my personal list for quite a while! I've already done some research on making the autocomplete function (one where you can import a .csv with prompt keys, trigger words and even effectiveness scale in some cases)
As for sd-webui-prompt-all-in-one thingy I had a bit similar "prompt constructor" concept in mind for COMPEL/CLIP compatible models where you can essentially convert positive/negative prompt fields in Linear UI into "building blocks".
In my thoughts, said blocks could have weight and be reusable, for example
("blue sky", "yellow sand, beach, palm trees", "crayon art, simple colors").and(1,1,1)could be converted into three separate blocks, back and forth, as long as syntax is correct. I thought of it as a simple UI additional, quite transparent and non-intrusive for the the overall InvokeAI workflow.I could work on this after my current QoL plan is done.
Wow, so you guys actually considered this matter? Then I'm relieved! In my opinion, for the time being, if Invoke wants to quickly gain popularity among anime artists, this feature is definitely essential.
Every time I've returned to Invoke, I have found myself looking for this feature, and every time it's still missing I drift away from Invoke. I work heavily with anime models (noob, illustrious) and this system is simply mandatory for these models. I've read the discussion about this on discord and can't agree at all with the idea that implementing this functionality (tag autocomplete based on a CSV import especially) is a technological dead end as newer models emerge. Not all of us own racks of 5090's for AI inference. A lot of people use 12GB cards, and rely on SDXL-based models and the vast ecosystem of loras and controlnets available.
Every time I've returned to Invoke, I have found myself looking for this feature, and every time it's still missing I drift away from Invoke. I work heavily with anime models (noob, illustrious) and this system is simply mandatory for these models. I've read the discussion about this on discord and can't agree at all with the idea that implementing this functionality (tag autocomplete based on a CSV import especially) is a technological dead end as newer models emerge. Not all of us own racks of 5090's for AI inference. A lot of people use 12GB cards, and rely on SDXL-based models and the vast ecosystem of loras and controlnets available.
FYI I'm still on it, And likewise, every time I use Invoke I imagine it having these tools. I even wrote myself a little addition in spare time that I use when I am using Invoke in browser.
(I have no plans on releasing this tool because it's just a draft of things to test out, but nearly everything from it will surely appear in feature or be built upon to grant some new skills to planned feature) I plan to either create all the thing myself or continue on someone's work, if any will appear.
thank you @FloeHetling for keeping the hope alive regarding this critical piece of functionality. I will keep an eye on this thread in hopes of seeing tag autocomplete hitting invoke sooner rather than later!