CyberPony version is the best I've used so far
Yeah, I find Cyberpony pretty good too and am using it to develop my own GenAI mod. I use 'Euler a' with 20-25 steps and get good balance between appearance, generation time and fidelity to prompts (according to
my taste at least). Just started on V14 myself and though it's a bit better in its range of appearances (faces, ages, skin tones ...) I've gotta disagree about ADetailer. I find using it with face_yolov8s.pt greatly improves faces, but at significant cost in image generation time. I also find CyberRealistic_Negative_PONY_V2-neg and CyberRealisticPony_POSV1 useful for nicer faces and less cartoons or monsters - especially in sex positions - with no extra processing overhead I've noticed.
A tip for using ADetailer without adding too much image lag.
When you've got a frequently interacted character with a commonly reused image (i.e. one that pops up immediately without regenerating), switch ADetailer on and Auto Generate off (if you're using more detailed images you can also switch on Upscaling). Return to the game and use "X" to delete the current image then "+" to regenerate it with ADetailer on. Then go back to GenAI Config and switch Auto Generate back on (if that's how you like it) and ADetailer and Upscaling off. Then you'll get the better quality image 'for free' whenever you return to it.
A tip for the 'hallucinations' (which I often find entertaining, TBH) - including the ones where you just get a flat 2D field of amorphous colour - is to 'Append' something fairly neutral (e.g. pretty) to the positive prompt then force an image regen. Often it's a permanent fix for that NPC.
if the GenAI mod could somehow send the next prompt to webui before we actually progress the scene.
Can't imagine how to do that unless they start rolling out quantum computers with precognition
. I've moved the toolbar up a bit and will have a think about your other toolbar suggestions and see what I can do, but don't expect anything soon. My 'to do' list is already receding into the hazy distance and the new LR2 update hasn't helped.
But hey, if you want me to give your suggestions more priority you could start testing my mod and making suggestions specific to it
. So far it's basically add-ons to AnotherMike's narration text version, though I can't guarantee I haven't introduced extra bugs. I'd recommend leaving Process Narration off to avoid the error when meeting Nora and Stephanie on Saturday night (still unfixed from AnotherMike's original). Even with it off my mod uses narration text to contextually inform a lot of image generation, hopefully with a bit more effect and relevance than merely appending the narration text to the prompt as AnotherMike does. Pony models aren't too clever with natural English.
