Your Place 0.20-0.21v (WIP)[game]/ Love in futa time 2 [video](WIP) jun/13
Hello, everyone!
Post only for patreons because it includes adult images.
This time, I've focused more on testing the AI than on other aspects. However, I've continued with the renders, but I've mainly dedicated myself to experimenting with the AI.
Regarding the progress on the renders, I'm still creating renders for the studio scene, including the renders I'll use as a base for the AI animations. The scene itself will work as follows: the MC will automatically ask Alex and Natalie a series of questions, and after these questions are finished, the MC must punish the one with the lowest score with a spanking. This will be Natalie, whom Alex will then ask the same questions again. If she answers incorrectly again, the MC will spank her; if she answers correctly, she'll be spared. The player must remember the correct answers, and if they spank when they should and hold back when Natalie answers correctly, a scene with the MC, Natalie, and Alex will take place. If the player makes a mistake, a lesbian scene will take place with the MC as a spectator.
The spanking scenes are already finished. To achieve this, I used several renders as inputs for an FLF2V workflow in ComfyUI. This involves an animation generated from two frames: the first and last frames of the animation. I did this to prevent the animation from displaying things that don't exist. Let me explain. If the camera is fixed on one point and the animation moves the camera, it would show parts of a room that don't exist but are randomly generated by the AI. This would be an unacceptable discontinuity flaw. Therefore, I used this two-frame system to control what is displayed. Problems still arise: six-fingered hands, arms appearing out of nowhere, wrists rotating like Megan's head in The Exorcist—that sort of thing.
I don't mean to say that the FLF2V workflow is useless, but as I've been using it, it's very inefficient. Generally, in AI, it's necessary to generate several results to find one that meets the user's needs. In my case, I've run over 600 attempts to get two animations of the spanking scene. But perhaps the problem of inefficiency and inadequate results can be solved in another way.
One alternative I had almost ruled out was animation with an I2V workflow. That is, creating the animation from a single frame. This would naturally generate discontinuity problems in the animations if I continue using the images I've been using until now. But what if the reference renders were more general? That is, not close-ups but wide shots that prevent the AI from getting too imaginative. This is what I'll start experimenting with starting today, Friday, as I write this post. If I get good animations of this type in the coming weeks, I hope to start posting animations of old scenes made with this new method on Twitter or Discord.
Perhaps the AI problem was more my own error in using inappropriate inputs than a lack of capacity in the system to generate what the scene needed. I've also stopped using Comfy's default prompts and started looking for more NSFW-focused ones on a site called Civitai. With these new prompts and by using different images from a wider perspective, I'll be able to make better use of AI.
Regarding Love in Futa Time 2:
I hope to finish writing this next week; I'm taking longer than I should. Once I'm done, it should take another week to complete everything.