My primary tool is still WAN2GP by DeepBeepMeep, specifically using the Wan2.1 model. It seems to give the best results and is trainable on a 24GB card, so there are lots of loras available and I can make my own for style. I vibe coded a python script to stabilize the colors between frames and one to do rife crossfades between clips via a blender plugin. I might take another crack at the color stabilization, it's still a problem and I've got some ideas.
I've experimented a tiny bit with seedance 2.0, and in some ways it's very impressive, but it also has a problem where animation regresses to the mean animation style, which is a crummy flash cartoon with minimal movement - which defeats the purpose. Also generating a seedance video costs $1 where generating a video locally costs $0.02. Also there's agressive content filtering. Still might be usefull for some scenes.
---
How long it takes is tricky to answer because there were different phases with different amounts of attention and effort required. First I cut the panels out and did some test runs trying to get a single animated sequence, but the results were pretty abstract so I decided to animate each frame individually. I drew keyframes for the flat chested versions, then a little bit of editing of the original panels to get clean plates, then I could use those as start and end keyframes. Then it was a matter of queuing up jobs to run over night and filtering them down to the good stuff. I generate a batch of clips, pick out the good ones, then queue up more depending on what's working and what isn't. I generated about 400 clips, sifted down to 95 good ones, then the final page uses 13. Probably 100+ gpu hours, which is a lot, but I made several mistakes and had to redo things.
Entry Level and Deep Breath both have some tight shots and overlapping panels that make it difficult to animate. I do want to get more into animated shorts, so for future projects I'm thinking about moving to more of a storyboard style where every panel is a clean 16:9 shot. I've got a little experimental 30 second short of the morale officer undressing where I drew keyframes specifically to animate with the AI, and it worked pretty well but stalled out a bit on some shots. It's got music and foley.
AI voice acting is not happening with current techniques. Sometimes I can get good results for a 10 second clip from LTX2, like maybe an instant loss or imagination teaser, but for longer stuff there's just too many variables that don't fit into a text prompt. Character consistency, speed, tone, emphasis, emotionality, pauses, there would have to be a dedicated timeline interface for these things and I haven't seen anything close to that. So we still need humans for now.
---
Organization is an ongoing problem. When I started the site I had ~200 pics and now there are > 4000. Animations like this are currently in the AI section: https://satinminions.com/AI.html I could make an "Animated Comics" subsection under animation, and probably get rid of the "ai enhanced" section which is Dain/RIFE frame interpolation. There are a couple of non-ai comic pages that have animation as well, which could fill out a subsection nicely. Though some people are touchy AI being co-mingled with traditional frame by frame and I want to be as transparent as possible. At some point AI, if used correctly, is just a tool. But what do you think?
This animation actually has 4 (I think) RIFE interpolated frames in it, but you can't even tell where they are. I only used them to change the timing, not to make it ultra smooth and uncanny.
Yeah on the first panel the comic has vague definition between his arm and her shoulder and the action isn't really loopable, but the other two panels came out well so I wanted to do the whole page.
Mmmmm... I'd love to see an AI animated version of this page one that shows the contents of this victim's scrotum being completely emptied-out before their penis gets transformed into a pussy against their will, especially if that animation can convincingly capture the conflicting experiential extremes our hapless and helpless protagonist is being subjected to here: the excruciating pain and sense of irreversible loss throughout this ordeal; and the epic and literally mind-blowing orgasmic pleasure being felt as a result of this forced ejaculation of all of their remaining semenal fluids and sperm contained therein!!
The choice to keep Allison short-haired reminds the audience that she once was male (even though her short-haired 'do has a decidedly girlish look to it)and was forcibly changed into a female to become the sex slave she now is for the demons she now serves.
I like the short hair. I agree with the poster above that I prefer long hair on a girl. But if you carry over the TG theme from the sister image I think the short hair fits better.
For me hair style is a bit along the line of clothes. If somebody were to change gender why would their clothes or hair style change? Over time yes, but those things are sort of separate from biology.
I really, REALLY love your art work and style. Infact I envy it over my realism. I do have to say I have never liked girls with short hair, I have been a huge long haired girl kind of guy, but this... Your art really makes me think twice :) again, thank you for your beautiful art.
It must be hard for her in several ways. One, in her current state, is obvious: She's been turned into a woman, was straight, and is just following along the path of attraction to be a straight woman so she's naturally attracted to all this.
On the other hand, another thought that may go through her head is "God... I used to be that strong... Not anymore...."
It's an abbreviation for a working title of a cyberpunk-esque story idea I'm toying with. I've got to name the files something and of course the golden rule is you don't abbreviate cyberpunk.
My primary tool is still WAN2GP by DeepBeepMeep, specifically using the Wan2.1 model. It seems to give the best results and is trainable on a 24GB card, so there are lots of loras available and I can make my own for style. I vibe coded a python script to stabilize the colors between frames and one to do rife crossfades between clips via a blender plugin. I might take another crack at the color stabilization, it's still a problem and I've got some ideas.
I've experimented a tiny bit with seedance 2.0, and in some ways it's very impressive, but it also has a problem where animation regresses to the mean animation style, which is a crummy flash cartoon with minimal movement - which defeats the purpose. Also generating a seedance video costs $1 where generating a video locally costs $0.02. Also there's agressive content filtering. Still might be usefull for some scenes.
---
How long it takes is tricky to answer because there were different phases with different amounts of attention and effort required. First I cut the panels out and did some test runs trying to get a single animated sequence, but the results were pretty abstract so I decided to animate each frame individually. I drew keyframes for the flat chested versions, then a little bit of editing of the original panels to get clean plates, then I could use those as start and end keyframes. Then it was a matter of queuing up jobs to run over night and filtering them down to the good stuff. I generate a batch of clips, pick out the good ones, then queue up more depending on what's working and what isn't. I generated about 400 clips, sifted down to 95 good ones, then the final page uses 13. Probably 100+ gpu hours, which is a lot, but I made several mistakes and had to redo things.
Entry Level and Deep Breath both have some tight shots and overlapping panels that make it difficult to animate. I do want to get more into animated shorts, so for future projects I'm thinking about moving to more of a storyboard style where every panel is a clean 16:9 shot. I've got a little experimental 30 second short of the morale officer undressing where I drew keyframes specifically to animate with the AI, and it worked pretty well but stalled out a bit on some shots. It's got music and foley.
AI voice acting is not happening with current techniques. Sometimes I can get good results for a 10 second clip from LTX2, like maybe an instant loss or imagination teaser, but for longer stuff there's just too many variables that don't fit into a text prompt. Character consistency, speed, tone, emphasis, emotionality, pauses, there would have to be a dedicated timeline interface for these things and I haven't seen anything close to that. So we still need humans for now.
---
Organization is an ongoing problem. When I started the site I had ~200 pics and now there are > 4000. Animations like this are currently in the AI section: https://satinminions.com/AI.html I could make an "Animated Comics" subsection under animation, and probably get rid of the "ai enhanced" section which is Dain/RIFE frame interpolation. There are a couple of non-ai comic pages that have animation as well, which could fill out a subsection nicely. Though some people are touchy AI being co-mingled with traditional frame by frame and I want to be as transparent as possible. At some point AI, if used correctly, is just a tool. But what do you think?