Act Two + Lipsync Test BE/Bimbo (featuring Hailey)

So you may or may not be aware, but I'm currently working on a full-length bimbo animation that features Ashton (from Dream Girl). I wanted her eyes to cross briefly every time she loses IQ, so this test was mainly to determine if Act Two can pull that off - it can't :(

The facial animation is pretty good though and I think the body dynamics are decent - even a bit of boob jiggle in there lol. I think my first choice would be just to animate any facial expressions and poses in Kling, but if I can't get what I want then Act Two is certainly an option.

Just so you guys understand what I'm testing here, this is the workflow for the first 30-second animation
1. Generate an image of Hailey with a fine-tuned Flux model
2. Generate audio with Elevenlabs V3 (TTS with instant voice clone)
3. Take video of myself acting out how I want her to move (captures body movements)
4. Plug the Hailey image and the video of myself into Act Two
5. Take that generated video and combine with the Elevenlabs audio in Pixverse to get the lips synced

Of course, I could lip-sync myself as part of Step 3 and skip that last Pixverse step, but I'm pretty bad at lip-syncing lol.

The workflow for the BE animation was similar:
1.Generate an image of Hailey with a fine-tuned Flux model
2. Inpaint her bustier (used Cyberrealistic Pony 12.5 in this case)
3. Use Kling to morph between initial image in as start frame and the busty version
4. Generate audio with Elevenlabs V3 (TTS with instant voice clone)
5. Take video of myself acting out her facial expressions
6. Plug the Kling BE video and the video of myself into Act Two
7. Take that generated video and combine with the Elevenlabs audio in Pixverse to get the lips synced

Conclusions:
Act Two (with body animation): Hands are bad as expected and Act Two only outputs at 720p. Even with chest up framing, it struggled to keep the face consistent in landscape aspect ratio. If I end up using this, I'd probably generate in portrait or square aspect ratio and then composite over the white background in Adobe Premiere. In it's current state, I don't think Act Two is super useful for me.

Act Two (face animation only): This feels about the same as Act One to me. Which is fine - that's what I used for The Right Match and latina.exe.

Lipsync: Pixverse lipsync seems really solid to me. This will save a bunch of time so I don't have to spend 50 takes trying to lip sync perfectly.

Eleven v3: I also was testing how my Hailey instant voice clone worked with the new v3 model. I think it's pretty good! I wasn't super happy with the giggling but one thing I saw is that they recommend matching your training data with what you want to output. So in other words, if my Ashton voice training data has some giggling in it, it will probably work much better! It's worth a try at least.